diff options
Diffstat (limited to 'src')
73 files changed, 2227 insertions, 1786 deletions
diff --git a/src/api/Cargo.toml b/src/api/Cargo.toml index 4d9a6ab6..dba0bbef 100644 --- a/src/api/Cargo.toml +++ b/src/api/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_api" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,11 +14,11 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_model = { version = "0.8.0", path = "../model" } -garage_table = { version = "0.8.0", path = "../table" } -garage_block = { version = "0.8.0", path = "../block" } -garage_util = { version = "0.8.0", path = "../util" } -garage_rpc = { version = "0.8.0", path = "../rpc" } +garage_model = { version = "0.8.1", path = "../model" } +garage_table = { version = "0.8.1", path = "../table" } +garage_block = { version = "0.8.1", path = "../block" } +garage_util = { version = "0.8.1", path = "../util" } +garage_rpc = { version = "0.8.1", path = "../rpc" } async-trait = "0.1.7" base64 = "0.13" diff --git a/src/api/admin/router.rs b/src/api/admin/router.rs index 3fa07b3c..62e6abc3 100644 --- a/src/api/admin/router.rs +++ b/src/api/admin/router.rs @@ -143,10 +143,13 @@ impl Endpoint { } generateQueryParameters! { - "format" => format, - "id" => id, - "search" => search, - "globalAlias" => global_alias, - "alias" => alias, - "accessKeyId" => access_key_id + keywords: [], + fields: [ + "format" => format, + "id" => id, + "search" => search, + "globalAlias" => global_alias, + "alias" => alias, + "accessKeyId" => access_key_id + ] } diff --git a/src/api/k2v/router.rs b/src/api/k2v/router.rs index 50e6965b..e7a3dd69 100644 --- a/src/api/k2v/router.rs +++ b/src/api/k2v/router.rs @@ -96,7 +96,7 @@ impl Endpoint { fn from_get(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ EMPTY if causality_token => PollItem (query::sort_key, query::causality_token, opt_parse::timeout), EMPTY => ReadItem (query::sort_key), @@ -111,7 +111,7 @@ impl Endpoint { fn from_search(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ ], no_key: [ @@ -125,7 +125,7 @@ impl Endpoint { fn from_head(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ EMPTY => HeadObject(opt_parse::part_number, query_opt::version_id), ], @@ -140,7 +140,7 @@ impl Endpoint { fn from_post(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ ], no_key: [ @@ -155,7 +155,7 @@ impl Endpoint { fn from_put(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ EMPTY => InsertItem (query::sort_key), @@ -169,7 +169,7 @@ impl Endpoint { fn from_delete(partition_key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), partition_key, query, None), + (query.keyword.take().unwrap_or_default(), partition_key, query, None), key: [ EMPTY => DeleteItem (query::sort_key), ], @@ -232,21 +232,18 @@ impl Endpoint { // parameter name => struct field generateQueryParameters! { - "prefix" => prefix, - "start" => start, - "causality_token" => causality_token, - "end" => end, - "limit" => limit, - "reverse" => reverse, - "sort_key" => sort_key, - "timeout" => timeout -} - -mod keywords { - //! This module contain all query parameters with no associated value - //! used to differentiate endpoints. - pub const EMPTY: &str = ""; - - pub const DELETE: &str = "delete"; - pub const SEARCH: &str = "search"; + keywords: [ + "delete" => DELETE, + "search" => SEARCH + ], + fields: [ + "prefix" => prefix, + "start" => start, + "causality_token" => causality_token, + "end" => end, + "limit" => limit, + "reverse" => reverse, + "sort_key" => sort_key, + "timeout" => timeout + ] } diff --git a/src/api/router_macros.rs b/src/api/router_macros.rs index 4c593300..07b5570c 100644 --- a/src/api/router_macros.rs +++ b/src/api/router_macros.rs @@ -4,10 +4,9 @@ macro_rules! router_match { (@match $enum:expr , [ $($endpoint:ident,)* ]) => {{ // usage: router_match {@match my_enum, [ VariantWithField1, VariantWithField2 ..] } // returns true if the variant was one of the listed variants, false otherwise. - use Endpoint::*; match $enum { $( - $endpoint { .. } => true, + Endpoint::$endpoint { .. } => true, )* _ => false } @@ -15,37 +14,35 @@ macro_rules! router_match { (@extract $enum:expr , $param:ident, [ $($endpoint:ident,)* ]) => {{ // usage: router_match {@extract my_enum, field_name, [ VariantWithField1, VariantWithField2 ..] } // returns Some(field_value), or None if the variant was not one of the listed variants. - use Endpoint::*; match $enum { $( - $endpoint {$param, ..} => Some($param), + Endpoint::$endpoint {$param, ..} => Some($param), )* _ => None } }}; - (@gen_path_parser ($method:expr, $reqpath:expr, $query:expr) - [ - $($meth:ident $path:pat $(if $required:ident)? => $api:ident $(($($conv:ident :: $param:ident),*))?,)* - ]) => {{ - { - use Endpoint::*; - match ($method, $reqpath) { - $( - (&Method::$meth, $path) if true $(&& $query.$required.is_some())? => $api { - $($( - $param: router_match!(@@parse_param $query, $conv, $param), - )*)? - }, - )* - (m, p) => { - return Err(Error::bad_request(format!( - "Unknown API endpoint: {} {}", - m, p - ))) - } - } - } - }}; + (@gen_path_parser ($method:expr, $reqpath:expr, $query:expr) + [ + $($meth:ident $path:pat $(if $required:ident)? => $api:ident $(($($conv:ident :: $param:ident),*))?,)* + ]) => {{ + { + match ($method, $reqpath) { + $( + (&Method::$meth, $path) if true $(&& $query.$required.is_some())? => Endpoint::$api { + $($( + $param: router_match!(@@parse_param $query, $conv, $param), + )*)? + }, + )* + (m, p) => { + return Err(Error::bad_request(format!( + "Unknown API endpoint: {} {}", + m, p + ))) + } + } + } + }}; (@gen_parser ($keyword:expr, $key:ident, $query:expr, $header:expr), key: [$($kw_k:ident $(if $required_k:ident)? $(header $header_k:expr)? => $api_k:ident $(($($conv_k:ident :: $param_k:ident),*))?,)*], no_key: [$($kw_nk:ident $(if $required_nk:ident)? $(if_header $header_nk:expr)? => $api_nk:ident $(($($conv_nk:ident :: $param_nk:ident),*))?,)*]) => {{ @@ -60,11 +57,9 @@ macro_rules! router_match { // ] // } // See in from_{method} for more detailed usage. - use Endpoint::*; - use keywords::*; match ($keyword, !$key.is_empty()){ $( - ($kw_k, true) if true $(&& $query.$required_k.is_some())? $(&& $header.contains_key($header_k))? => Ok($api_k { + (Keyword::$kw_k, true) if true $(&& $query.$required_k.is_some())? $(&& $header.contains_key($header_k))? => Ok(Endpoint::$api_k { $key, $($( $param_k: router_match!(@@parse_param $query, $conv_k, $param_k), @@ -72,7 +67,7 @@ macro_rules! router_match { }), )* $( - ($kw_nk, false) $(if $query.$required_nk.is_some())? $(if $header.contains($header_nk))? => Ok($api_nk { + (Keyword::$kw_nk, false) $(if $query.$required_nk.is_some())? $(if $header.contains($header_nk))? => Ok(Endpoint::$api_nk { $($( $param_nk: router_match!(@@parse_param $query, $conv_nk, $param_nk), )*)? @@ -84,7 +79,7 @@ macro_rules! router_match { (@@parse_param $query:expr, query_opt, $param:ident) => {{ // extract optional query parameter - $query.$param.take().map(|param| param.into_owned()) + $query.$param.take().map(|param| param.into_owned()) }}; (@@parse_param $query:expr, query, $param:ident) => {{ // extract mendatory query parameter @@ -93,7 +88,7 @@ macro_rules! router_match { (@@parse_param $query:expr, opt_parse, $param:ident) => {{ // extract and parse optional query parameter // missing parameter is file, however parse error is reported as an error - $query.$param + $query.$param .take() .map(|param| param.parse()) .transpose() @@ -144,14 +139,40 @@ macro_rules! router_match { /// This macro is used to generate part of the code in this module. It must be called only one, and /// is useless outside of this module. macro_rules! generateQueryParameters { - ( $($rest:expr => $name:ident),* ) => { + ( + keywords: [ $($kw_param:expr => $kw_name: ident),* ], + fields: [ $($f_param:expr => $f_name:ident),* ] + ) => { + #[derive(Debug)] + #[allow(non_camel_case_types)] + #[allow(clippy::upper_case_acronyms)] + enum Keyword { + EMPTY, + $( $kw_name, )* + } + + impl std::fmt::Display for Keyword { + fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result { + match self { + Keyword::EMPTY => write!(f, "``"), + $( Keyword::$kw_name => write!(f, "`{}`", $kw_param), )* + } + } + } + + impl Default for Keyword { + fn default() -> Self { + Keyword::EMPTY + } + } + /// Struct containing all query parameters used in endpoints. Think of it as an HashMap, /// but with keys statically known. #[derive(Debug, Default)] struct QueryParameters<'a> { - keyword: Option<Cow<'a, str>>, + keyword: Option<Keyword>, $( - $name: Option<Cow<'a, str>>, + $f_name: Option<Cow<'a, str>>, )* } @@ -160,34 +181,29 @@ macro_rules! generateQueryParameters { fn from_query(query: &'a str) -> Result<Self, Error> { let mut res: Self = Default::default(); for (k, v) in url::form_urlencoded::parse(query.as_bytes()) { - let repeated = match k.as_ref() { + match k.as_ref() { $( - $rest => if !v.is_empty() { - res.$name.replace(v).is_some() - } else { - false + $kw_param => if let Some(prev_kw) = res.keyword.replace(Keyword::$kw_name) { + return Err(Error::bad_request(format!( + "Multiple keywords: '{}' and '{}'", prev_kw, $kw_param + ))); }, )* - _ => { - if k.starts_with("response-") || k.starts_with("X-Amz-") { - false - } else if v.as_ref().is_empty() { - if res.keyword.replace(k).is_some() { - return Err(Error::bad_request("Multiple keywords")); + $( + $f_param => if !v.is_empty() { + if res.$f_name.replace(v).is_some() { + return Err(Error::bad_request(format!( + "Query parameter repeated: '{}'", k + ))); } - continue; - } else { + }, + )* + _ => { + if !(k.starts_with("response-") || k.starts_with("X-Amz-")) { debug!("Received an unknown query parameter: '{}'", k); - false } } }; - if repeated { - return Err(Error::bad_request(format!( - "Query parameter repeated: '{}'", - k - ))); - } } Ok(res) } @@ -198,8 +214,8 @@ macro_rules! generateQueryParameters { if self.keyword.is_some() { Some("Keyword not used") } $( - else if self.$name.is_some() { - Some(concat!("'", $rest, "'")) + else if self.$f_name.is_some() { + Some(concat!("'", $f_param, "'")) } )* else { None diff --git a/src/api/s3/bucket.rs b/src/api/s3/bucket.rs index 3ac6a6ec..8471385f 100644 --- a/src/api/s3/bucket.rs +++ b/src/api/s3/bucket.rs @@ -161,6 +161,15 @@ pub async fn handle_create_bucket( return Err(CommonError::BucketAlreadyExists.into()); } } else { + // Check user is allowed to create bucket + if !key_params.allow_create_bucket.get() { + return Err(CommonError::Forbidden(format!( + "Access key {} is not allowed to create buckets", + api_key.key_id + )) + .into()); + } + // Create the bucket! if !is_valid_bucket_name(&bucket_name) { return Err(Error::bad_request(format!( diff --git a/src/api/s3/router.rs b/src/api/s3/router.rs index 44f581ff..821b0e07 100644 --- a/src/api/s3/router.rs +++ b/src/api/s3/router.rs @@ -355,7 +355,7 @@ impl Endpoint { fn from_get(key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), key, query, None), + (query.keyword.take().unwrap_or_default(), key, query, None), key: [ EMPTY if upload_id => ListParts (query::upload_id, opt_parse::max_parts, opt_parse::part_number_marker), EMPTY => GetObject (query_opt::version_id, opt_parse::part_number), @@ -412,7 +412,7 @@ impl Endpoint { fn from_head(key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), key, query, None), + (query.keyword.take().unwrap_or_default(), key, query, None), key: [ EMPTY => HeadObject(opt_parse::part_number, query_opt::version_id), ], @@ -426,7 +426,7 @@ impl Endpoint { fn from_post(key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), key, query, None), + (query.keyword.take().unwrap_or_default(), key, query, None), key: [ EMPTY if upload_id => CompleteMultipartUpload (query::upload_id), RESTORE => RestoreObject (query_opt::version_id), @@ -448,7 +448,7 @@ impl Endpoint { ) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), key, query, headers), + (query.keyword.take().unwrap_or_default(), key, query, headers), key: [ EMPTY if part_number header "x-amz-copy-source" => UploadPartCopy (parse::part_number, query::upload_id), EMPTY header "x-amz-copy-source" => CopyObject, @@ -490,7 +490,7 @@ impl Endpoint { fn from_delete(key: String, query: &mut QueryParameters<'_>) -> Result<Self, Error> { router_match! { @gen_parser - (query.keyword.take().unwrap_or_default().as_ref(), key, query, None), + (query.keyword.take().unwrap_or_default(), key, query, None), key: [ EMPTY if upload_id => AbortMultipartUpload (query::upload_id), EMPTY => DeleteObject (query_opt::version_id), @@ -624,63 +624,60 @@ impl Endpoint { // parameter name => struct field generateQueryParameters! { - "continuation-token" => continuation_token, - "delimiter" => delimiter, - "encoding-type" => encoding_type, - "fetch-owner" => fetch_owner, - "id" => id, - "key-marker" => key_marker, - "list-type" => list_type, - "marker" => marker, - "max-keys" => max_keys, - "max-parts" => max_parts, - "max-uploads" => max_uploads, - "partNumber" => part_number, - "part-number-marker" => part_number_marker, - "prefix" => prefix, - "select-type" => select_type, - "start-after" => start_after, - "uploadId" => upload_id, - "upload-id-marker" => upload_id_marker, - "versionId" => version_id, - "version-id-marker" => version_id_marker -} - -mod keywords { - //! This module contain all query parameters with no associated value S3 uses to differentiate - //! endpoints. - pub const EMPTY: &str = ""; - - pub const ACCELERATE: &str = "accelerate"; - pub const ACL: &str = "acl"; - pub const ANALYTICS: &str = "analytics"; - pub const CORS: &str = "cors"; - pub const DELETE: &str = "delete"; - pub const ENCRYPTION: &str = "encryption"; - pub const INTELLIGENT_TIERING: &str = "intelligent-tiering"; - pub const INVENTORY: &str = "inventory"; - pub const LEGAL_HOLD: &str = "legal-hold"; - pub const LIFECYCLE: &str = "lifecycle"; - pub const LOCATION: &str = "location"; - pub const LOGGING: &str = "logging"; - pub const METRICS: &str = "metrics"; - pub const NOTIFICATION: &str = "notification"; - pub const OBJECT_LOCK: &str = "object-lock"; - pub const OWNERSHIP_CONTROLS: &str = "ownershipControls"; - pub const POLICY: &str = "policy"; - pub const POLICY_STATUS: &str = "policyStatus"; - pub const PUBLIC_ACCESS_BLOCK: &str = "publicAccessBlock"; - pub const REPLICATION: &str = "replication"; - pub const REQUEST_PAYMENT: &str = "requestPayment"; - pub const RESTORE: &str = "restore"; - pub const RETENTION: &str = "retention"; - pub const SELECT: &str = "select"; - pub const TAGGING: &str = "tagging"; - pub const TORRENT: &str = "torrent"; - pub const UPLOADS: &str = "uploads"; - pub const VERSIONING: &str = "versioning"; - pub const VERSIONS: &str = "versions"; - pub const WEBSITE: &str = "website"; + keywords: [ + "accelerate" => ACCELERATE, + "acl" => ACL, + "analytics" => ANALYTICS, + "cors" => CORS, + "delete" => DELETE, + "encryption" => ENCRYPTION, + "intelligent-tiering" => INTELLIGENT_TIERING, + "inventory" => INVENTORY, + "legal-hold" => LEGAL_HOLD, + "lifecycle" => LIFECYCLE, + "location" => LOCATION, + "logging" => LOGGING, + "metrics" => METRICS, + "notification" => NOTIFICATION, + "object-lock" => OBJECT_LOCK, + "ownershipControls" => OWNERSHIP_CONTROLS, + "policy" => POLICY, + "policyStatus" => POLICY_STATUS, + "publicAccessBlock" => PUBLIC_ACCESS_BLOCK, + "replication" => REPLICATION, + "requestPayment" => REQUEST_PAYMENT, + "restore" => RESTORE, + "retention" => RETENTION, + "select" => SELECT, + "tagging" => TAGGING, + "torrent" => TORRENT, + "uploads" => UPLOADS, + "versioning" => VERSIONING, + "versions" => VERSIONS, + "website" => WEBSITE + ], + fields: [ + "continuation-token" => continuation_token, + "delimiter" => delimiter, + "encoding-type" => encoding_type, + "fetch-owner" => fetch_owner, + "id" => id, + "key-marker" => key_marker, + "list-type" => list_type, + "marker" => marker, + "max-keys" => max_keys, + "max-parts" => max_parts, + "max-uploads" => max_uploads, + "partNumber" => part_number, + "part-number-marker" => part_number_marker, + "prefix" => prefix, + "select-type" => select_type, + "start-after" => start_after, + "uploadId" => upload_id, + "upload-id-marker" => upload_id_marker, + "versionId" => version_id, + "version-id-marker" => version_id_marker + ] } #[cfg(test)] diff --git a/src/block/Cargo.toml b/src/block/Cargo.toml index cd409001..1e4eb64e 100644 --- a/src/block/Cargo.toml +++ b/src/block/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_block" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,10 +14,10 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_db = { version = "0.8.0", path = "../db" } -garage_rpc = { version = "0.8.0", path = "../rpc" } -garage_util = { version = "0.8.0", path = "../util" } -garage_table = { version = "0.8.0", path = "../table" } +garage_db = { version = "0.8.1", path = "../db" } +garage_rpc = { version = "0.8.1", path = "../rpc" } +garage_util = { version = "0.8.1", path = "../util" } +garage_table = { version = "0.8.1", path = "../table" } opentelemetry = "0.17" @@ -31,7 +31,6 @@ rand = "0.8" async-compression = { version = "0.3", features = ["tokio", "zstd"] } zstd = { version = "0.9", default-features = false } -rmp-serde = "0.15" serde = { version = "1.0", default-features = false, features = ["derive", "rc"] } serde_bytes = "0.11" diff --git a/src/block/manager.rs b/src/block/manager.rs index 7f439b96..1b5a5df0 100644 --- a/src/block/manager.rs +++ b/src/block/manager.rs @@ -3,6 +3,7 @@ use std::pin::Pin; use std::sync::Arc; use std::time::Duration; +use arc_swap::ArcSwapOption; use async_trait::async_trait; use bytes::Bytes; use serde::{Deserialize, Serialize}; @@ -22,6 +23,7 @@ use garage_rpc::rpc_helper::netapp::stream::{stream_asyncread, ByteStream}; use garage_db as db; +use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_util::error::*; use garage_util::metrics::RecordDuration; @@ -87,7 +89,16 @@ pub struct BlockManager { pub(crate) metrics: BlockManagerMetrics, - tx_scrub_command: mpsc::Sender<ScrubWorkerCommand>, + tx_scrub_command: ArcSwapOption<mpsc::Sender<ScrubWorkerCommand>>, +} + +#[derive(Serialize, Deserialize, Clone, Debug)] +pub struct BlockResyncErrorInfo { + pub hash: Hash, + pub refcount: u64, + pub error_count: u64, + pub last_try: u64, + pub next_try: u64, } // This custom struct contains functions that must only be ran @@ -114,9 +125,8 @@ impl BlockManager { .netapp .endpoint("garage_block/manager.rs/Rpc".to_string()); - let metrics = BlockManagerMetrics::new(resync.queue.clone(), resync.errors.clone()); - - let (scrub_tx, scrub_rx) = mpsc::channel(1); + let metrics = + BlockManagerMetrics::new(rc.rc.clone(), resync.queue.clone(), resync.errors.clone()); let block_manager = Arc::new(Self { replication, @@ -128,21 +138,24 @@ impl BlockManager { system, endpoint, metrics, - tx_scrub_command: scrub_tx, + tx_scrub_command: ArcSwapOption::new(None), }); block_manager.endpoint.set_handler(block_manager.clone()); + block_manager + } + + pub fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) { // Spawn a bunch of resync workers for index in 0..MAX_RESYNC_WORKERS { - let worker = ResyncWorker::new(index, block_manager.clone()); - block_manager.system.background.spawn_worker(worker); + let worker = ResyncWorker::new(index, self.clone()); + bg.spawn_worker(worker); } // Spawn scrub worker - let scrub_worker = ScrubWorker::new(block_manager.clone(), scrub_rx); - block_manager.system.background.spawn_worker(scrub_worker); - - block_manager + let (scrub_tx, scrub_rx) = mpsc::channel(1); + self.tx_scrub_command.store(Some(Arc::new(scrub_tx))); + bg.spawn_worker(ScrubWorker::new(self.clone(), scrub_rx)); } /// Ask nodes that might have a (possibly compressed) block for it @@ -309,9 +322,42 @@ impl BlockManager { Ok(self.rc.rc.len()?) } + /// Get number of items in the refcount table + pub fn rc_fast_len(&self) -> Result<Option<usize>, Error> { + Ok(self.rc.rc.fast_len()?) + } + /// Send command to start/stop/manager scrub worker - pub async fn send_scrub_command(&self, cmd: ScrubWorkerCommand) { - let _ = self.tx_scrub_command.send(cmd).await; + pub async fn send_scrub_command(&self, cmd: ScrubWorkerCommand) -> Result<(), Error> { + let tx = self.tx_scrub_command.load(); + let tx = tx.as_ref().ok_or_message("scrub worker is not running")?; + tx.send(cmd).await.ok_or_message("send error")?; + Ok(()) + } + + /// Get the reference count of a block + pub fn get_block_rc(&self, hash: &Hash) -> Result<u64, Error> { + Ok(self.rc.get_block_rc(hash)?.as_u64()) + } + + /// List all resync errors + pub fn list_resync_errors(&self) -> Result<Vec<BlockResyncErrorInfo>, Error> { + let mut blocks = Vec::with_capacity(self.resync.errors.len()); + for ent in self.resync.errors.iter()? { + let (hash, cnt) = ent?; + let cnt = ErrorCounter::decode(&cnt); + blocks.push(BlockResyncErrorInfo { + hash: Hash::try_from(&hash).unwrap(), + refcount: 0, + error_count: cnt.errors, + last_try: cnt.last_try, + next_try: cnt.next_try(), + }); + } + for block in blocks.iter_mut() { + block.refcount = self.get_block_rc(&block.hash)?; + } + Ok(blocks) } //// ----- Managing the reference counter ---- diff --git a/src/block/metrics.rs b/src/block/metrics.rs index 477add66..fbef95af 100644 --- a/src/block/metrics.rs +++ b/src/block/metrics.rs @@ -1,9 +1,11 @@ use opentelemetry::{global, metrics::*}; +use garage_db as db; use garage_db::counted_tree_hack::CountedTree; /// TableMetrics reference all counter used for metrics pub struct BlockManagerMetrics { + pub(crate) _rc_size: ValueObserver<u64>, pub(crate) _resync_queue_len: ValueObserver<u64>, pub(crate) _resync_errored_blocks: ValueObserver<u64>, @@ -23,9 +25,17 @@ pub struct BlockManagerMetrics { } impl BlockManagerMetrics { - pub fn new(resync_queue: CountedTree, resync_errors: CountedTree) -> Self { + pub fn new(rc_tree: db::Tree, resync_queue: CountedTree, resync_errors: CountedTree) -> Self { let meter = global::meter("garage_model/block"); Self { + _rc_size: meter + .u64_value_observer("block.rc_size", move |observer| { + if let Ok(Some(v)) = rc_tree.fast_len() { + observer.observe(v as u64, &[]) + } + }) + .with_description("Number of blocks known to the reference counter") + .init(), _resync_queue_len: meter .u64_value_observer("block.resync_queue_length", move |observer| { observer.observe(resync_queue.len() as u64, &[]) diff --git a/src/block/rc.rs b/src/block/rc.rs index ce6defad..8dae3960 100644 --- a/src/block/rc.rs +++ b/src/block/rc.rs @@ -169,4 +169,11 @@ impl RcEntry { pub(crate) fn is_needed(&self) -> bool { !self.is_deletable() } + + pub(crate) fn as_u64(&self) -> u64 { + match self { + RcEntry::Present { count } => *count, + _ => 0, + } + } } diff --git a/src/block/repair.rs b/src/block/repair.rs index e2884b69..a6ded65a 100644 --- a/src/block/repair.rs +++ b/src/block/repair.rs @@ -53,7 +53,7 @@ impl Worker for RepairWorker { "Block repair worker".into() } - fn info(&self) -> Option<String> { + fn status(&self) -> WorkerStatus { match self.block_iter.as_ref() { None => { let idx_bytes = self @@ -66,9 +66,20 @@ impl Worker for RepairWorker { } else { idx_bytes }; - Some(format!("Phase 1: {}", hex::encode(idx_bytes))) + WorkerStatus { + progress: Some("0.00%".into()), + freeform: vec![format!( + "Currently in phase 1, iterator position: {}", + hex::encode(idx_bytes) + )], + ..Default::default() + } } - Some(bi) => Some(format!("Phase 2: {:.2}% done", bi.progress() * 100.)), + Some(bi) => WorkerStatus { + progress: Some(format!("{:.2}%", bi.progress() * 100.)), + freeform: vec!["Currently in phase 2".into()], + ..Default::default() + }, } } @@ -137,7 +148,7 @@ impl Worker for RepairWorker { } } - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { + async fn wait_for_work(&mut self) -> WorkerState { unreachable!() } } @@ -167,6 +178,7 @@ struct ScrubWorkerPersisted { time_last_complete_scrub: u64, corruptions_detected: u64, } +impl garage_util::migrate::InitialFormat for ScrubWorkerPersisted {} enum ScrubWorkerState { Running(BlockStoreIterator), @@ -271,29 +283,28 @@ impl Worker for ScrubWorker { "Block scrub worker".into() } - fn info(&self) -> Option<String> { - let s = match &self.work { - ScrubWorkerState::Running(bsi) => format!( - "{:.2}% done (tranquility = {})", - bsi.progress() * 100., - self.persisted.tranquility - ), + fn status(&self) -> WorkerStatus { + let mut s = WorkerStatus { + persistent_errors: Some(self.persisted.corruptions_detected), + tranquility: Some(self.persisted.tranquility), + ..Default::default() + }; + match &self.work { + ScrubWorkerState::Running(bsi) => { + s.progress = Some(format!("{:.2}%", bsi.progress() * 100.)); + } ScrubWorkerState::Paused(bsi, rt) => { - format!( - "Paused, {:.2}% done, resumes at {}", - bsi.progress() * 100., - msec_to_rfc3339(*rt) - ) + s.progress = Some(format!("{:.2}%", bsi.progress() * 100.)); + s.freeform = vec![format!("Scrub paused, resumes at {}", msec_to_rfc3339(*rt))]; } - ScrubWorkerState::Finished => format!( - "Last completed scrub: {}", - msec_to_rfc3339(self.persisted.time_last_complete_scrub) - ), - }; - Some(format!( - "{} ; corruptions detected: {}", - s, self.persisted.corruptions_detected - )) + ScrubWorkerState::Finished => { + s.freeform = vec![format!( + "Last scrub completed at {}", + msec_to_rfc3339(self.persisted.time_last_complete_scrub) + )]; + } + } + s } async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { @@ -331,7 +342,7 @@ impl Worker for ScrubWorker { } } - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { + async fn wait_for_work(&mut self) -> WorkerState { let (wait_until, command) = match &self.work { ScrubWorkerState::Running(_) => return WorkerState::Busy, ScrubWorkerState::Paused(_, resume_time) => (*resume_time, ScrubWorkerCommand::Resume), diff --git a/src/block/resync.rs b/src/block/resync.rs index ada3ac54..9c7b3b0e 100644 --- a/src/block/resync.rs +++ b/src/block/resync.rs @@ -63,6 +63,7 @@ struct ResyncPersistedConfig { n_workers: usize, tranquility: u32, } +impl garage_util::migrate::InitialFormat for ResyncPersistedConfig {} enum ResyncIterResult { BusyDidSomething, @@ -123,6 +124,24 @@ impl BlockResyncManager { Ok(self.errors.len()) } + /// Clear the error counter for a block and put it in queue immediately + pub fn clear_backoff(&self, hash: &Hash) -> Result<(), Error> { + let now = now_msec(); + if let Some(ec) = self.errors.get(hash)? { + let mut ec = ErrorCounter::decode(&ec); + if ec.errors > 0 { + ec.last_try = now - ec.delay_msec(); + self.errors.insert(hash, ec.encode())?; + self.put_to_resync_at(hash, now)?; + return Ok(()); + } + } + Err(Error::Message(format!( + "Block {:?} was not in an errored state", + hash + ))) + } + // ---- Resync loop ---- // This part manages a queue of blocks that need to be @@ -257,7 +276,7 @@ impl BlockResyncManager { if let Err(e) = &res { manager.metrics.resync_error_counter.add(1); - warn!("Error when resyncing {:?}: {}", hash, e); + error!("Error when resyncing {:?}: {}", hash, e); let err_counter = match self.errors.get(hash.as_slice())? { Some(ec) => ErrorCounter::decode(&ec).add1(now + 1), @@ -477,27 +496,22 @@ impl Worker for ResyncWorker { format!("Block resync worker #{}", self.index + 1) } - fn info(&self) -> Option<String> { + fn status(&self) -> WorkerStatus { let persisted = self.manager.resync.persisted.load(); if self.index >= persisted.n_workers { - return Some("(unused)".into()); - } - - let mut ret = vec![]; - ret.push(format!("tranquility = {}", persisted.tranquility)); - - let qlen = self.manager.resync.queue_len().unwrap_or(0); - if qlen > 0 { - ret.push(format!("{} blocks in queue", qlen)); + return WorkerStatus { + freeform: vec!["This worker is currently disabled".into()], + ..Default::default() + }; } - let elen = self.manager.resync.errors_len().unwrap_or(0); - if elen > 0 { - ret.push(format!("{} blocks in error state", elen)); + WorkerStatus { + queue_length: Some(self.manager.resync.queue_len().unwrap_or(0) as u64), + tranquility: Some(persisted.tranquility), + persistent_errors: Some(self.manager.resync.errors_len().unwrap_or(0) as u64), + ..Default::default() } - - Some(ret.join(", ")) } async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { @@ -527,7 +541,7 @@ impl Worker for ResyncWorker { } } - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { + async fn wait_for_work(&mut self) -> WorkerState { while self.index >= self.manager.resync.persisted.load().n_workers { self.manager.resync.notify.notified().await } @@ -545,9 +559,9 @@ impl Worker for ResyncWorker { /// and the time of the last try. /// Used to implement exponential backoff. #[derive(Clone, Copy, Debug)] -struct ErrorCounter { - errors: u64, - last_try: u64, +pub(crate) struct ErrorCounter { + pub(crate) errors: u64, + pub(crate) last_try: u64, } impl ErrorCounter { @@ -558,12 +572,13 @@ impl ErrorCounter { } } - fn decode(data: &[u8]) -> Self { + pub(crate) fn decode(data: &[u8]) -> Self { Self { errors: u64::from_be_bytes(data[0..8].try_into().unwrap()), last_try: u64::from_be_bytes(data[8..16].try_into().unwrap()), } } + fn encode(&self) -> Vec<u8> { [ u64::to_be_bytes(self.errors), @@ -583,7 +598,8 @@ impl ErrorCounter { (RESYNC_RETRY_DELAY.as_millis() as u64) << std::cmp::min(self.errors - 1, RESYNC_RETRY_DELAY_MAX_BACKOFF_POWER) } - fn next_try(&self) -> u64 { + + pub(crate) fn next_try(&self) -> u64 { self.last_try + self.delay_msec() } } diff --git a/src/db/Cargo.toml b/src/db/Cargo.toml index 82cf49dc..c479d9d1 100644 --- a/src/db/Cargo.toml +++ b/src/db/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_db" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" diff --git a/src/db/lib.rs b/src/db/lib.rs index 5304c195..22bd9364 100644 --- a/src/db/lib.rs +++ b/src/db/lib.rs @@ -178,6 +178,10 @@ impl Tree { pub fn len(&self) -> Result<usize> { self.0.len(self.1) } + #[inline] + pub fn fast_len(&self) -> Result<Option<usize>> { + self.0.fast_len(self.1) + } #[inline] pub fn first(&self) -> Result<Option<(Value, Value)>> { @@ -320,6 +324,9 @@ pub(crate) trait IDb: Send + Sync { fn get(&self, tree: usize, key: &[u8]) -> Result<Option<Value>>; fn len(&self, tree: usize) -> Result<usize>; + fn fast_len(&self, _tree: usize) -> Result<Option<usize>> { + Ok(None) + } fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>>; fn remove(&self, tree: usize, key: &[u8]) -> Result<Option<Value>>; diff --git a/src/db/lmdb_adapter.rs b/src/db/lmdb_adapter.rs index c036c990..31956612 100644 --- a/src/db/lmdb_adapter.rs +++ b/src/db/lmdb_adapter.rs @@ -121,6 +121,10 @@ impl IDb for LmdbDb { Ok(tree.len(&tx)?.try_into().unwrap()) } + fn fast_len(&self, tree: usize) -> Result<Option<usize>> { + Ok(Some(self.len(tree)?)) + } + fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>> { let tree = self.get_tree(tree)?; let mut tx = self.db.write_txn()?; diff --git a/src/db/sqlite_adapter.rs b/src/db/sqlite_adapter.rs index 886fda6e..63b4506e 100644 --- a/src/db/sqlite_adapter.rs +++ b/src/db/sqlite_adapter.rs @@ -144,6 +144,10 @@ impl IDb for SqliteDb { } } + fn fast_len(&self, tree: usize) -> Result<Option<usize>> { + Ok(Some(self.len(tree)?)) + } + fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<Option<Value>> { trace!("insert {}: lock db", tree); let this = self.0.lock().unwrap(); diff --git a/src/garage/Cargo.toml b/src/garage/Cargo.toml index f9d7cf3a..b43b0242 100644 --- a/src/garage/Cargo.toml +++ b/src/garage/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -21,14 +21,14 @@ path = "tests/lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_db = { version = "0.8.0", path = "../db" } -garage_api = { version = "0.8.0", path = "../api" } -garage_block = { version = "0.8.0", path = "../block" } -garage_model = { version = "0.8.0", path = "../model" } -garage_rpc = { version = "0.8.0", path = "../rpc" } -garage_table = { version = "0.8.0", path = "../table" } -garage_util = { version = "0.8.0", path = "../util" } -garage_web = { version = "0.8.0", path = "../web" } +garage_db = { version = "0.8.1", path = "../db" } +garage_api = { version = "0.8.1", path = "../api" } +garage_block = { version = "0.8.1", path = "../block" } +garage_model = { version = "0.8.1", path = "../model" } +garage_rpc = { version = "0.8.1", path = "../rpc" } +garage_table = { version = "0.8.1", path = "../table" } +garage_util = { version = "0.8.1", path = "../util" } +garage_web = { version = "0.8.1", path = "../web" } backtrace = "0.3" bytes = "1.0" @@ -42,7 +42,6 @@ rand = "0.8" async-trait = "0.1.7" sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" } -rmp-serde = "0.15" serde = { version = "1.0", default-features = false, features = ["derive", "rc"] } serde_bytes = "0.11" structopt = { version = "0.3", default-features = false } @@ -74,7 +73,7 @@ base64 = "0.13" [features] -default = [ "bundled-libs", "metrics", "sled" ] +default = [ "bundled-libs", "metrics", "sled", "k2v" ] k2v = [ "garage_util/k2v", "garage_api/k2v" ] diff --git a/src/garage/admin.rs b/src/garage/admin.rs index e973cfe7..c669b5e6 100644 --- a/src/garage/admin.rs +++ b/src/garage/admin.rs @@ -5,9 +5,11 @@ use std::sync::Arc; use async_trait::async_trait; use serde::{Deserialize, Serialize}; +use garage_util::background::BackgroundRunner; use garage_util::crdt::*; use garage_util::data::*; use garage_util::error::Error as GarageError; +use garage_util::formater::format_table_to_string; use garage_util::time::*; use garage_table::replication::*; @@ -15,6 +17,7 @@ use garage_table::*; use garage_rpc::*; +use garage_block::manager::BlockResyncErrorInfo; use garage_block::repair::ScrubWorkerCommand; use garage_model::bucket_alias_table::*; @@ -24,6 +27,8 @@ use garage_model::helper::error::{Error, OkOrBadRequest}; use garage_model::key_table::*; use garage_model::migrate::Migrate; use garage_model::permission::*; +use garage_model::s3::object_table::*; +use garage_model::s3::version_table::Version; use crate::cli::*; use crate::repair::online::launch_online_repair; @@ -38,7 +43,8 @@ pub enum AdminRpc { LaunchRepair(RepairOpt), Migrate(MigrateOpt), Stats(StatsOpt), - Worker(WorkerOpt), + Worker(WorkerOperation), + BlockOperation(BlockOperation), // Replies Ok(String), @@ -54,6 +60,13 @@ pub enum AdminRpc { HashMap<usize, garage_util::background::WorkerInfo>, WorkerListOpt, ), + WorkerInfo(usize, garage_util::background::WorkerInfo), + BlockErrorList(Vec<BlockResyncErrorInfo>), + BlockInfo { + hash: Hash, + refcount: u64, + versions: Vec<Result<Version, Uuid>>, + }, } impl Rpc for AdminRpc { @@ -62,17 +75,24 @@ impl Rpc for AdminRpc { pub struct AdminRpcHandler { garage: Arc<Garage>, + background: Arc<BackgroundRunner>, endpoint: Arc<Endpoint<AdminRpc, Self>>, } impl AdminRpcHandler { - pub fn new(garage: Arc<Garage>) -> Arc<Self> { + pub fn new(garage: Arc<Garage>, background: Arc<BackgroundRunner>) -> Arc<Self> { let endpoint = garage.system.netapp.endpoint(ADMIN_RPC_PATH.into()); - let admin = Arc::new(Self { garage, endpoint }); + let admin = Arc::new(Self { + garage, + background, + endpoint, + }); admin.endpoint.set_handler(admin.clone()); admin } + // ================ BUCKET COMMANDS ==================== + async fn handle_bucket_cmd(&self, cmd: &BucketOperation) -> Result<AdminRpc, Error> { match cmd { BucketOperation::List => self.handle_list_buckets().await, @@ -551,6 +571,8 @@ impl AdminRpcHandler { Ok(AdminRpc::Ok(ret)) } + // ================ KEY COMMANDS ==================== + async fn handle_key_cmd(&self, cmd: &KeyOperation) -> Result<AdminRpc, Error> { match cmd { KeyOperation::List => self.handle_list_keys().await, @@ -688,6 +710,8 @@ impl AdminRpcHandler { Ok(AdminRpc::KeyInfo(key, relevant_buckets)) } + // ================ MIGRATION COMMANDS ==================== + async fn handle_migrate(self: &Arc<Self>, opt: MigrateOpt) -> Result<AdminRpc, Error> { if !opt.yes { return Err(Error::BadRequest( @@ -704,6 +728,8 @@ impl AdminRpcHandler { Ok(AdminRpc::Ok("Migration successfull.".into())) } + // ================ REPAIR COMMANDS ==================== + async fn handle_launch_repair(self: &Arc<Self>, opt: RepairOpt) -> Result<AdminRpc, Error> { if !opt.yes { return Err(Error::BadRequest( @@ -739,7 +765,7 @@ impl AdminRpcHandler { ))) } } else { - launch_online_repair(self.garage.clone(), opt).await; + launch_online_repair(&self.garage, &self.background, opt).await?; Ok(AdminRpc::Ok(format!( "Repair launched on {:?}", self.garage.system.id @@ -747,6 +773,8 @@ impl AdminRpcHandler { } } + // ================ STATS COMMANDS ==================== + async fn handle_stats(&self, opt: StatsOpt) -> Result<AdminRpc, Error> { if opt.all_nodes { let mut ret = String::new(); @@ -763,11 +791,12 @@ impl AdminRpcHandler { match self .endpoint .call(&node_id, AdminRpc::Stats(opt), PRIO_NORMAL) - .await? + .await { - Ok(AdminRpc::Ok(s)) => writeln!(&mut ret, "{}", s).unwrap(), - Ok(x) => writeln!(&mut ret, "Bad answer: {:?}", x).unwrap(), - Err(e) => writeln!(&mut ret, "Error: {}", e).unwrap(), + Ok(Ok(AdminRpc::Ok(s))) => writeln!(&mut ret, "{}", s).unwrap(), + Ok(Ok(x)) => writeln!(&mut ret, "Bad answer: {:?}", x).unwrap(), + Ok(Err(e)) => writeln!(&mut ret, "Remote error: {}", e).unwrap(), + Err(e) => writeln!(&mut ret, "Network error: {}", e).unwrap(), } } Ok(AdminRpc::Ok(ret)) @@ -787,6 +816,7 @@ impl AdminRpcHandler { .unwrap_or_else(|| "(unknown)".into()), ) .unwrap(); + writeln!(&mut ret, "\nDatabase engine: {}", self.garage.db.engine()).unwrap(); // Gather ring statistics @@ -805,21 +835,38 @@ impl AdminRpcHandler { writeln!(&mut ret, " {:?} {}", n, c).unwrap(); } - self.gather_table_stats(&mut ret, &self.garage.bucket_table, &opt)?; - self.gather_table_stats(&mut ret, &self.garage.key_table, &opt)?; - self.gather_table_stats(&mut ret, &self.garage.object_table, &opt)?; - self.gather_table_stats(&mut ret, &self.garage.version_table, &opt)?; - self.gather_table_stats(&mut ret, &self.garage.block_ref_table, &opt)?; + // Gather table statistics + let mut table = vec![" Table\tItems\tMklItems\tMklTodo\tGcTodo".into()]; + table.push(self.gather_table_stats(&self.garage.bucket_table, opt.detailed)?); + table.push(self.gather_table_stats(&self.garage.key_table, opt.detailed)?); + table.push(self.gather_table_stats(&self.garage.object_table, opt.detailed)?); + table.push(self.gather_table_stats(&self.garage.version_table, opt.detailed)?); + table.push(self.gather_table_stats(&self.garage.block_ref_table, opt.detailed)?); + write!( + &mut ret, + "\nTable stats:\n{}", + format_table_to_string(table) + ) + .unwrap(); + // Gather block manager statistics writeln!(&mut ret, "\nBlock manager stats:").unwrap(); - if opt.detailed { - writeln!( - &mut ret, - " number of RC entries (~= number of blocks): {}", - self.garage.block_manager.rc_len()? - ) - .unwrap(); - } + let rc_len = if opt.detailed { + self.garage.block_manager.rc_len()?.to_string() + } else { + self.garage + .block_manager + .rc_fast_len()? + .map(|x| x.to_string()) + .unwrap_or_else(|| "NC".into()) + }; + + writeln!( + &mut ret, + " number of RC entries (~= number of blocks): {}", + rc_len + ) + .unwrap(); writeln!( &mut ret, " resync queue length: {}", @@ -833,67 +880,83 @@ impl AdminRpcHandler { ) .unwrap(); + if !opt.detailed { + writeln!(&mut ret, "\nIf values are missing (marked as NC), consider adding the --detailed flag - this will be slow.").unwrap(); + } + Ok(ret) } fn gather_table_stats<F, R>( &self, - to: &mut String, t: &Arc<Table<F, R>>, - opt: &StatsOpt, - ) -> Result<(), Error> + detailed: bool, + ) -> Result<String, Error> where F: TableSchema + 'static, R: TableReplication + 'static, { - writeln!(to, "\nTable stats for {}", F::TABLE_NAME).unwrap(); - if opt.detailed { - writeln!( - to, - " number of items: {}", - t.data.store.len().map_err(GarageError::from)? + let (data_len, mkl_len) = if detailed { + ( + t.data.store.len().map_err(GarageError::from)?.to_string(), + t.merkle_updater.merkle_tree_len()?.to_string(), ) - .unwrap(); - writeln!( - to, - " Merkle tree size: {}", - t.merkle_updater.merkle_tree_len()? + } else { + ( + t.data + .store + .fast_len() + .map_err(GarageError::from)? + .map(|x| x.to_string()) + .unwrap_or_else(|| "NC".into()), + t.merkle_updater + .merkle_tree_fast_len()? + .map(|x| x.to_string()) + .unwrap_or_else(|| "NC".into()), ) - .unwrap(); - } - writeln!( - to, - " Merkle updater todo queue length: {}", - t.merkle_updater.todo_len()? - ) - .unwrap(); - writeln!(to, " GC todo queue length: {}", t.data.gc_todo_len()?).unwrap(); + }; - Ok(()) + Ok(format!( + " {}\t{}\t{}\t{}\t{}", + F::TABLE_NAME, + data_len, + mkl_len, + t.merkle_updater.todo_len()?, + t.data.gc_todo_len()? + )) } - // ---- + // ================ WORKER COMMANDS ==================== - async fn handle_worker_cmd(&self, opt: WorkerOpt) -> Result<AdminRpc, Error> { - match opt.cmd { - WorkerCmd::List { opt } => { - let workers = self.garage.background.get_worker_info(); - Ok(AdminRpc::WorkerList(workers, opt)) + async fn handle_worker_cmd(&self, cmd: &WorkerOperation) -> Result<AdminRpc, Error> { + match cmd { + WorkerOperation::List { opt } => { + let workers = self.background.get_worker_info(); + Ok(AdminRpc::WorkerList(workers, *opt)) + } + WorkerOperation::Info { tid } => { + let info = self + .background + .get_worker_info() + .get(tid) + .ok_or_bad_request(format!("No worker with TID {}", tid))? + .clone(); + Ok(AdminRpc::WorkerInfo(*tid, info)) } - WorkerCmd::Set { opt } => match opt { + WorkerOperation::Set { opt } => match opt { WorkerSetCmd::ScrubTranquility { tranquility } => { - let scrub_command = ScrubWorkerCommand::SetTranquility(tranquility); + let scrub_command = ScrubWorkerCommand::SetTranquility(*tranquility); self.garage .block_manager .send_scrub_command(scrub_command) - .await; + .await?; Ok(AdminRpc::Ok("Scrub tranquility updated".into())) } - WorkerSetCmd::ResyncNWorkers { n_workers } => { + WorkerSetCmd::ResyncWorkerCount { worker_count } => { self.garage .block_manager .resync - .set_n_workers(n_workers) + .set_n_workers(*worker_count) .await?; Ok(AdminRpc::Ok("Number of resync workers updated".into())) } @@ -901,13 +964,154 @@ impl AdminRpcHandler { self.garage .block_manager .resync - .set_tranquility(tranquility) + .set_tranquility(*tranquility) .await?; Ok(AdminRpc::Ok("Resync tranquility updated".into())) } }, } } + + // ================ BLOCK COMMANDS ==================== + + async fn handle_block_cmd(&self, cmd: &BlockOperation) -> Result<AdminRpc, Error> { + match cmd { + BlockOperation::ListErrors => Ok(AdminRpc::BlockErrorList( + self.garage.block_manager.list_resync_errors()?, + )), + BlockOperation::Info { hash } => { + let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?; + let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?; + let refcount = self.garage.block_manager.get_block_rc(&hash)?; + let block_refs = self + .garage + .block_ref_table + .get_range(&hash, None, None, 10000, Default::default()) + .await?; + let mut versions = vec![]; + for br in block_refs { + if let Some(v) = self + .garage + .version_table + .get(&br.version, &EmptyKey) + .await? + { + versions.push(Ok(v)); + } else { + versions.push(Err(br.version)); + } + } + Ok(AdminRpc::BlockInfo { + hash, + refcount, + versions, + }) + } + BlockOperation::RetryNow { all, blocks } => { + if *all { + if !blocks.is_empty() { + return Err(Error::BadRequest( + "--all was specified, cannot also specify blocks".into(), + )); + } + let blocks = self.garage.block_manager.list_resync_errors()?; + for b in blocks.iter() { + self.garage.block_manager.resync.clear_backoff(&b.hash)?; + } + Ok(AdminRpc::Ok(format!( + "{} blocks returned in queue for a retry now (check logs to see results)", + blocks.len() + ))) + } else { + for hash in blocks { + let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?; + let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?; + self.garage.block_manager.resync.clear_backoff(&hash)?; + } + Ok(AdminRpc::Ok(format!( + "{} blocks returned in queue for a retry now (check logs to see results)", + blocks.len() + ))) + } + } + BlockOperation::Purge { yes, blocks } => { + if !yes { + return Err(Error::BadRequest( + "Pass the --yes flag to confirm block purge operation.".into(), + )); + } + + let mut obj_dels = 0; + let mut ver_dels = 0; + + for hash in blocks { + let hash = hex::decode(hash).ok_or_bad_request("invalid hash")?; + let hash = Hash::try_from(&hash).ok_or_bad_request("invalid hash")?; + let block_refs = self + .garage + .block_ref_table + .get_range(&hash, None, None, 10000, Default::default()) + .await?; + + for br in block_refs { + let version = match self + .garage + .version_table + .get(&br.version, &EmptyKey) + .await? + { + Some(v) => v, + None => continue, + }; + + if let Some(object) = self + .garage + .object_table + .get(&version.bucket_id, &version.key) + .await? + { + let ov = object.versions().iter().rev().find(|v| v.is_complete()); + if let Some(ov) = ov { + if ov.uuid == br.version { + let del_uuid = gen_uuid(); + let deleted_object = Object::new( + version.bucket_id, + version.key.clone(), + vec![ObjectVersion { + uuid: del_uuid, + timestamp: ov.timestamp + 1, + state: ObjectVersionState::Complete( + ObjectVersionData::DeleteMarker, + ), + }], + ); + self.garage.object_table.insert(&deleted_object).await?; + obj_dels += 1; + } + } + } + + if !version.deleted.get() { + let deleted_version = Version::new( + version.uuid, + version.bucket_id, + version.key.clone(), + true, + ); + self.garage.version_table.insert(&deleted_version).await?; + ver_dels += 1; + } + } + } + Ok(AdminRpc::Ok(format!( + "{} blocks were purged: {} object deletion markers added, {} versions marked deleted", + blocks.len(), + obj_dels, + ver_dels + ))) + } + } + } } #[async_trait] @@ -923,7 +1127,8 @@ impl EndpointHandler<AdminRpc> for AdminRpcHandler { AdminRpc::Migrate(opt) => self.handle_migrate(opt.clone()).await, AdminRpc::LaunchRepair(opt) => self.handle_launch_repair(opt.clone()).await, AdminRpc::Stats(opt) => self.handle_stats(opt.clone()).await, - AdminRpc::Worker(opt) => self.handle_worker_cmd(opt.clone()).await, + AdminRpc::Worker(wo) => self.handle_worker_cmd(wo).await, + AdminRpc::BlockOperation(bo) => self.handle_block_cmd(bo).await, m => Err(GarageError::unexpected_rpc_message(m).into()), } } diff --git a/src/garage/cli/cmd.rs b/src/garage/cli/cmd.rs index e352ddf2..0d180ecd 100644 --- a/src/garage/cli/cmd.rs +++ b/src/garage/cli/cmd.rs @@ -41,6 +41,9 @@ pub async fn cli_command_dispatch( } Command::Stats(so) => cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::Stats(so)).await, Command::Worker(wo) => cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::Worker(wo)).await, + Command::Block(bo) => { + cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::BlockOperation(bo)).await + } _ => unreachable!(), } } @@ -186,7 +189,20 @@ pub async fn cmd_admin( print_key_info(&key, &rb); } AdminRpc::WorkerList(wi, wlo) => { - print_worker_info(wi, wlo); + print_worker_list(wi, wlo); + } + AdminRpc::WorkerInfo(tid, wi) => { + print_worker_info(tid, wi); + } + AdminRpc::BlockErrorList(el) => { + print_block_error_list(el); + } + AdminRpc::BlockInfo { + hash, + refcount, + versions, + } => { + print_block_info(hash, refcount, versions); } r => { error!("Unexpected response: {:?}", r); diff --git a/src/garage/cli/structs.rs b/src/garage/cli/structs.rs index 49a1f267..825c1813 100644 --- a/src/garage/cli/structs.rs +++ b/src/garage/cli/structs.rs @@ -49,7 +49,11 @@ pub enum Command { /// Manage background workers #[structopt(name = "worker", version = garage_version())] - Worker(WorkerOpt), + Worker(WorkerOperation), + + /// Low-level debug operations on data blocks + #[structopt(name = "block", version = garage_version())] + Block(BlockOperation), } #[derive(StructOpt, Debug)] @@ -513,20 +517,17 @@ pub struct StatsOpt { pub detailed: bool, } -#[derive(Serialize, Deserialize, StructOpt, Debug, Clone)] -pub struct WorkerOpt { - #[structopt(subcommand)] - pub cmd: WorkerCmd, -} - #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)] -pub enum WorkerCmd { +pub enum WorkerOperation { /// List all workers on Garage node #[structopt(name = "list", version = garage_version())] List { #[structopt(flatten)] opt: WorkerListOpt, }, + /// Get detailed information about a worker + #[structopt(name = "info", version = garage_version())] + Info { tid: usize }, /// Set worker parameter #[structopt(name = "set", version = garage_version())] Set { @@ -551,9 +552,41 @@ pub enum WorkerSetCmd { #[structopt(name = "scrub-tranquility", version = garage_version())] ScrubTranquility { tranquility: u32 }, /// Set number of concurrent block resync workers - #[structopt(name = "resync-n-workers", version = garage_version())] - ResyncNWorkers { n_workers: usize }, + #[structopt(name = "resync-worker-count", version = garage_version())] + ResyncWorkerCount { worker_count: usize }, /// Set tranquility of block resync operations #[structopt(name = "resync-tranquility", version = garage_version())] ResyncTranquility { tranquility: u32 }, } + +#[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)] +pub enum BlockOperation { + /// List all blocks that currently have a resync error + #[structopt(name = "list-errors", version = garage_version())] + ListErrors, + /// Get detailed information about a single block + #[structopt(name = "info", version = garage_version())] + Info { + /// Hash of the block for which to retrieve information + hash: String, + }, + /// Retry now the resync of one or many blocks + #[structopt(name = "retry-now", version = garage_version())] + RetryNow { + /// Retry all blocks that have a resync error + #[structopt(long = "all")] + all: bool, + /// Hashes of the block to retry to resync now + blocks: Vec<String>, + }, + /// Delete all objects referencing a missing block + #[structopt(name = "purge", version = garage_version())] + Purge { + /// Mandatory to confirm this operation + #[structopt(long = "yes")] + yes: bool, + /// Hashes of the block to purge + #[structopt(required = true)] + blocks: Vec<String>, + }, +} diff --git a/src/garage/cli/util.rs b/src/garage/cli/util.rs index 396938ae..63fd9eba 100644 --- a/src/garage/cli/util.rs +++ b/src/garage/cli/util.rs @@ -3,14 +3,17 @@ use std::time::Duration; use garage_util::background::*; use garage_util::crdt::*; -use garage_util::data::Uuid; +use garage_util::data::*; use garage_util::error::*; use garage_util::formater::format_table; use garage_util::time::*; +use garage_block::manager::BlockResyncErrorInfo; + use garage_model::bucket_table::*; use garage_model::key_table::*; use garage_model::s3::object_table::{BYTES, OBJECTS, UNFINISHED_UPLOADS}; +use garage_model::s3::version_table::Version; use crate::cli::structs::WorkerListOpt; @@ -241,7 +244,7 @@ pub fn find_matching_node( } } -pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) { +pub fn print_worker_list(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) { let mut wi = wi.into_iter().collect::<Vec<_>>(); wi.sort_by_key(|(tid, info)| { ( @@ -254,7 +257,7 @@ pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) { ) }); - let mut table = vec![]; + let mut table = vec!["TID\tState\tName\tTranq\tDone\tQueue\tErrors\tConsec\tLast".to_string()]; for (tid, info) in wi.iter() { if wlo.busy && !matches!(info.state, WorkerState::Busy | WorkerState::Throttled(_)) { continue; @@ -263,33 +266,147 @@ pub fn print_worker_info(wi: HashMap<usize, WorkerInfo>, wlo: WorkerListOpt) { continue; } - table.push(format!("{}\t{}\t{}", tid, info.state, info.name)); - if let Some(i) = &info.info { - table.push(format!("\t\t {}", i)); - } let tf = timeago::Formatter::new(); - let (err_ago, err_msg) = info + let err_ago = info .last_error .as_ref() - .map(|(m, t)| { - ( - tf.convert(Duration::from_millis(now_msec() - t)), - m.as_str(), - ) - }) - .unwrap_or(("(?) ago".into(), "(?)")); - if info.consecutive_errors > 0 { + .map(|(_, t)| tf.convert(Duration::from_millis(now_msec() - t))) + .unwrap_or_default(); + let (total_err, consec_err) = if info.errors > 0 { + (info.errors.to_string(), info.consecutive_errors.to_string()) + } else { + ("-".into(), "-".into()) + }; + + table.push(format!( + "{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}", + tid, + info.state, + info.name, + info.status + .tranquility + .as_ref() + .map(ToString::to_string) + .unwrap_or_else(|| "-".into()), + info.status.progress.as_deref().unwrap_or("-"), + info.status + .queue_length + .as_ref() + .map(ToString::to_string) + .unwrap_or_else(|| "-".into()), + total_err, + consec_err, + err_ago, + )); + } + format_table(table); +} + +pub fn print_worker_info(tid: usize, info: WorkerInfo) { + let mut table = vec![]; + table.push(format!("Task id:\t{}", tid)); + table.push(format!("Worker name:\t{}", info.name)); + match info.state { + WorkerState::Throttled(t) => { table.push(format!( - "\t\t {} consecutive errors ({} total), last {}", - info.consecutive_errors, info.errors, err_ago, + "Worker state:\tBusy (throttled, paused for {:.3}s)", + t )); - table.push(format!("\t\t {}", err_msg)); - } else if info.errors > 0 { - table.push(format!("\t\t ({} errors, last {})", info.errors, err_ago,)); - if wlo.errors { - table.push(format!("\t\t {}", err_msg)); + } + s => { + table.push(format!("Worker state:\t{}", s)); + } + }; + if let Some(tql) = info.status.tranquility { + table.push(format!("Tranquility:\t{}", tql)); + } + + table.push("".into()); + table.push(format!("Total errors:\t{}", info.errors)); + table.push(format!("Consecutive errs:\t{}", info.consecutive_errors)); + if let Some((s, t)) = info.last_error { + table.push(format!("Last error:\t{}", s)); + let tf = timeago::Formatter::new(); + table.push(format!( + "Last error time:\t{}", + tf.convert(Duration::from_millis(now_msec() - t)) + )); + } + + table.push("".into()); + if let Some(p) = info.status.progress { + table.push(format!("Progress:\t{}", p)); + } + if let Some(ql) = info.status.queue_length { + table.push(format!("Queue length:\t{}", ql)); + } + if let Some(pe) = info.status.persistent_errors { + table.push(format!("Persistent errors:\t{}", pe)); + } + + for (i, s) in info.status.freeform.iter().enumerate() { + if i == 0 { + if table.last() != Some(&"".into()) { + table.push("".into()); } + table.push(format!("Message:\t{}", s)); + } else { + table.push(format!("\t{}", s)); } } format_table(table); } + +pub fn print_block_error_list(el: Vec<BlockResyncErrorInfo>) { + let now = now_msec(); + let tf = timeago::Formatter::new(); + let mut tf2 = timeago::Formatter::new(); + tf2.ago(""); + + let mut table = vec!["Hash\tRC\tErrors\tLast error\tNext try".into()]; + for e in el { + table.push(format!( + "{}\t{}\t{}\t{}\tin {}", + hex::encode(e.hash.as_slice()), + e.refcount, + e.error_count, + tf.convert(Duration::from_millis(now - e.last_try)), + tf2.convert(Duration::from_millis(e.next_try - now)) + )); + } + format_table(table); +} + +pub fn print_block_info(hash: Hash, refcount: u64, versions: Vec<Result<Version, Uuid>>) { + println!("Block hash: {}", hex::encode(hash.as_slice())); + println!("Refcount: {}", refcount); + println!(); + + let mut table = vec!["Version\tBucket\tKey\tDeleted".into()]; + let mut nondeleted_count = 0; + for v in versions.iter() { + match v { + Ok(ver) => { + table.push(format!( + "{:?}\t{:?}\t{}\t{:?}", + ver.uuid, + ver.bucket_id, + ver.key, + ver.deleted.get() + )); + if !ver.deleted.get() { + nondeleted_count += 1; + } + } + Err(vh) => { + table.push(format!("{:?}\t\t\tyes", vh)); + } + } + } + format_table(table); + + if refcount != nondeleted_count { + println!(); + println!("Warning: refcount does not match number of non-deleted versions"); + } +} diff --git a/src/garage/main.rs b/src/garage/main.rs index 8e64273f..cd1d6228 100644 --- a/src/garage/main.rs +++ b/src/garage/main.rs @@ -130,9 +130,16 @@ async fn main() { std::process::abort(); })); + // Parse arguments and dispatch command line + let opt = Opt::from_clap(&Opt::clap().version(version.as_str()).get_matches()); + // Initialize logging as well as other libraries used in Garage if std::env::var("RUST_LOG").is_err() { - std::env::set_var("RUST_LOG", "netapp=info,garage=info") + let default_log = match &opt.cmd { + Command::Server => "netapp=info,garage=info", + _ => "netapp=warn,garage=warn", + }; + std::env::set_var("RUST_LOG", default_log) } tracing_subscriber::fmt() .with_writer(std::io::stderr) @@ -140,9 +147,6 @@ async fn main() { .init(); sodiumoxide::init().expect("Unable to init sodiumoxide"); - // Parse arguments and dispatch command line - let opt = Opt::from_clap(&Opt::clap().version(version.as_str()).get_matches()); - let res = match opt.cmd { Command::Server => server::run_server(opt.config_file).await, Command::OfflineRepair(repair_opt) => { @@ -185,9 +189,9 @@ async fn cli_command(opt: Opt) -> Result<(), Error> { let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, sk); // Find and parse the address of the target host - let (id, addr) = if let Some(h) = opt.rpc_host { + let (id, addr, is_default_addr) = if let Some(h) = opt.rpc_host { let (id, addrs) = parse_and_resolve_peer_addr(&h).ok_or_else(|| format!("Invalid RPC remote node identifier: {}. Expected format is <pubkey>@<IP or hostname>:<port>.", h))?; - (id, addrs[0]) + (id, addrs[0], false) } else { let node_id = garage_rpc::system::read_node_id(&config.as_ref().unwrap().metadata_dir) .err_context(READ_KEY_ERROR)?; @@ -198,24 +202,26 @@ async fn cli_command(opt: Opt) -> Result<(), Error> { .ok_or_message("unable to resolve rpc_public_addr specified in config file")? .next() .ok_or_message("unable to resolve rpc_public_addr specified in config file")?; - (node_id, a) + (node_id, a, false) } else { let default_addr = SocketAddr::new( "127.0.0.1".parse().unwrap(), config.as_ref().unwrap().rpc_bind_addr.port(), ); - warn!( - "Trying to contact Garage node at default address {}", - default_addr - ); - warn!("If this doesn't work, consider adding rpc_public_addr in your config file or specifying the -h command line parameter."); - (node_id, default_addr) + (node_id, default_addr, true) } }; // Connect to target host - netapp.clone().try_connect(addr, id).await - .err_context("Unable to connect to destination RPC host. Check that you are using the same value of rpc_secret as them, and that you have their correct public key.")?; + if let Err(e) = netapp.clone().try_connect(addr, id).await { + if is_default_addr { + warn!( + "Tried to contact Garage node at default address {}, which didn't work. If that address is wrong, consider setting rpc_public_addr in your config file.", + addr + ); + } + Err(e).err_context("Unable to connect to destination RPC host. Check that you are using the same value of rpc_secret as them, and that you have their correct public key.")?; + } let system_rpc_endpoint = netapp.endpoint::<SystemRpc, ()>(SYSTEM_RPC_PATH.into()); let admin_rpc_endpoint = netapp.endpoint::<AdminRpc, ()>(ADMIN_RPC_PATH.into()); diff --git a/src/garage/repair/offline.rs b/src/garage/repair/offline.rs index 7760a8bd..25193e4a 100644 --- a/src/garage/repair/offline.rs +++ b/src/garage/repair/offline.rs @@ -1,8 +1,5 @@ use std::path::PathBuf; -use tokio::sync::watch; - -use garage_util::background::*; use garage_util::config::*; use garage_util::error::*; @@ -20,12 +17,8 @@ pub async fn offline_repair(config_file: PathBuf, opt: OfflineRepairOpt) -> Resu info!("Loading configuration..."); let config = read_config(config_file)?; - info!("Initializing background runner..."); - let (done_tx, done_rx) = watch::channel(false); - let (background, await_background_done) = BackgroundRunner::new(16, done_rx); - info!("Initializing Garage main data store..."); - let garage = Garage::new(config.clone(), background)?; + let garage = Garage::new(config)?; info!("Launching repair operation..."); match opt.what { @@ -43,13 +36,7 @@ pub async fn offline_repair(config_file: PathBuf, opt: OfflineRepairOpt) -> Resu } } - info!("Repair operation finished, shutting down Garage internals..."); - done_tx.send(true).unwrap(); - drop(garage); - - await_background_done.await?; - - info!("Cleaning up..."); + info!("Repair operation finished, shutting down..."); Ok(()) } diff --git a/src/garage/repair/online.rs b/src/garage/repair/online.rs index e33cf097..627e3bf3 100644 --- a/src/garage/repair/online.rs +++ b/src/garage/repair/online.rs @@ -12,38 +12,37 @@ use garage_model::s3::version_table::*; use garage_table::*; use garage_util::background::*; use garage_util::error::Error; +use garage_util::migrate::Migrate; use crate::*; -pub async fn launch_online_repair(garage: Arc<Garage>, opt: RepairOpt) { +pub async fn launch_online_repair( + garage: &Arc<Garage>, + bg: &BackgroundRunner, + opt: RepairOpt, +) -> Result<(), Error> { match opt.what { RepairWhat::Tables => { info!("Launching a full sync of tables"); - garage.bucket_table.syncer.add_full_sync(); - garage.object_table.syncer.add_full_sync(); - garage.version_table.syncer.add_full_sync(); - garage.block_ref_table.syncer.add_full_sync(); - garage.key_table.syncer.add_full_sync(); + garage.bucket_table.syncer.add_full_sync()?; + garage.object_table.syncer.add_full_sync()?; + garage.version_table.syncer.add_full_sync()?; + garage.block_ref_table.syncer.add_full_sync()?; + garage.key_table.syncer.add_full_sync()?; } RepairWhat::Versions => { info!("Repairing the versions table"); - garage - .background - .spawn_worker(RepairVersionsWorker::new(garage.clone())); + bg.spawn_worker(RepairVersionsWorker::new(garage.clone())); } RepairWhat::BlockRefs => { info!("Repairing the block refs table"); - garage - .background - .spawn_worker(RepairBlockrefsWorker::new(garage.clone())); + bg.spawn_worker(RepairBlockrefsWorker::new(garage.clone())); } RepairWhat::Blocks => { info!("Repairing the stored blocks"); - garage - .background - .spawn_worker(garage_block::repair::RepairWorker::new( - garage.block_manager.clone(), - )); + bg.spawn_worker(garage_block::repair::RepairWorker::new( + garage.block_manager.clone(), + )); } RepairWhat::Scrub { cmd } => { let cmd = match cmd { @@ -56,9 +55,10 @@ pub async fn launch_online_repair(garage: Arc<Garage>, opt: RepairOpt) { } }; info!("Sending command to scrub worker: {:?}", cmd); - garage.block_manager.send_scrub_command(cmd).await; + garage.block_manager.send_scrub_command(cmd).await?; } } + Ok(()) } // ---- @@ -85,25 +85,23 @@ impl Worker for RepairVersionsWorker { "Version repair worker".into() } - fn info(&self) -> Option<String> { - Some(format!("{} items done", self.counter)) + fn status(&self) -> WorkerStatus { + WorkerStatus { + progress: Some(self.counter.to_string()), + ..Default::default() + } } async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { - let item_bytes = match self.garage.version_table.data.store.get_gt(&self.pos)? { - Some((k, v)) => { - self.pos = k; - v - } + let (item_bytes, next_pos) = match self.garage.version_table.data.store.get_gt(&self.pos)? { + Some((k, v)) => (v, k), None => { info!("repair_versions: finished, done {}", self.counter); return Ok(WorkerState::Done); } }; - self.counter += 1; - - let version = rmp_serde::decode::from_read_ref::<_, Version>(&item_bytes)?; + let version = Version::decode(&item_bytes).ok_or_message("Cannot decode Version")?; if !version.deleted.get() { let object = self .garage @@ -131,10 +129,13 @@ impl Worker for RepairVersionsWorker { } } + self.counter += 1; + self.pos = next_pos; + Ok(WorkerState::Busy) } - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { + async fn wait_for_work(&mut self) -> WorkerState { unreachable!() } } @@ -163,25 +164,24 @@ impl Worker for RepairBlockrefsWorker { "Block refs repair worker".into() } - fn info(&self) -> Option<String> { - Some(format!("{} items done", self.counter)) + fn status(&self) -> WorkerStatus { + WorkerStatus { + progress: Some(self.counter.to_string()), + ..Default::default() + } } async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { - let item_bytes = match self.garage.block_ref_table.data.store.get_gt(&self.pos)? { - Some((k, v)) => { - self.pos = k; - v - } - None => { - info!("repair_block_ref: finished, done {}", self.counter); - return Ok(WorkerState::Done); - } - }; - - self.counter += 1; + let (item_bytes, next_pos) = + match self.garage.block_ref_table.data.store.get_gt(&self.pos)? { + Some((k, v)) => (v, k), + None => { + info!("repair_block_ref: finished, done {}", self.counter); + return Ok(WorkerState::Done); + } + }; - let block_ref = rmp_serde::decode::from_read_ref::<_, BlockRef>(&item_bytes)?; + let block_ref = BlockRef::decode(&item_bytes).ok_or_message("Cannot decode BlockRef")?; if !block_ref.deleted.get() { let version = self .garage @@ -206,10 +206,13 @@ impl Worker for RepairBlockrefsWorker { } } + self.counter += 1; + self.pos = next_pos; + Ok(WorkerState::Busy) } - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { + async fn wait_for_work(&mut self) -> WorkerState { unreachable!() } } diff --git a/src/garage/server.rs b/src/garage/server.rs index d4099a97..16f1b625 100644 --- a/src/garage/server.rs +++ b/src/garage/server.rs @@ -35,12 +35,15 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> { #[cfg(feature = "metrics")] let metrics_exporter = opentelemetry_prometheus::exporter().init(); + info!("Initializing Garage main data store..."); + let garage = Garage::new(config.clone())?; + info!("Initializing background runner..."); let watch_cancel = watch_shutdown_signal(); - let (background, await_background_done) = BackgroundRunner::new(16, watch_cancel.clone()); + let (background, await_background_done) = BackgroundRunner::new(watch_cancel.clone()); - info!("Initializing Garage main data store..."); - let garage = Garage::new(config.clone(), background)?; + info!("Spawning Garage workers..."); + garage.spawn_workers(&background); if config.admin.trace_sink.is_some() { info!("Initialize tracing..."); @@ -63,7 +66,7 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> { let run_system = tokio::spawn(garage.system.clone().run(watch_cancel.clone())); info!("Create admin RPC handler..."); - AdminRpcHandler::new(garage.clone()); + AdminRpcHandler::new(garage.clone(), background.clone()); // ---- Launch public-facing API servers ---- diff --git a/src/garage/tests/bucket.rs b/src/garage/tests/bucket.rs index b32af068..9c363013 100644 --- a/src/garage/tests/bucket.rs +++ b/src/garage/tests/bucket.rs @@ -1,4 +1,5 @@ use crate::common; +use crate::common::ext::CommandExt; use aws_sdk_s3::model::BucketLocationConstraint; use aws_sdk_s3::output::DeleteBucketOutput; @@ -8,6 +9,27 @@ async fn test_bucket_all() { let bucket_name = "hello"; { + // Check bucket cannot be created if not authorized + ctx.garage + .command() + .args(["key", "deny"]) + .args(["--create-bucket", &ctx.garage.key.id]) + .quiet() + .expect_success_output("Could not deny key to create buckets"); + + // Try create bucket, should fail + let r = ctx.client.create_bucket().bucket(bucket_name).send().await; + assert!(r.is_err()); + } + { + // Now allow key to create bucket + ctx.garage + .command() + .args(["key", "allow"]) + .args(["--create-bucket", &ctx.garage.key.id]) + .quiet() + .expect_success_output("Could not deny key to create buckets"); + // Create bucket //@TODO check with an invalid bucket name + with an already existing bucket let r = ctx diff --git a/src/garage/tests/s3/streaming_signature.rs b/src/garage/tests/s3/streaming_signature.rs index c68f7dfc..48da7607 100644 --- a/src/garage/tests/s3/streaming_signature.rs +++ b/src/garage/tests/s3/streaming_signature.rs @@ -1,6 +1,7 @@ use std::collections::HashMap; use crate::common; +use crate::common::ext::CommandExt; use common::custom_requester::BodySignature; use hyper::Method; @@ -105,6 +106,13 @@ async fn test_create_bucket_streaming() { let ctx = common::context(); let bucket = "createbucket-streaming"; + ctx.garage + .command() + .args(["key", "allow"]) + .args(["--create-bucket", &ctx.garage.key.id]) + .quiet() + .expect_success_output("Could not allow key to create buckets"); + { // create bucket let _ = ctx diff --git a/src/k2v-client/Cargo.toml b/src/k2v-client/Cargo.toml index 9d2b4e30..f57ce849 100644 --- a/src/k2v-client/Cargo.toml +++ b/src/k2v-client/Cargo.toml @@ -22,7 +22,7 @@ tokio = "1.17.0" # cli deps clap = { version = "3.1.18", optional = true, features = ["derive", "env"] } -garage_util = { version = "0.8.0", path = "../util", optional = true } +garage_util = { version = "0.8.1", path = "../util", optional = true } [features] diff --git a/src/model/Cargo.toml b/src/model/Cargo.toml index 08baf81f..323c2d64 100644 --- a/src/model/Cargo.toml +++ b/src/model/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_model" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,11 +14,11 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_db = { version = "0.8.0", default-features = false, path = "../db" } -garage_rpc = { version = "0.8.0", path = "../rpc" } -garage_table = { version = "0.8.0", path = "../table" } -garage_block = { version = "0.8.0", path = "../block" } -garage_util = { version = "0.8.0", path = "../util" } +garage_db = { version = "0.8.1", default-features = false, path = "../db" } +garage_rpc = { version = "0.8.1", path = "../rpc" } +garage_table = { version = "0.8.1", path = "../table" } +garage_block = { version = "0.8.1", path = "../block" } +garage_util = { version = "0.8.1", path = "../util" } async-trait = "0.1.7" arc-swap = "1.0" @@ -30,7 +30,6 @@ tracing = "0.1.30" rand = "0.8" zstd = { version = "0.9", default-features = false } -rmp-serde = "0.15" serde = { version = "1.0", default-features = false, features = ["derive", "rc"] } serde_bytes = "0.11" diff --git a/src/model/bucket_alias_table.rs b/src/model/bucket_alias_table.rs index fcd1536e..54d7fbad 100644 --- a/src/model/bucket_alias_table.rs +++ b/src/model/bucket_alias_table.rs @@ -1,18 +1,26 @@ -use serde::{Deserialize, Serialize}; - use garage_util::data::*; use garage_table::crdt::*; use garage_table::*; -/// The bucket alias table holds the names given to buckets -/// in the global namespace. -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct BucketAlias { - name: String, - pub state: crdt::Lww<Option<Uuid>>, +mod v08 { + use garage_util::crdt; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + + /// The bucket alias table holds the names given to buckets + /// in the global namespace. + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct BucketAlias { + pub(super) name: String, + pub state: crdt::Lww<Option<Uuid>>, + } + + impl garage_util::migrate::InitialFormat for BucketAlias {} } +pub use v08::*; + impl BucketAlias { pub fn new(name: String, ts: u64, bucket_id: Option<Uuid>) -> Option<Self> { if !is_valid_bucket_name(&name) { diff --git a/src/model/bucket_table.rs b/src/model/bucket_table.rs index 7be42702..ac163736 100644 --- a/src/model/bucket_table.rs +++ b/src/model/bucket_table.rs @@ -1,5 +1,3 @@ -use serde::{Deserialize, Serialize}; - use garage_table::crdt::*; use garage_table::*; use garage_util::data::*; @@ -7,72 +5,83 @@ use garage_util::time::*; use crate::permission::BucketKeyPerm; -/// A bucket is a collection of objects -/// -/// Its parameters are not directly accessible as: -/// - It must be possible to merge paramaters, hence the use of a LWW CRDT. -/// - A bucket has 2 states, Present or Deleted and parameters make sense only if present. -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Bucket { - /// ID of the bucket - pub id: Uuid, - /// State, and configuration if not deleted, of the bucket - pub state: crdt::Deletable<BucketParams>, -} +mod v08 { + use crate::permission::BucketKeyPerm; + use garage_util::crdt; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + + /// A bucket is a collection of objects + /// + /// Its parameters are not directly accessible as: + /// - It must be possible to merge paramaters, hence the use of a LWW CRDT. + /// - A bucket has 2 states, Present or Deleted and parameters make sense only if present. + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Bucket { + /// ID of the bucket + pub id: Uuid, + /// State, and configuration if not deleted, of the bucket + pub state: crdt::Deletable<BucketParams>, + } -/// Configuration for a bucket -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct BucketParams { - /// Bucket's creation date - pub creation_date: u64, - /// Map of key with access to the bucket, and what kind of access they give - pub authorized_keys: crdt::Map<String, BucketKeyPerm>, - - /// Map of aliases that are or have been given to this bucket - /// in the global namespace - /// (not authoritative: this is just used as an indication to - /// map back to aliases when doing ListBuckets) - pub aliases: crdt::LwwMap<String, bool>, - /// Map of aliases that are or have been given to this bucket - /// in namespaces local to keys - /// key = (access key id, alias name) - pub local_aliases: crdt::LwwMap<(String, String), bool>, - - /// Whether this bucket is allowed for website access - /// (under all of its global alias names), - /// and if so, the website configuration XML document - pub website_config: crdt::Lww<Option<WebsiteConfig>>, - /// CORS rules - pub cors_config: crdt::Lww<Option<Vec<CorsRule>>>, - /// Bucket quotas - #[serde(default)] - pub quotas: crdt::Lww<BucketQuotas>, -} + /// Configuration for a bucket + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct BucketParams { + /// Bucket's creation date + pub creation_date: u64, + /// Map of key with access to the bucket, and what kind of access they give + pub authorized_keys: crdt::Map<String, BucketKeyPerm>, + + /// Map of aliases that are or have been given to this bucket + /// in the global namespace + /// (not authoritative: this is just used as an indication to + /// map back to aliases when doing ListBuckets) + pub aliases: crdt::LwwMap<String, bool>, + /// Map of aliases that are or have been given to this bucket + /// in namespaces local to keys + /// key = (access key id, alias name) + pub local_aliases: crdt::LwwMap<(String, String), bool>, + + /// Whether this bucket is allowed for website access + /// (under all of its global alias names), + /// and if so, the website configuration XML document + pub website_config: crdt::Lww<Option<WebsiteConfig>>, + /// CORS rules + pub cors_config: crdt::Lww<Option<Vec<CorsRule>>>, + /// Bucket quotas + #[serde(default)] + pub quotas: crdt::Lww<BucketQuotas>, + } -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct WebsiteConfig { - pub index_document: String, - pub error_document: Option<String>, -} + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct WebsiteConfig { + pub index_document: String, + pub error_document: Option<String>, + } -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct CorsRule { - pub id: Option<String>, - pub max_age_seconds: Option<u64>, - pub allow_origins: Vec<String>, - pub allow_methods: Vec<String>, - pub allow_headers: Vec<String>, - pub expose_headers: Vec<String>, -} + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct CorsRule { + pub id: Option<String>, + pub max_age_seconds: Option<u64>, + pub allow_origins: Vec<String>, + pub allow_methods: Vec<String>, + pub allow_headers: Vec<String>, + pub expose_headers: Vec<String>, + } -#[derive(Default, PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub struct BucketQuotas { - /// Maximum size in bytes (bucket size = sum of sizes of objects in the bucket) - pub max_size: Option<u64>, - /// Maximum number of non-deleted objects in the bucket - pub max_objects: Option<u64>, + #[derive(Default, PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] + pub struct BucketQuotas { + /// Maximum size in bytes (bucket size = sum of sizes of objects in the bucket) + pub max_size: Option<u64>, + /// Maximum number of non-deleted objects in the bucket + pub max_objects: Option<u64>, + } + + impl garage_util::migrate::InitialFormat for Bucket {} } +pub use v08::*; + impl AutoCrdt for BucketQuotas { const WARN_IF_DIFFERENT: bool = true; } diff --git a/src/model/garage.rs b/src/model/garage.rs index e34d034f..5bea6b4f 100644 --- a/src/model/garage.rs +++ b/src/model/garage.rs @@ -39,8 +39,6 @@ pub struct Garage { /// The local database pub db: db::Db, - /// A background job runner - pub background: Arc<BackgroundRunner>, /// The membership manager pub system: Arc<System>, /// The block manager @@ -78,7 +76,7 @@ pub struct GarageK2V { impl Garage { /// Create and run garage - pub fn new(config: Config, background: Arc<BackgroundRunner>) -> Result<Arc<Self>, Error> { + pub fn new(config: Config) -> Result<Arc<Self>, Error> { // Create meta dir and data dir if they don't exist already std::fs::create_dir_all(&config.metadata_dir) .ok_or_message("Unable to create Garage metadata directory")?; @@ -167,7 +165,7 @@ impl Garage { .expect("Invalid replication_mode in config file."); info!("Initialize membership management system..."); - let system = System::new(network_key, background.clone(), replication_mode, &config)?; + let system = System::new(network_key, replication_mode, &config)?; let data_rep_param = TableShardedReplication { system: system.clone(), @@ -225,7 +223,6 @@ impl Garage { info!("Initialize version_table..."); let version_table = Table::new( VersionTable { - background: background.clone(), block_ref_table: block_ref_table.clone(), }, meta_rep_param.clone(), @@ -240,7 +237,6 @@ impl Garage { #[allow(clippy::redundant_clone)] let object_table = Table::new( ObjectTable { - background: background.clone(), version_table: version_table.clone(), object_counter_table: object_counter_table.clone(), }, @@ -258,7 +254,6 @@ impl Garage { config, replication_mode, db, - background, system, block_manager, bucket_table, @@ -273,6 +268,22 @@ impl Garage { })) } + pub fn spawn_workers(&self, bg: &BackgroundRunner) { + self.block_manager.spawn_workers(bg); + + self.bucket_table.spawn_workers(bg); + self.bucket_alias_table.spawn_workers(bg); + self.key_table.spawn_workers(bg); + + self.object_table.spawn_workers(bg); + self.object_counter_table.spawn_workers(bg); + self.version_table.spawn_workers(bg); + self.block_ref_table.spawn_workers(bg); + + #[cfg(feature = "k2v")] + self.k2v.spawn_workers(bg); + } + pub fn bucket_helper(&self) -> helper::bucket::BucketHelper { helper::bucket::BucketHelper(self) } @@ -307,4 +318,9 @@ impl GarageK2V { rpc, } } + + pub fn spawn_workers(&self, bg: &BackgroundRunner) { + self.item_table.spawn_workers(bg); + self.counter_table.spawn_workers(bg); + } } diff --git a/src/model/index_counter.rs b/src/model/index_counter.rs index e6394f0c..35d6596d 100644 --- a/src/model/index_counter.rs +++ b/src/model/index_counter.rs @@ -1,19 +1,18 @@ use core::ops::Bound; -use std::collections::{hash_map, BTreeMap, HashMap}; +use std::collections::{BTreeMap, HashMap}; use std::marker::PhantomData; use std::sync::Arc; -use async_trait::async_trait; use serde::{Deserialize, Serialize}; -use tokio::sync::{mpsc, watch}; use garage_db as db; use garage_rpc::ring::Ring; use garage_rpc::system::System; -use garage_util::background::*; +use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_util::error::*; +use garage_util::migrate::Migrate; use garage_util::time::*; use garage_table::crdt::*; @@ -31,14 +30,44 @@ pub trait CountedItem: Clone + PartialEq + Send + Sync + 'static { fn counts(&self) -> Vec<(&'static str, i64)>; } -/// A counter entry in the global table -#[derive(Clone, PartialEq, Debug, Serialize, Deserialize)] -pub struct CounterEntry<T: CountedItem> { - pub pk: T::CP, - pub sk: T::CS, - pub values: BTreeMap<String, CounterValue>, +mod v08 { + use super::CountedItem; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + use std::collections::BTreeMap; + + // ---- Global part (the table everyone queries) ---- + + /// A counter entry in the global table + #[derive(Clone, PartialEq, Debug, Serialize, Deserialize)] + pub struct CounterEntry<T: CountedItem> { + pub pk: T::CP, + pub sk: T::CS, + pub values: BTreeMap<String, CounterValue>, + } + + /// A counter entry in the global table + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct CounterValue { + pub node_values: BTreeMap<Uuid, (u64, i64)>, + } + + impl<T: CountedItem> garage_util::migrate::InitialFormat for CounterEntry<T> {} + + // ---- Local part (the counter we maintain transactionnaly on each node) ---- + + #[derive(PartialEq, Clone, Debug, Serialize, Deserialize)] + pub(super) struct LocalCounterEntry<T: CountedItem> { + pub(super) pk: T::CP, + pub(super) sk: T::CS, + pub(super) values: BTreeMap<String, (u64, i64)>, + } + + impl<T: CountedItem> garage_util::migrate::InitialFormat for LocalCounterEntry<T> {} } +pub use v08::*; + impl<T: CountedItem> Entry<T::CP, T::CS> for CounterEntry<T> { fn partition_key(&self) -> &T::CP { &self.pk @@ -80,12 +109,6 @@ impl<T: CountedItem> CounterEntry<T> { } } -/// A counter entry in the global table -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct CounterValue { - pub node_values: BTreeMap<Uuid, (u64, i64)>, -} - impl<T: CountedItem> Crdt for CounterEntry<T> { fn merge(&mut self, other: &Self) { for (name, e2) in other.values.iter() { @@ -142,7 +165,6 @@ impl<T: CountedItem> TableSchema for CounterTable<T> { pub struct IndexCounter<T: CountedItem> { this_node: Uuid, local_counter: db::Tree, - propagate_tx: mpsc::UnboundedSender<(T::CP, T::CS, LocalCounterEntry<T>)>, pub table: Arc<Table<CounterTable<T>, TableShardedReplication>>, } @@ -152,16 +174,11 @@ impl<T: CountedItem> IndexCounter<T> { replication: TableShardedReplication, db: &db::Db, ) -> Arc<Self> { - let background = system.background.clone(); - - let (propagate_tx, propagate_rx) = mpsc::unbounded_channel(); - - let this = Arc::new(Self { + Arc::new(Self { this_node: system.id, local_counter: db .open_tree(format!("local_counter_v2:{}", T::COUNTER_TABLE_NAME)) .expect("Unable to open local counter tree"), - propagate_tx, table: Table::new( CounterTable { _phantom_t: Default::default(), @@ -170,16 +187,11 @@ impl<T: CountedItem> IndexCounter<T> { system, db, ), - }); - - background.spawn_worker(IndexPropagatorWorker { - index_counter: this.clone(), - propagate_rx, - buf: HashMap::new(), - errors: 0, - }); + }) + } - this + pub fn spawn_workers(&self, bg: &BackgroundRunner) { + self.table.spawn_workers(bg); } pub fn count( @@ -208,11 +220,9 @@ impl<T: CountedItem> IndexCounter<T> { let tree_key = self.table.data.tree_key(pk, sk); let mut entry = match tx.get(&self.local_counter, &tree_key[..])? { - Some(old_bytes) => { - rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>(&old_bytes) - .map_err(Error::RmpDecode) - .map_err(db::TxError::Abort)? - } + Some(old_bytes) => LocalCounterEntry::<T>::decode(&old_bytes) + .ok_or_message("Cannot decode local counter entry") + .map_err(db::TxError::Abort)?, None => LocalCounterEntry { pk: pk.clone(), sk: sk.clone(), @@ -227,17 +237,14 @@ impl<T: CountedItem> IndexCounter<T> { ent.1 += *inc; } - let new_entry_bytes = rmp_to_vec_all_named(&entry) + let new_entry_bytes = entry + .encode() .map_err(Error::RmpEncode) .map_err(db::TxError::Abort)?; tx.insert(&self.local_counter, &tree_key[..], new_entry_bytes)?; - if let Err(e) = self.propagate_tx.send((pk.clone(), sk.clone(), entry)) { - error!( - "Could not propagate updated counter values, failed to send to channel: {}", - e - ); - } + let dist_entry = entry.into_counter_entry(self.this_node); + self.table.queue_insert(tx, &dist_entry)?; Ok(()) } @@ -250,23 +257,6 @@ impl<T: CountedItem> IndexCounter<T> { TS: TableSchema<E = T>, TR: TableReplication, { - let save_counter_entry = |entry: CounterEntry<T>| -> Result<(), Error> { - let entry_k = self - .table - .data - .tree_key(entry.partition_key(), entry.sort_key()); - self.table - .data - .update_entry_with(&entry_k, |ent| match ent { - Some(mut ent) => { - ent.merge(&entry); - ent - } - None => entry.clone(), - })?; - Ok(()) - }; - // 1. Set all old local counters to zero let now = now_msec(); let mut next_start: Option<Vec<u8>> = None; @@ -289,20 +279,22 @@ impl<T: CountedItem> IndexCounter<T> { info!("zeroing old counters... ({})", hex::encode(&batch[0].0)); for (local_counter_k, local_counter) in batch { - let mut local_counter = - rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>(&local_counter)?; + let mut local_counter = LocalCounterEntry::<T>::decode(&local_counter) + .ok_or_message("Cannot decode local counter entry")?; for (_, tv) in local_counter.values.iter_mut() { tv.0 = std::cmp::max(tv.0 + 1, now); tv.1 = 0; } - let local_counter_bytes = rmp_to_vec_all_named(&local_counter)?; + let local_counter_bytes = local_counter.encode()?; self.local_counter .insert(&local_counter_k, &local_counter_bytes)?; let counter_entry = local_counter.into_counter_entry(self.this_node); - save_counter_entry(counter_entry)?; + self.local_counter + .db() + .transaction(|mut tx| self.table.queue_insert(&mut tx, &counter_entry))?; next_start = Some(local_counter_k); } @@ -343,9 +335,8 @@ impl<T: CountedItem> IndexCounter<T> { let local_counter_key = self.table.data.tree_key(pk, sk); let mut local_counter = match self.local_counter.get(&local_counter_key)? { Some(old_bytes) => { - let ent = rmp_serde::decode::from_read_ref::<_, LocalCounterEntry<T>>( - &old_bytes, - )?; + let ent = LocalCounterEntry::<T>::decode(&old_bytes) + .ok_or_message("Cannot decode local counter entry")?; assert!(ent.pk == *pk); assert!(ent.sk == *sk); ent @@ -362,12 +353,14 @@ impl<T: CountedItem> IndexCounter<T> { tv.1 += v; } - let local_counter_bytes = rmp_to_vec_all_named(&local_counter)?; + let local_counter_bytes = local_counter.encode()?; self.local_counter .insert(&local_counter_key, local_counter_bytes)?; let counter_entry = local_counter.into_counter_entry(self.this_node); - save_counter_entry(counter_entry)?; + self.local_counter + .db() + .transaction(|mut tx| self.table.queue_insert(&mut tx, &counter_entry))?; next_start = Some(counted_entry_k); } @@ -378,104 +371,7 @@ impl<T: CountedItem> IndexCounter<T> { } } -struct IndexPropagatorWorker<T: CountedItem> { - index_counter: Arc<IndexCounter<T>>, - propagate_rx: mpsc::UnboundedReceiver<(T::CP, T::CS, LocalCounterEntry<T>)>, - - buf: HashMap<Vec<u8>, CounterEntry<T>>, - errors: usize, -} - -impl<T: CountedItem> IndexPropagatorWorker<T> { - fn add_ent(&mut self, pk: T::CP, sk: T::CS, counters: LocalCounterEntry<T>) { - let tree_key = self.index_counter.table.data.tree_key(&pk, &sk); - let dist_entry = counters.into_counter_entry(self.index_counter.this_node); - match self.buf.entry(tree_key) { - hash_map::Entry::Vacant(e) => { - e.insert(dist_entry); - } - hash_map::Entry::Occupied(mut e) => { - e.get_mut().merge(&dist_entry); - } - } - } -} - -#[async_trait] -impl<T: CountedItem> Worker for IndexPropagatorWorker<T> { - fn name(&self) -> String { - format!("{} index counter propagator", T::COUNTER_TABLE_NAME) - } - - fn info(&self) -> Option<String> { - if !self.buf.is_empty() { - Some(format!("{} items in queue", self.buf.len())) - } else { - None - } - } - - async fn work(&mut self, must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { - // This loop batches updates to counters to be sent all at once. - // They are sent once the propagate_rx channel has been emptied (or is closed). - let closed = loop { - match self.propagate_rx.try_recv() { - Ok((pk, sk, counters)) => { - self.add_ent(pk, sk, counters); - } - Err(mpsc::error::TryRecvError::Empty) => break false, - Err(mpsc::error::TryRecvError::Disconnected) => break true, - } - }; - - if !self.buf.is_empty() { - let entries_k = self.buf.keys().take(100).cloned().collect::<Vec<_>>(); - let entries = entries_k.iter().map(|k| self.buf.get(k).unwrap()); - if let Err(e) = self.index_counter.table.insert_many(entries).await { - self.errors += 1; - if self.errors >= 2 && *must_exit.borrow() { - error!("({}) Could not propagate {} counter values: {}, these counters will not be updated correctly.", T::COUNTER_TABLE_NAME, self.buf.len(), e); - return Ok(WorkerState::Done); - } - // Propagate error up to worker manager, it will log it, increment a counter, - // and sleep for a certain delay (with exponential backoff), waiting for - // things to go back to normal - return Err(e); - } else { - for k in entries_k { - self.buf.remove(&k); - } - self.errors = 0; - } - - return Ok(WorkerState::Busy); - } else if closed { - return Ok(WorkerState::Done); - } else { - return Ok(WorkerState::Idle); - } - } - - async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState { - match self.propagate_rx.recv().await { - Some((pk, sk, counters)) => { - self.add_ent(pk, sk, counters); - WorkerState::Busy - } - None => match self.buf.is_empty() { - false => WorkerState::Busy, - true => WorkerState::Done, - }, - } - } -} - -#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)] -struct LocalCounterEntry<T: CountedItem> { - pk: T::CP, - sk: T::CS, - values: BTreeMap<String, (u64, i64)>, -} +// ---- impl<T: CountedItem> LocalCounterEntry<T> { fn into_counter_entry(self, this_node: Uuid) -> CounterEntry<T> { diff --git a/src/model/k2v/item_table.rs b/src/model/k2v/item_table.rs index 7860cb17..ce3e4129 100644 --- a/src/model/k2v/item_table.rs +++ b/src/model/k2v/item_table.rs @@ -1,7 +1,8 @@ -use serde::{Deserialize, Serialize}; use std::collections::BTreeMap; use std::sync::Arc; +use serde::{Deserialize, Serialize}; + use garage_db as db; use garage_util::data::*; @@ -17,32 +18,43 @@ pub const CONFLICTS: &str = "conflicts"; pub const VALUES: &str = "values"; pub const BYTES: &str = "bytes"; -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct K2VItem { - pub partition: K2VItemPartition, - pub sort_key: String, +mod v08 { + use crate::k2v::causality::K2VNodeId; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + use std::collections::BTreeMap; - items: BTreeMap<K2VNodeId, DvvsEntry>, -} + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct K2VItem { + pub partition: K2VItemPartition, + pub sort_key: String, -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize, Hash)] -pub struct K2VItemPartition { - pub bucket_id: Uuid, - pub partition_key: String, -} + pub(super) items: BTreeMap<K2VNodeId, DvvsEntry>, + } -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -struct DvvsEntry { - t_discard: u64, - values: Vec<(u64, DvvsValue)>, -} + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize, Hash)] + pub struct K2VItemPartition { + pub bucket_id: Uuid, + pub partition_key: String, + } + + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct DvvsEntry { + pub(super) t_discard: u64, + pub(super) values: Vec<(u64, DvvsValue)>, + } -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub enum DvvsValue { - Value(#[serde(with = "serde_bytes")] Vec<u8>), - Deleted, + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub enum DvvsValue { + Value(#[serde(with = "serde_bytes")] Vec<u8>), + Deleted, + } + + impl garage_util::migrate::InitialFormat for K2VItem {} } +pub use v08::*; + impl K2VItem { /// Creates a new K2VItem when no previous entry existed in the db pub fn new(bucket_id: Uuid, partition_key: String, sort_key: String) -> Self { diff --git a/src/model/k2v/rpc.rs b/src/model/k2v/rpc.rs index a74df277..f64a7984 100644 --- a/src/model/k2v/rpc.rs +++ b/src/model/k2v/rpc.rs @@ -273,14 +273,9 @@ impl K2VRpcHandler { } fn local_insert(&self, item: &InsertedItem) -> Result<Option<K2VItem>, Error> { - let tree_key = self - .item_table - .data - .tree_key(&item.partition, &item.sort_key); - self.item_table .data - .update_entry_with(&tree_key[..], |ent| { + .update_entry_with(&item.partition, &item.sort_key, |ent| { let mut ent = ent.unwrap_or_else(|| { K2VItem::new( item.partition.bucket_id, diff --git a/src/model/key_table.rs b/src/model/key_table.rs index 9d2fc783..bb5334a3 100644 --- a/src/model/key_table.rs +++ b/src/model/key_table.rs @@ -1,45 +1,121 @@ use serde::{Deserialize, Serialize}; -use garage_table::crdt::*; -use garage_table::*; +use garage_util::crdt::{self, Crdt}; use garage_util::data::*; +use garage_table::{DeletedFilter, EmptyKey, Entry, TableSchema}; + use crate::permission::BucketKeyPerm; -use crate::prev::v051::key_table as old; +pub(crate) mod v05 { + use garage_util::crdt; + use serde::{Deserialize, Serialize}; -/// An api key -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Key { - /// The id of the key (immutable), used as partition key - pub key_id: String, + /// An api key + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Key { + /// The id of the key (immutable), used as partition key + pub key_id: String, - /// Internal state of the key - pub state: crdt::Deletable<KeyParams>, -} + /// The secret_key associated + pub secret_key: String, -/// Configuration for a key -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct KeyParams { - /// The secret_key associated (immutable) - pub secret_key: String, + /// Name for the key + pub name: crdt::Lww<String>, - /// Name for the key - pub name: crdt::Lww<String>, + /// Is the key deleted + pub deleted: crdt::Bool, + + /// Buckets in which the key is authorized. Empty if `Key` is deleted + // CRDT interaction: deleted implies authorized_buckets is empty + pub authorized_buckets: crdt::LwwMap<String, PermissionSet>, + } - /// Flag to allow users having this key to create buckets - pub allow_create_bucket: crdt::Lww<bool>, + /// Permission given to a key in a bucket + #[derive(PartialOrd, Ord, PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct PermissionSet { + /// The key can be used to read the bucket + pub allow_read: bool, + /// The key can be used to write in the bucket + pub allow_write: bool, + } + + impl crdt::AutoCrdt for PermissionSet { + const WARN_IF_DIFFERENT: bool = true; + } - /// If the key is present: it gives some permissions, - /// a map of bucket IDs (uuids) to permissions. - /// Otherwise no permissions are granted to key - pub authorized_buckets: crdt::Map<Uuid, BucketKeyPerm>, + impl garage_util::migrate::InitialFormat for Key {} +} - /// A key can have a local view of buckets names it is - /// the only one to see, this is the namespace for these aliases - pub local_aliases: crdt::LwwMap<String, Option<Uuid>>, +mod v08 { + use super::v05; + use crate::permission::BucketKeyPerm; + use garage_util::crdt; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + + /// An api key + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Key { + /// The id of the key (immutable), used as partition key + pub key_id: String, + + /// Internal state of the key + pub state: crdt::Deletable<KeyParams>, + } + + /// Configuration for a key + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct KeyParams { + /// The secret_key associated (immutable) + pub secret_key: String, + + /// Name for the key + pub name: crdt::Lww<String>, + + /// Flag to allow users having this key to create buckets + pub allow_create_bucket: crdt::Lww<bool>, + + /// If the key is present: it gives some permissions, + /// a map of bucket IDs (uuids) to permissions. + /// Otherwise no permissions are granted to key + pub authorized_buckets: crdt::Map<Uuid, BucketKeyPerm>, + + /// A key can have a local view of buckets names it is + /// the only one to see, this is the namespace for these aliases + pub local_aliases: crdt::LwwMap<String, Option<Uuid>>, + } + + impl garage_util::migrate::Migrate for Key { + type Previous = v05::Key; + + fn migrate(old_k: v05::Key) -> Key { + let name = crdt::Lww::raw(old_k.name.timestamp(), old_k.name.get().clone()); + + let state = if old_k.deleted.get() { + crdt::Deletable::Deleted + } else { + // Authorized buckets is ignored here, + // migration is performed in specific migration code in + // garage/migrate.rs + crdt::Deletable::Present(KeyParams { + secret_key: old_k.secret_key, + name, + allow_create_bucket: crdt::Lww::new(false), + authorized_buckets: crdt::Map::new(), + local_aliases: crdt::LwwMap::new(), + }) + }; + Key { + key_id: old_k.key_id, + state, + } + } + } } +pub use v08::*; + impl KeyParams { fn new(secret_key: &str, name: &str) -> Self { KeyParams { @@ -173,28 +249,4 @@ impl TableSchema for KeyTable { } } } - - fn try_migrate(bytes: &[u8]) -> Option<Self::E> { - let old_k = rmp_serde::decode::from_read_ref::<_, old::Key>(bytes).ok()?; - let name = crdt::Lww::raw(old_k.name.timestamp(), old_k.name.get().clone()); - - let state = if old_k.deleted.get() { - crdt::Deletable::Deleted - } else { - // Authorized buckets is ignored here, - // migration is performed in specific migration code in - // garage/migrate.rs - crdt::Deletable::Present(KeyParams { - secret_key: old_k.secret_key, - name, - allow_create_bucket: crdt::Lww::new(false), - authorized_buckets: crdt::Map::new(), - local_aliases: crdt::LwwMap::new(), - }) - }; - Some(Key { - key_id: old_k.key_id, - state, - }) - } } diff --git a/src/model/migrate.rs b/src/model/migrate.rs index cd6ad26a..6b4c3eed 100644 --- a/src/model/migrate.rs +++ b/src/model/migrate.rs @@ -2,6 +2,7 @@ use std::sync::Arc; use garage_util::crdt::*; use garage_util::data::*; +use garage_util::encode::nonversioned_decode; use garage_util::error::Error as GarageError; use garage_util::time::*; @@ -28,8 +29,8 @@ impl Migrate { let mut old_buckets = vec![]; for res in tree.iter().map_err(GarageError::from)? { let (_k, v) = res.map_err(GarageError::from)?; - let bucket = rmp_serde::decode::from_read_ref::<_, old_bucket::Bucket>(&v[..]) - .map_err(GarageError::from)?; + let bucket = + nonversioned_decode::<old_bucket::Bucket>(&v[..]).map_err(GarageError::from)?; old_buckets.push(bucket); } diff --git a/src/model/prev/v051/bucket_table.rs b/src/model/prev/v051/bucket_table.rs index 628a49dd..19893458 100644 --- a/src/model/prev/v051/bucket_table.rs +++ b/src/model/prev/v051/bucket_table.rs @@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize}; use garage_table::crdt::Crdt; use garage_table::*; -use super::key_table::PermissionSet; +use crate::key_table::v05::PermissionSet; /// A bucket is a collection of objects /// diff --git a/src/model/prev/v051/key_table.rs b/src/model/prev/v051/key_table.rs deleted file mode 100644 index 37516b1c..00000000 --- a/src/model/prev/v051/key_table.rs +++ /dev/null @@ -1,50 +0,0 @@ -use serde::{Deserialize, Serialize}; - -use garage_table::crdt::*; -use garage_table::*; - -/// An api key -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Key { - /// The id of the key (immutable), used as partition key - pub key_id: String, - - /// The secret_key associated - pub secret_key: String, - - /// Name for the key - pub name: crdt::Lww<String>, - - /// Is the key deleted - pub deleted: crdt::Bool, - - /// Buckets in which the key is authorized. Empty if `Key` is deleted - // CRDT interaction: deleted implies authorized_buckets is empty - pub authorized_buckets: crdt::LwwMap<String, PermissionSet>, -} - -/// Permission given to a key in a bucket -#[derive(PartialOrd, Ord, PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct PermissionSet { - /// The key can be used to read the bucket - pub allow_read: bool, - /// The key can be used to write in the bucket - pub allow_write: bool, -} - -impl AutoCrdt for PermissionSet { - const WARN_IF_DIFFERENT: bool = true; -} - -impl Crdt for Key { - fn merge(&mut self, other: &Self) { - self.name.merge(&other.name); - self.deleted.merge(&other.deleted); - - if self.deleted.get() { - self.authorized_buckets.clear(); - } else { - self.authorized_buckets.merge(&other.authorized_buckets); - } - } -} diff --git a/src/model/prev/v051/mod.rs b/src/model/prev/v051/mod.rs index 7a954752..8c1335a5 100644 --- a/src/model/prev/v051/mod.rs +++ b/src/model/prev/v051/mod.rs @@ -1,4 +1 @@ pub(crate) mod bucket_table; -pub(crate) mod key_table; -pub(crate) mod object_table; -pub(crate) mod version_table; diff --git a/src/model/prev/v051/object_table.rs b/src/model/prev/v051/object_table.rs deleted file mode 100644 index e79e5787..00000000 --- a/src/model/prev/v051/object_table.rs +++ /dev/null @@ -1,149 +0,0 @@ -use serde::{Deserialize, Serialize}; -use std::collections::BTreeMap; - -use garage_util::data::*; - -use garage_table::crdt::*; - -/// An object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Object { - /// The bucket in which the object is stored, used as partition key - pub bucket: String, - - /// The key at which the object is stored in its bucket, used as sorting key - pub key: String, - - /// The list of currenty stored versions of the object - versions: Vec<ObjectVersion>, -} - -impl Object { - /// Get a list of currently stored versions of `Object` - pub fn versions(&self) -> &[ObjectVersion] { - &self.versions[..] - } -} - -/// Informations about a version of an object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersion { - /// Id of the version - pub uuid: Uuid, - /// Timestamp of when the object was created - pub timestamp: u64, - /// State of the version - pub state: ObjectVersionState, -} - -/// State of an object version -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub enum ObjectVersionState { - /// The version is being received - Uploading(ObjectVersionHeaders), - /// The version is fully received - Complete(ObjectVersionData), - /// The version uploaded containded errors or the upload was explicitly aborted - Aborted, -} - -impl Crdt for ObjectVersionState { - fn merge(&mut self, other: &Self) { - use ObjectVersionState::*; - match other { - Aborted => { - *self = Aborted; - } - Complete(b) => match self { - Aborted => {} - Complete(a) => { - a.merge(b); - } - Uploading(_) => { - *self = Complete(b.clone()); - } - }, - Uploading(_) => {} - } - } -} - -/// Data stored in object version -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub enum ObjectVersionData { - /// The object was deleted, this Version is a tombstone to mark it as such - DeleteMarker, - /// The object is short, it's stored inlined - Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>), - /// The object is not short, Hash of first block is stored here, next segments hashes are - /// stored in the version table - FirstBlock(ObjectVersionMeta, Hash), -} - -impl AutoCrdt for ObjectVersionData { - const WARN_IF_DIFFERENT: bool = true; -} - -/// Metadata about the object version -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersionMeta { - /// Headers to send to the client - pub headers: ObjectVersionHeaders, - /// Size of the object - pub size: u64, - /// etag of the object - pub etag: String, -} - -/// Additional headers for an object -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersionHeaders { - /// Content type of the object - pub content_type: String, - /// Any other http headers to send - pub other: BTreeMap<String, String>, -} - -impl ObjectVersion { - fn cmp_key(&self) -> (u64, Uuid) { - (self.timestamp, self.uuid) - } - - /// Is the object version completely received - pub fn is_complete(&self) -> bool { - matches!(self.state, ObjectVersionState::Complete(_)) - } -} - -impl Crdt for Object { - fn merge(&mut self, other: &Self) { - // Merge versions from other into here - for other_v in other.versions.iter() { - match self - .versions - .binary_search_by(|v| v.cmp_key().cmp(&other_v.cmp_key())) - { - Ok(i) => { - self.versions[i].state.merge(&other_v.state); - } - Err(i) => { - self.versions.insert(i, other_v.clone()); - } - } - } - - // Remove versions which are obsolete, i.e. those that come - // before the last version which .is_complete(). - let last_complete = self - .versions - .iter() - .enumerate() - .rev() - .find(|(_, v)| v.is_complete()) - .map(|(vi, _)| vi); - - if let Some(last_vi) = last_complete { - self.versions = self.versions.drain(last_vi..).collect::<Vec<_>>(); - } - } -} diff --git a/src/model/prev/v051/version_table.rs b/src/model/prev/v051/version_table.rs deleted file mode 100644 index c11c62d5..00000000 --- a/src/model/prev/v051/version_table.rs +++ /dev/null @@ -1,79 +0,0 @@ -use serde::{Deserialize, Serialize}; - -use garage_util::data::*; - -use garage_table::crdt::*; -use garage_table::*; - -/// A version of an object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Version { - /// UUID of the version, used as partition key - pub uuid: Uuid, - - // Actual data: the blocks for this version - // In the case of a multipart upload, also store the etags - // of individual parts and check them when doing CompleteMultipartUpload - /// Is this version deleted - pub deleted: crdt::Bool, - /// list of blocks of data composing the version - pub blocks: crdt::Map<VersionBlockKey, VersionBlock>, - /// Etag of each part in case of a multipart upload, empty otherwise - pub parts_etags: crdt::Map<u64, String>, - - // Back link to bucket+key so that we can figure if - // this was deleted later on - /// Bucket in which the related object is stored - pub bucket: String, - /// Key in which the related object is stored - pub key: String, -} - -#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)] -pub struct VersionBlockKey { - /// Number of the part - pub part_number: u64, - /// Offset of this sub-segment in its part - pub offset: u64, -} - -impl Ord for VersionBlockKey { - fn cmp(&self, other: &Self) -> std::cmp::Ordering { - self.part_number - .cmp(&other.part_number) - .then(self.offset.cmp(&other.offset)) - } -} - -impl PartialOrd for VersionBlockKey { - fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> { - Some(self.cmp(other)) - } -} - -/// Informations about a single block -#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)] -pub struct VersionBlock { - /// Blake2 sum of the block - pub hash: Hash, - /// Size of the block - pub size: u64, -} - -impl AutoCrdt for VersionBlock { - const WARN_IF_DIFFERENT: bool = true; -} - -impl Crdt for Version { - fn merge(&mut self, other: &Self) { - self.deleted.merge(&other.deleted); - - if self.deleted.get() { - self.blocks.clear(); - self.parts_etags.clear(); - } else { - self.blocks.merge(&other.blocks); - self.parts_etags.merge(&other.parts_etags); - } - } -} diff --git a/src/model/s3/block_ref_table.rs b/src/model/s3/block_ref_table.rs index c7017409..7b023d87 100644 --- a/src/model/s3/block_ref_table.rs +++ b/src/model/s3/block_ref_table.rs @@ -1,4 +1,3 @@ -use serde::{Deserialize, Serialize}; use std::sync::Arc; use garage_db as db; @@ -10,19 +9,29 @@ use garage_table::*; use garage_block::manager::*; -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct BlockRef { - /// Hash (blake2 sum) of the block, used as partition key - pub block: Hash, +mod v08 { + use garage_util::crdt; + use garage_util::data::{Hash, Uuid}; + use serde::{Deserialize, Serialize}; - /// Id of the Version for the object containing this block, used as sorting key - pub version: Uuid, + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct BlockRef { + /// Hash (blake2 sum) of the block, used as partition key + pub block: Hash, - // Keep track of deleted status - /// Is the Version that contains this block deleted - pub deleted: crdt::Bool, + /// Id of the Version for the object containing this block, used as sorting key + pub version: Uuid, + + // Keep track of deleted status + /// Is the Version that contains this block deleted + pub deleted: crdt::Bool, + } + + impl garage_util::migrate::InitialFormat for BlockRef {} } +pub use v08::*; + impl Entry<Hash, Uuid> for BlockRef { fn partition_key(&self) -> &Hash { &self.block diff --git a/src/model/s3/object_table.rs b/src/model/s3/object_table.rs index 26ff57f6..518acc95 100644 --- a/src/model/s3/object_table.rs +++ b/src/model/s3/object_table.rs @@ -1,10 +1,8 @@ use serde::{Deserialize, Serialize}; -use std::collections::BTreeMap; use std::sync::Arc; use garage_db as db; -use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_table::crdt::*; @@ -14,25 +12,126 @@ use garage_table::*; use crate::index_counter::*; use crate::s3::version_table::*; -use crate::prev::v051::object_table as old; - pub const OBJECTS: &str = "objects"; pub const UNFINISHED_UPLOADS: &str = "unfinished_uploads"; pub const BYTES: &str = "bytes"; -/// An object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Object { - /// The bucket in which the object is stored, used as partition key - pub bucket_id: Uuid, +mod v05 { + use garage_util::data::{Hash, Uuid}; + use serde::{Deserialize, Serialize}; + use std::collections::BTreeMap; + + /// An object + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Object { + /// The bucket in which the object is stored, used as partition key + pub bucket: String, + + /// The key at which the object is stored in its bucket, used as sorting key + pub key: String, + + /// The list of currenty stored versions of the object + pub(super) versions: Vec<ObjectVersion>, + } + + /// Informations about a version of an object + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct ObjectVersion { + /// Id of the version + pub uuid: Uuid, + /// Timestamp of when the object was created + pub timestamp: u64, + /// State of the version + pub state: ObjectVersionState, + } + + /// State of an object version + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub enum ObjectVersionState { + /// The version is being received + Uploading(ObjectVersionHeaders), + /// The version is fully received + Complete(ObjectVersionData), + /// The version uploaded containded errors or the upload was explicitly aborted + Aborted, + } + + /// Data stored in object version + #[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] + pub enum ObjectVersionData { + /// The object was deleted, this Version is a tombstone to mark it as such + DeleteMarker, + /// The object is short, it's stored inlined + Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>), + /// The object is not short, Hash of first block is stored here, next segments hashes are + /// stored in the version table + FirstBlock(ObjectVersionMeta, Hash), + } + + /// Metadata about the object version + #[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] + pub struct ObjectVersionMeta { + /// Headers to send to the client + pub headers: ObjectVersionHeaders, + /// Size of the object + pub size: u64, + /// etag of the object + pub etag: String, + } - /// The key at which the object is stored in its bucket, used as sorting key - pub key: String, + /// Additional headers for an object + #[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] + pub struct ObjectVersionHeaders { + /// Content type of the object + pub content_type: String, + /// Any other http headers to send + pub other: BTreeMap<String, String>, + } - /// The list of currenty stored versions of the object - versions: Vec<ObjectVersion>, + impl garage_util::migrate::InitialFormat for Object {} } +mod v08 { + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + + use super::v05; + + pub use v05::{ + ObjectVersion, ObjectVersionData, ObjectVersionHeaders, ObjectVersionMeta, + ObjectVersionState, + }; + + /// An object + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Object { + /// The bucket in which the object is stored, used as partition key + pub bucket_id: Uuid, + + /// The key at which the object is stored in its bucket, used as sorting key + pub key: String, + + /// The list of currenty stored versions of the object + pub(super) versions: Vec<ObjectVersion>, + } + + impl garage_util::migrate::Migrate for Object { + type Previous = v05::Object; + + fn migrate(old: v05::Object) -> Object { + use garage_util::data::blake2sum; + + Object { + bucket_id: blake2sum(old.bucket.as_bytes()), + key: old.key, + versions: old.versions, + } + } + } +} + +pub use v08::*; + impl Object { /// Initialize an Object struct from parts pub fn new(bucket_id: Uuid, key: String, versions: Vec<ObjectVersion>) -> Self { @@ -69,28 +168,6 @@ impl Object { } } -/// Informations about a version of an object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersion { - /// Id of the version - pub uuid: Uuid, - /// Timestamp of when the object was created - pub timestamp: u64, - /// State of the version - pub state: ObjectVersionState, -} - -/// State of an object version -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub enum ObjectVersionState { - /// The version is being received - Uploading(ObjectVersionHeaders), - /// The version is fully received - Complete(ObjectVersionData), - /// The version uploaded containded errors or the upload was explicitly aborted - Aborted, -} - impl Crdt for ObjectVersionState { fn merge(&mut self, other: &Self) { use ObjectVersionState::*; @@ -112,42 +189,10 @@ impl Crdt for ObjectVersionState { } } -/// Data stored in object version -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub enum ObjectVersionData { - /// The object was deleted, this Version is a tombstone to mark it as such - DeleteMarker, - /// The object is short, it's stored inlined - Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>), - /// The object is not short, Hash of first block is stored here, next segments hashes are - /// stored in the version table - FirstBlock(ObjectVersionMeta, Hash), -} - impl AutoCrdt for ObjectVersionData { const WARN_IF_DIFFERENT: bool = true; } -/// Metadata about the object version -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersionMeta { - /// Headers to send to the client - pub headers: ObjectVersionHeaders, - /// Size of the object - pub size: u64, - /// etag of the object - pub etag: String, -} - -/// Additional headers for an object -#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)] -pub struct ObjectVersionHeaders { - /// Content type of the object - pub content_type: String, - /// Any other http headers to send - pub other: BTreeMap<String, String>, -} - impl ObjectVersion { fn cmp_key(&self) -> (u64, Uuid) { (self.timestamp, self.uuid) @@ -221,7 +266,6 @@ impl Crdt for Object { } pub struct ObjectTable { - pub background: Arc<BackgroundRunner>, pub version_table: Arc<Table<VersionTable, TableShardedReplication>>, pub object_counter_table: Arc<IndexCounter<Object>>, } @@ -255,34 +299,34 @@ impl TableSchema for ObjectTable { ); } - // 2. Spawn threads that propagates deletions to version table - let version_table = self.version_table.clone(); - let old = old.cloned(); - let new = new.cloned(); - - self.background.spawn(async move { - if let (Some(old_v), Some(new_v)) = (old, new) { - // Propagate deletion of old versions - for v in old_v.versions.iter() { - let newly_deleted = match new_v - .versions - .binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key())) - { - Err(_) => true, - Ok(i) => { - new_v.versions[i].state == ObjectVersionState::Aborted - && v.state != ObjectVersionState::Aborted - } - }; - if newly_deleted { - let deleted_version = - Version::new(v.uuid, old_v.bucket_id, old_v.key.clone(), true); - version_table.insert(&deleted_version).await?; + // 2. Enqueue propagation deletions to version table + if let (Some(old_v), Some(new_v)) = (old, new) { + // Propagate deletion of old versions + for v in old_v.versions.iter() { + let newly_deleted = match new_v + .versions + .binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key())) + { + Err(_) => true, + Ok(i) => { + new_v.versions[i].state == ObjectVersionState::Aborted + && v.state != ObjectVersionState::Aborted + } + }; + if newly_deleted { + let deleted_version = + Version::new(v.uuid, old_v.bucket_id, old_v.key.clone(), true); + let res = self.version_table.queue_insert(tx, &deleted_version); + if let Err(e) = db::unabort(res)? { + error!( + "Unable to enqueue version deletion propagation: {}. A repair will be needed.", + e + ); } } } - Ok(()) - }); + } + Ok(()) } @@ -292,11 +336,6 @@ impl TableSchema for ObjectTable { ObjectFilter::IsUploading => entry.versions.iter().any(|v| v.is_uploading()), } } - - fn try_migrate(bytes: &[u8]) -> Option<Self::E> { - let old_obj = rmp_serde::decode::from_read_ref::<_, old::Object>(bytes).ok()?; - Some(migrate_object(old_obj)) - } } impl CountedItem for Object { @@ -341,64 +380,3 @@ impl CountedItem for Object { ] } } - -// vvvvvvvv migration code, stupid stuff vvvvvvvvvvvv -// (we just want to change bucket into bucket_id by hashing it) - -fn migrate_object(o: old::Object) -> Object { - let versions = o - .versions() - .iter() - .cloned() - .map(migrate_object_version) - .collect(); - Object { - bucket_id: blake2sum(o.bucket.as_bytes()), - key: o.key, - versions, - } -} - -fn migrate_object_version(v: old::ObjectVersion) -> ObjectVersion { - ObjectVersion { - uuid: Uuid::try_from(v.uuid.as_slice()).unwrap(), - timestamp: v.timestamp, - state: match v.state { - old::ObjectVersionState::Uploading(h) => { - ObjectVersionState::Uploading(migrate_object_version_headers(h)) - } - old::ObjectVersionState::Complete(d) => { - ObjectVersionState::Complete(migrate_object_version_data(d)) - } - old::ObjectVersionState::Aborted => ObjectVersionState::Aborted, - }, - } -} - -fn migrate_object_version_headers(h: old::ObjectVersionHeaders) -> ObjectVersionHeaders { - ObjectVersionHeaders { - content_type: h.content_type, - other: h.other, - } -} - -fn migrate_object_version_data(d: old::ObjectVersionData) -> ObjectVersionData { - match d { - old::ObjectVersionData::DeleteMarker => ObjectVersionData::DeleteMarker, - old::ObjectVersionData::Inline(m, b) => { - ObjectVersionData::Inline(migrate_object_version_meta(m), b) - } - old::ObjectVersionData::FirstBlock(m, h) => ObjectVersionData::FirstBlock( - migrate_object_version_meta(m), - Hash::try_from(h.as_slice()).unwrap(), - ), - } -} - -fn migrate_object_version_meta(m: old::ObjectVersionMeta) -> ObjectVersionMeta { - ObjectVersionMeta { - headers: migrate_object_version_headers(m.headers), - size: m.size, - etag: m.etag, - } -} diff --git a/src/model/s3/version_table.rs b/src/model/s3/version_table.rs index 6bc2ecd1..6edc83f4 100644 --- a/src/model/s3/version_table.rs +++ b/src/model/s3/version_table.rs @@ -1,9 +1,7 @@ -use serde::{Deserialize, Serialize}; use std::sync::Arc; use garage_db as db; -use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_table::crdt::*; @@ -12,32 +10,108 @@ use garage_table::*; use crate::s3::block_ref_table::*; -use crate::prev::v051::version_table as old; - -/// A version of an object -#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] -pub struct Version { - /// UUID of the version, used as partition key - pub uuid: Uuid, - - // Actual data: the blocks for this version - // In the case of a multipart upload, also store the etags - // of individual parts and check them when doing CompleteMultipartUpload - /// Is this version deleted - pub deleted: crdt::Bool, - /// list of blocks of data composing the version - pub blocks: crdt::Map<VersionBlockKey, VersionBlock>, - /// Etag of each part in case of a multipart upload, empty otherwise - pub parts_etags: crdt::Map<u64, String>, - - // Back link to bucket+key so that we can figure if - // this was deleted later on - /// Bucket in which the related object is stored - pub bucket_id: Uuid, - /// Key in which the related object is stored - pub key: String, +mod v05 { + use garage_util::crdt; + use garage_util::data::{Hash, Uuid}; + use serde::{Deserialize, Serialize}; + + /// A version of an object + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Version { + /// UUID of the version, used as partition key + pub uuid: Uuid, + + // Actual data: the blocks for this version + // In the case of a multipart upload, also store the etags + // of individual parts and check them when doing CompleteMultipartUpload + /// Is this version deleted + pub deleted: crdt::Bool, + /// list of blocks of data composing the version + pub blocks: crdt::Map<VersionBlockKey, VersionBlock>, + /// Etag of each part in case of a multipart upload, empty otherwise + pub parts_etags: crdt::Map<u64, String>, + + // Back link to bucket+key so that we can figure if + // this was deleted later on + /// Bucket in which the related object is stored + pub bucket: String, + /// Key in which the related object is stored + pub key: String, + } + + #[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)] + pub struct VersionBlockKey { + /// Number of the part + pub part_number: u64, + /// Offset of this sub-segment in its part + pub offset: u64, + } + + /// Informations about a single block + #[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)] + pub struct VersionBlock { + /// Blake2 sum of the block + pub hash: Hash, + /// Size of the block + pub size: u64, + } + + impl garage_util::migrate::InitialFormat for Version {} } +mod v08 { + use garage_util::crdt; + use garage_util::data::Uuid; + use serde::{Deserialize, Serialize}; + + use super::v05; + + /// A version of an object + #[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)] + pub struct Version { + /// UUID of the version, used as partition key + pub uuid: Uuid, + + // Actual data: the blocks for this version + // In the case of a multipart upload, also store the etags + // of individual parts and check them when doing CompleteMultipartUpload + /// Is this version deleted + pub deleted: crdt::Bool, + /// list of blocks of data composing the version + pub blocks: crdt::Map<VersionBlockKey, VersionBlock>, + /// Etag of each part in case of a multipart upload, empty otherwise + pub parts_etags: crdt::Map<u64, String>, + + // Back link to bucket+key so that we can figure if + // this was deleted later on + /// Bucket in which the related object is stored + pub bucket_id: Uuid, + /// Key in which the related object is stored + pub key: String, + } + + pub use v05::{VersionBlock, VersionBlockKey}; + + impl garage_util::migrate::Migrate for Version { + type Previous = v05::Version; + + fn migrate(old: v05::Version) -> Version { + use garage_util::data::blake2sum; + + Version { + uuid: old.uuid, + deleted: old.deleted, + blocks: old.blocks, + parts_etags: old.parts_etags, + bucket_id: blake2sum(old.bucket.as_bytes()), + key: old.key, + } + } + } +} + +pub use v08::*; + impl Version { pub fn new(uuid: Uuid, bucket_id: Uuid, key: String, deleted: bool) -> Self { Self { @@ -65,14 +139,6 @@ impl Version { } } -#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)] -pub struct VersionBlockKey { - /// Number of the part - pub part_number: u64, - /// Offset of this sub-segment in its part - pub offset: u64, -} - impl Ord for VersionBlockKey { fn cmp(&self, other: &Self) -> std::cmp::Ordering { self.part_number @@ -87,15 +153,6 @@ impl PartialOrd for VersionBlockKey { } } -/// Informations about a single block -#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)] -pub struct VersionBlock { - /// Blake2 sum of the block - pub hash: Hash, - /// Size of the block - pub size: u64, -} - impl AutoCrdt for VersionBlock { const WARN_IF_DIFFERENT: bool = true; } @@ -127,7 +184,6 @@ impl Crdt for Version { } pub struct VersionTable { - pub background: Arc<BackgroundRunner>, pub block_ref_table: Arc<Table<BlockRefTable, TableShardedReplication>>, } @@ -141,33 +197,26 @@ impl TableSchema for VersionTable { fn updated( &self, - _tx: &mut db::Transaction, + tx: &mut db::Transaction, old: Option<&Self::E>, new: Option<&Self::E>, ) -> db::TxOpResult<()> { - let block_ref_table = self.block_ref_table.clone(); - let old = old.cloned(); - let new = new.cloned(); - - self.background.spawn(async move { - if let (Some(old_v), Some(new_v)) = (old, new) { - // Propagate deletion of version blocks - if new_v.deleted.get() && !old_v.deleted.get() { - let deleted_block_refs = old_v - .blocks - .items() - .iter() - .map(|(_k, vb)| BlockRef { - block: vb.hash, - version: old_v.uuid, - deleted: true.into(), - }) - .collect::<Vec<_>>(); - block_ref_table.insert_many(&deleted_block_refs[..]).await?; + if let (Some(old_v), Some(new_v)) = (old, new) { + // Propagate deletion of version blocks + if new_v.deleted.get() && !old_v.deleted.get() { + let deleted_block_refs = old_v.blocks.items().iter().map(|(_k, vb)| BlockRef { + block: vb.hash, + version: old_v.uuid, + deleted: true.into(), + }); + for block_ref in deleted_block_refs { + let res = self.block_ref_table.queue_insert(tx, &block_ref); + if let Err(e) = db::unabort(res)? { + error!("Unable to enqueue block ref deletion propagation: {}. A repair will be needed.", e); + } } } - Ok(()) - }); + } Ok(()) } @@ -175,42 +224,4 @@ impl TableSchema for VersionTable { fn matches_filter(entry: &Self::E, filter: &Self::Filter) -> bool { filter.apply(entry.deleted.get()) } - - fn try_migrate(bytes: &[u8]) -> Option<Self::E> { - let old = rmp_serde::decode::from_read_ref::<_, old::Version>(bytes).ok()?; - - let blocks = old - .blocks - .items() - .iter() - .map(|(k, v)| { - ( - VersionBlockKey { - part_number: k.part_number, - offset: k.offset, - }, - VersionBlock { - hash: Hash::try_from(v.hash.as_slice()).unwrap(), - size: v.size, - }, - ) - }) - .collect::<crdt::Map<_, _>>(); - - let parts_etags = old - .parts_etags - .items() - .iter() - .map(|(k, v)| (*k, v.clone())) - .collect::<crdt::Map<_, _>>(); - - Some(Version { - uuid: Hash::try_from(old.uuid.as_slice()).unwrap(), - deleted: crdt::Bool::new(old.deleted.get()), - blocks, - parts_etags, - bucket_id: blake2sum(old.bucket.as_bytes()), - key: old.key, - }) - } } diff --git a/src/rpc/Cargo.toml b/src/rpc/Cargo.toml index 1b411c6a..3d4d3ff5 100644 --- a/src/rpc/Cargo.toml +++ b/src/rpc/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_rpc" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,7 +14,7 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_util = { version = "0.8.0", path = "../util" } +garage_util = { version = "0.8.1", path = "../util" } arc-swap = "1.0" bytes = "1.0" @@ -27,7 +27,6 @@ itertools="0.10" sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" } async-trait = "0.1.7" -rmp-serde = "0.15" serde = { version = "1.0", default-features = false, features = ["derive", "rc"] } serde_bytes = "0.11" serde_json = "1.0" diff --git a/src/rpc/layout.rs b/src/rpc/layout.rs index 1cef44d1..d756f0aa 100644 --- a/src/rpc/layout.rs +++ b/src/rpc/layout.rs @@ -9,6 +9,7 @@ use serde::{Deserialize, Serialize}; use garage_util::crdt::{AutoCrdt, Crdt, Lww, LwwMap}; use garage_util::data::*; +use garage_util::encode::nonversioned_encode; use garage_util::error::*; use crate::graph_algo::*; @@ -58,6 +59,7 @@ pub struct ClusterLayout { pub staging_roles: LwwMap<Uuid, NodeRoleV>, pub staging_hash: Hash, } +impl garage_util::migrate::InitialFormat for ClusterLayout {} /// This struct is used to set the parameters to be used in the assignation computation /// algorithm. It is stored as a Crdt. @@ -134,7 +136,7 @@ impl ClusterLayout { fn calculate_staging_hash(&self) -> Hash { let hashed_tuple = (&self.staging_roles, &self.staging_parameters); - blake2sum(&rmp_to_vec_all_named(&hashed_tuple).unwrap()[..]) + blake2sum(&nonversioned_encode(&hashed_tuple).unwrap()[..]) } pub fn merge(&mut self, other: &ClusterLayout) -> bool { diff --git a/src/rpc/rpc_helper.rs b/src/rpc/rpc_helper.rs index 949aced6..1ec250c3 100644 --- a/src/rpc/rpc_helper.rs +++ b/src/rpc/rpc_helper.rs @@ -5,7 +5,6 @@ use std::time::Duration; use futures::future::join_all; use futures::stream::futures_unordered::FuturesUnordered; use futures::stream::StreamExt; -use futures_util::future::FutureExt; use tokio::select; use tokio::sync::watch; @@ -24,7 +23,6 @@ pub use netapp::message::{ use netapp::peering::fullmesh::FullMeshPeeringStrategy; pub use netapp::{self, NetApp, NodeID}; -use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_util::error::Error; use garage_util::metrics::RecordDuration; @@ -94,7 +92,6 @@ pub struct RpcHelper(Arc<RpcHelperInner>); struct RpcHelperInner { our_node_id: Uuid, fullmesh: Arc<FullMeshPeeringStrategy>, - background: Arc<BackgroundRunner>, ring: watch::Receiver<Arc<Ring>>, metrics: RpcMetrics, rpc_timeout: Duration, @@ -104,7 +101,6 @@ impl RpcHelper { pub(crate) fn new( our_node_id: Uuid, fullmesh: Arc<FullMeshPeeringStrategy>, - background: Arc<BackgroundRunner>, ring: watch::Receiver<Arc<Ring>>, rpc_timeout: Option<Duration>, ) -> Self { @@ -113,7 +109,6 @@ impl RpcHelper { Self(Arc::new(RpcHelperInner { our_node_id, fullmesh, - background, ring, metrics, rpc_timeout: rpc_timeout.unwrap_or(DEFAULT_TIMEOUT), @@ -377,16 +372,13 @@ impl RpcHelper { if !resp_stream.is_empty() { // Continue remaining requests in background. - // Continue the remaining requests immediately using tokio::spawn - // but enqueue a task in the background runner - // to ensure that the process won't exit until the requests are done - // (if we had just enqueued the resp_stream.collect directly in the background runner, - // the requests might have been put on hold in the background runner's queue, - // in which case they might timeout or otherwise fail) - let wait_finished_fut = tokio::spawn(async move { + // Note: these requests can get interrupted on process shutdown, + // we must not count on them being executed for certain. + // For all background things that have to happen with certainty, + // they have to be put in a proper queue that is persisted to disk. + tokio::spawn(async move { resp_stream.collect::<Vec<Result<_, _>>>().await; }); - self.0.background.spawn(wait_finished_fut.map(|_| Ok(()))); } } diff --git a/src/rpc/system.rs b/src/rpc/system.rs index ea92b570..1f4d86e7 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -21,7 +21,6 @@ use netapp::peering::fullmesh::FullMeshPeeringStrategy; use netapp::util::parse_and_resolve_peer_addr_async; use netapp::{NetApp, NetworkKey, NodeID, NodeKey}; -use garage_util::background::BackgroundRunner; use garage_util::config::Config; #[cfg(feature = "kubernetes-discovery")] use garage_util::config::KubernetesDiscoveryConfig; @@ -50,8 +49,6 @@ pub const GARAGE_VERSION_TAG: u64 = 0x6761726167650008; // garage 0x0008 /// RPC endpoint used for calls related to membership pub const SYSTEM_RPC_PATH: &str = "garage_rpc/membership.rs/SystemRpc"; -pub const CONNECT_ERROR_MESSAGE: &str = "Error establishing RPC connection to remote node. This can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret"; - /// RPC messages related to membership #[derive(Debug, Serialize, Deserialize, Clone)] pub enum SystemRpc { @@ -76,13 +73,17 @@ impl Rpc for SystemRpc { type Response = Result<SystemRpc, Error>; } +#[derive(Serialize, Deserialize)] +pub struct PeerList(Vec<(Uuid, SocketAddr)>); +impl garage_util::migrate::InitialFormat for PeerList {} + /// This node's membership manager pub struct System { /// The id of this node pub id: Uuid, persist_cluster_layout: Persister<ClusterLayout>, - persist_peer_list: Persister<Vec<(Uuid, SocketAddr)>>, + persist_peer_list: Persister<PeerList>, local_status: ArcSwap<NodeStatus>, node_status: RwLock<HashMap<Uuid, (u64, NodeStatus)>>, @@ -110,9 +111,6 @@ pub struct System { pub ring: watch::Receiver<Arc<Ring>>, update_ring: Mutex<watch::Sender<Arc<Ring>>>, - /// The job runner of this node - pub background: Arc<BackgroundRunner>, - /// Path to metadata directory pub metadata_dir: PathBuf, } @@ -232,7 +230,6 @@ impl System { /// Create this node's membership manager pub fn new( network_key: NetworkKey, - background: Arc<BackgroundRunner>, replication_mode: ReplicationMode, config: &Config, ) -> Result<Arc<Self>, Error> { @@ -354,7 +351,6 @@ impl System { rpc: RpcHelper::new( netapp.id.into(), fullmesh, - background.clone(), ring.clone(), config.rpc_timeout_msec.map(Duration::from_millis), ), @@ -372,7 +368,6 @@ impl System { ring, update_ring: Mutex::new(update_ring), - background, metadata_dir: config.metadata_dir.clone(), }); sys.system_endpoint.set_handler(sys.clone()); @@ -444,17 +439,14 @@ impl System { )) })?; let mut errors = vec![]; - for ip in addrs.iter() { - match self - .netapp - .clone() - .try_connect(*ip, pubkey) - .await - .err_context(CONNECT_ERROR_MESSAGE) - { + for addr in addrs.iter() { + match self.netapp.clone().try_connect(*addr, pubkey).await { Ok(()) => return Ok(()), Err(e) => { - errors.push((*ip, e)); + errors.push(( + *addr, + Error::Message(connect_error_message(*addr, pubkey, e)), + )); } } } @@ -529,56 +521,61 @@ impl System { // ---- INTERNALS ---- #[cfg(feature = "consul-discovery")] - async fn advertise_to_consul(self: Arc<Self>) -> Result<(), Error> { + async fn advertise_to_consul(self: Arc<Self>) { let c = match &self.consul_discovery { Some(c) => c, - _ => return Ok(()), + _ => return, }; let rpc_public_addr = match self.rpc_public_addr { Some(addr) => addr, None => { warn!("Not advertising to Consul because rpc_public_addr is not defined in config file and could not be autodetected."); - return Ok(()); + return; } }; - c.publish_consul_service( - self.netapp.id, - &self.local_status.load_full().hostname, - rpc_public_addr, - ) - .await - .err_context("Error while publishing Consul service") + if let Err(e) = c + .publish_consul_service( + self.netapp.id, + &self.local_status.load_full().hostname, + rpc_public_addr, + ) + .await + { + error!("Error while publishing Consul service: {}", e); + } } #[cfg(feature = "kubernetes-discovery")] - async fn advertise_to_kubernetes(self: Arc<Self>) -> Result<(), Error> { + async fn advertise_to_kubernetes(self: Arc<Self>) { let k = match &self.kubernetes_discovery { Some(k) => k, - _ => return Ok(()), + _ => return, }; let rpc_public_addr = match self.rpc_public_addr { Some(addr) => addr, None => { warn!("Not advertising to Kubernetes because rpc_public_addr is not defined in config file and could not be autodetected."); - return Ok(()); + return; } }; - publish_kubernetes_node( + if let Err(e) = publish_kubernetes_node( k, self.netapp.id, &self.local_status.load_full().hostname, rpc_public_addr, ) .await - .err_context("Error while publishing node to kubernetes") + { + error!("Error while publishing node to Kubernetes: {}", e); + } } /// Save network configuration to disc - async fn save_cluster_layout(self: Arc<Self>) -> Result<(), Error> { + async fn save_cluster_layout(&self) -> Result<(), Error> { let ring: Arc<Ring> = self.ring.borrow().clone(); self.persist_cluster_layout .save_async(&ring.layout) @@ -630,11 +627,7 @@ impl System { if info.cluster_layout_version > local_info.cluster_layout_version || info.cluster_layout_staging_hash != local_info.cluster_layout_staging_hash { - let self2 = self.clone(); - self.background.spawn_cancellable(async move { - self2.pull_cluster_layout(from).await; - Ok(()) - }); + tokio::spawn(self.clone().pull_cluster_layout(from)); } self.node_status @@ -676,18 +669,21 @@ impl System { drop(update_ring); let self2 = self.clone(); - self.background.spawn_cancellable(async move { - self2 + tokio::spawn(async move { + if let Err(e) = self2 .rpc .broadcast( &self2.system_endpoint, SystemRpc::AdvertiseClusterLayout(layout), RequestStrategy::with_priority(PRIO_HIGH), ) - .await?; - Ok(()) + .await + { + warn!("Error while broadcasting new cluster layout: {}", e); + } }); - self.background.spawn(self.clone().save_cluster_layout()); + + self.save_cluster_layout().await?; } Ok(SystemRpc::Ok) @@ -734,7 +730,7 @@ impl System { // Add peer list from list stored on disk if let Ok(peers) = self.persist_peer_list.load_async().await { - ping_list.extend(peers.iter().map(|(id, addr)| ((*id).into(), *addr))) + ping_list.extend(peers.0.iter().map(|(id, addr)| ((*id).into(), *addr))) } // Fetch peer list from Consul @@ -773,12 +769,12 @@ impl System { } for (node_id, node_addr) in ping_list { - tokio::spawn( - self.netapp - .clone() - .try_connect(node_addr, node_id) - .map(|r| r.err_context(CONNECT_ERROR_MESSAGE)), - ); + let self2 = self.clone(); + tokio::spawn(async move { + if let Err(e) = self2.netapp.clone().try_connect(node_addr, node_id).await { + error!("{}", connect_error_message(node_addr, node_id, e)); + } + }); } } @@ -787,11 +783,10 @@ impl System { } #[cfg(feature = "consul-discovery")] - self.background.spawn(self.clone().advertise_to_consul()); + tokio::spawn(self.clone().advertise_to_consul()); #[cfg(feature = "kubernetes-discovery")] - self.background - .spawn(self.clone().advertise_to_kubernetes()); + tokio::spawn(self.clone().advertise_to_kubernetes()); let restart_at = tokio::time::sleep(DISCOVERY_INTERVAL); select! { @@ -815,12 +810,16 @@ impl System { // and append it to the list we are about to save, // so that no peer ID gets lost in the process. if let Ok(mut prev_peer_list) = self.persist_peer_list.load_async().await { - prev_peer_list.retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id)); - peer_list.extend(prev_peer_list); + prev_peer_list + .0 + .retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id)); + peer_list.extend(prev_peer_list.0); } // Save new peer list to file - self.persist_peer_list.save_async(&peer_list).await + self.persist_peer_list + .save_async(&PeerList(peer_list)) + .await } async fn pull_cluster_layout(self: Arc<Self>, peer: Uuid) { @@ -881,3 +880,11 @@ async fn resolve_peers(peers: &[String]) -> Vec<(NodeID, SocketAddr)> { ret } + +fn connect_error_message( + addr: SocketAddr, + pubkey: ed25519::PublicKey, + e: netapp::error::Error, +) -> String { + format!("Error establishing RPC connection to remote node: {}@{}.\nThis can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret.\n{}", hex::encode(pubkey), addr, e) +} diff --git a/src/table/Cargo.toml b/src/table/Cargo.toml index 38c6b41c..3911c945 100644 --- a/src/table/Cargo.toml +++ b/src/table/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_table" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,20 +14,20 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_db = { version = "0.8.0", path = "../db" } -garage_rpc = { version = "0.8.0", path = "../rpc" } -garage_util = { version = "0.8.0", path = "../util" } +garage_db = { version = "0.8.1", path = "../db" } +garage_rpc = { version = "0.8.1", path = "../rpc" } +garage_util = { version = "0.8.1", path = "../util" } opentelemetry = "0.17" async-trait = "0.1.7" +arc-swap = "1.0" bytes = "1.0" hex = "0.4" hexdump = "0.1" tracing = "0.1.30" rand = "0.8" -rmp-serde = "0.15" serde = { version = "1.0", default-features = false, features = ["derive", "rc"] } serde_bytes = "0.11" diff --git a/src/table/data.rs b/src/table/data.rs index 3212e82b..5c792f1f 100644 --- a/src/table/data.rs +++ b/src/table/data.rs @@ -10,6 +10,7 @@ use garage_db::counted_tree_hack::CountedTree; use garage_util::data::*; use garage_util::error::*; +use garage_util::migrate::Migrate; use garage_rpc::system::System; @@ -31,16 +32,16 @@ pub struct TableData<F: TableSchema, R: TableReplication> { pub(crate) merkle_tree: db::Tree, pub(crate) merkle_todo: db::Tree, pub(crate) merkle_todo_notify: Notify, + + pub(crate) insert_queue: db::Tree, + pub(crate) insert_queue_notify: Notify, + pub(crate) gc_todo: CountedTree, pub(crate) metrics: TableMetrics, } -impl<F, R> TableData<F, R> -where - F: TableSchema, - R: TableReplication, -{ +impl<F: TableSchema, R: TableReplication> TableData<F, R> { pub fn new(system: Arc<System>, instance: F, replication: R, db: &db::Db) -> Arc<Self> { let store = db .open_tree(&format!("{}:table", F::TABLE_NAME)) @@ -53,12 +54,22 @@ where .open_tree(&format!("{}:merkle_todo", F::TABLE_NAME)) .expect("Unable to open DB Merkle TODO tree"); + let insert_queue = db + .open_tree(&format!("{}:insert_queue", F::TABLE_NAME)) + .expect("Unable to open insert queue DB tree"); + let gc_todo = db .open_tree(&format!("{}:gc_todo_v2", F::TABLE_NAME)) - .expect("Unable to open DB tree"); + .expect("Unable to open GC DB tree"); let gc_todo = CountedTree::new(gc_todo).expect("Cannot count gc_todo_v2"); - let metrics = TableMetrics::new(F::TABLE_NAME, merkle_todo.clone(), gc_todo.clone()); + let metrics = TableMetrics::new( + F::TABLE_NAME, + store.clone(), + merkle_tree.clone(), + merkle_todo.clone(), + gc_todo.clone(), + ); Arc::new(Self { system, @@ -68,6 +79,8 @@ where merkle_tree, merkle_todo, merkle_todo_notify: Notify::new(), + insert_queue, + insert_queue_notify: Notify::new(), gc_todo, metrics, }) @@ -167,9 +180,8 @@ where pub(crate) fn update_entry(&self, update_bytes: &[u8]) -> Result<(), Error> { let update = self.decode_entry(update_bytes)?; - let tree_key = self.tree_key(update.partition_key(), update.sort_key()); - self.update_entry_with(&tree_key[..], |ent| match ent { + self.update_entry_with(update.partition_key(), update.sort_key(), |ent| match ent { Some(mut ent) => { ent.merge(&update); ent @@ -181,11 +193,14 @@ where pub fn update_entry_with( &self, - tree_key: &[u8], + partition_key: &F::P, + sort_key: &F::S, f: impl Fn(Option<F::E>) -> F::E, ) -> Result<Option<F::E>, Error> { + let tree_key = self.tree_key(partition_key, sort_key); + let changed = self.store.db().transaction(|mut tx| { - let (old_entry, old_bytes, new_entry) = match tx.get(&self.store, tree_key)? { + let (old_entry, old_bytes, new_entry) = match tx.get(&self.store, &tree_key)? { Some(old_bytes) => { let old_entry = self.decode_entry(&old_bytes).map_err(db::TxError::Abort)?; let new_entry = f(Some(old_entry.clone())); @@ -194,23 +209,24 @@ where None => (None, None, f(None)), }; - // Scenario 1: the value changed, so of course there is a change - let value_changed = Some(&new_entry) != old_entry.as_ref(); - + // Changed can be true in two scenarios + // Scenario 1: the actual represented value changed, + // so of course the messagepack encoding changed as well // Scenario 2: the value didn't change but due to a migration in the - // data format, the messagepack encoding changed. In this case - // we have to write the migrated value in the table and update - // the associated Merkle tree entry. - let new_bytes = rmp_to_vec_all_named(&new_entry) + // data format, the messagepack encoding changed. In this case, + // we also have to write the migrated value in the table and update + // the associated Merkle tree entry. + let new_bytes = new_entry + .encode() .map_err(Error::RmpEncode) .map_err(db::TxError::Abort)?; - let encoding_changed = Some(&new_bytes[..]) != old_bytes.as_ref().map(|x| &x[..]); + let changed = Some(&new_bytes[..]) != old_bytes.as_deref(); drop(old_bytes); - if value_changed || encoding_changed { - let new_bytes_hash = blake2sum(&new_bytes[..]); - tx.insert(&self.merkle_todo, tree_key, new_bytes_hash.as_slice())?; - tx.insert(&self.store, tree_key, new_bytes)?; + if changed { + let new_bytes_hash = blake2sum(&new_bytes); + tx.insert(&self.merkle_todo, &tree_key, new_bytes_hash.as_slice())?; + tx.insert(&self.store, &tree_key, new_bytes)?; self.instance .updated(&mut tx, old_entry.as_ref(), Some(&new_entry))?; @@ -236,7 +252,7 @@ where let pk_hash = Hash::try_from(&tree_key[..32]).unwrap(); let nodes = self.replication.write_nodes(&pk_hash); if nodes.first() == Some(&self.system.id) { - GcTodoEntry::new(tree_key.to_vec(), new_bytes_hash).save(&self.gc_todo)?; + GcTodoEntry::new(tree_key, new_bytes_hash).save(&self.gc_todo)?; } } @@ -252,10 +268,11 @@ where .db() .transaction(|mut tx| match tx.get(&self.store, k)? { Some(cur_v) if cur_v == v => { + let old_entry = self.decode_entry(v).map_err(db::TxError::Abort)?; + tx.remove(&self.store, k)?; tx.insert(&self.merkle_todo, k, vec![])?; - let old_entry = self.decode_entry(v).map_err(db::TxError::Abort)?; self.instance.updated(&mut tx, Some(&old_entry), None)?; Ok(true) } @@ -279,10 +296,11 @@ where .db() .transaction(|mut tx| match tx.get(&self.store, k)? { Some(cur_v) if blake2sum(&cur_v[..]) == vhash => { + let old_entry = self.decode_entry(&cur_v[..]).map_err(db::TxError::Abort)?; + tx.remove(&self.store, k)?; tx.insert(&self.merkle_todo, k, vec![])?; - let old_entry = self.decode_entry(&cur_v[..]).map_err(db::TxError::Abort)?; self.instance.updated(&mut tx, Some(&old_entry), None)?; Ok(true) } @@ -296,6 +314,32 @@ where Ok(removed) } + // ---- Insert queue functions ---- + + pub(crate) fn queue_insert( + &self, + tx: &mut db::Transaction, + ins: &F::E, + ) -> db::TxResult<(), Error> { + let tree_key = self.tree_key(ins.partition_key(), ins.sort_key()); + + let new_entry = match tx.get(&self.insert_queue, &tree_key)? { + Some(old_v) => { + let mut entry = self.decode_entry(&old_v).map_err(db::TxError::Abort)?; + entry.merge(ins); + entry.encode() + } + None => ins.encode(), + }; + let new_entry = new_entry + .map_err(Error::RmpEncode) + .map_err(db::TxError::Abort)?; + tx.insert(&self.insert_queue, &tree_key, new_entry)?; + self.insert_queue_notify.notify_one(); + + Ok(()) + } + // ---- Utility functions ---- pub fn tree_key(&self, p: &F::P, s: &F::S) -> Vec<u8> { @@ -305,18 +349,18 @@ where } pub fn decode_entry(&self, bytes: &[u8]) -> Result<F::E, Error> { - match rmp_serde::decode::from_read_ref::<_, F::E>(bytes) { - Ok(x) => Ok(x), - Err(e) => match F::try_migrate(bytes) { - Some(x) => Ok(x), - None => { - warn!("Unable to decode entry of {}: {}", F::TABLE_NAME, e); - for line in hexdump::hexdump_iter(bytes) { - debug!("{}", line); - } - Err(e.into()) + match F::E::decode(bytes) { + Some(x) => Ok(x), + None => { + error!("Unable to decode entry of {}", F::TABLE_NAME); + for line in hexdump::hexdump_iter(bytes) { + debug!("{}", line); } - }, + Err(Error::Message(format!( + "Unable to decode entry of {}", + F::TABLE_NAME + ))) + } } } diff --git a/src/table/gc.rs b/src/table/gc.rs index 83e7eeff..5b9124a7 100644 --- a/src/table/gc.rs +++ b/src/table/gc.rs @@ -31,7 +31,7 @@ const TABLE_GC_BATCH_SIZE: usize = 1024; // and the moment the garbage collection actually happens) const TABLE_GC_DELAY: Duration = Duration::from_secs(24 * 3600); -pub(crate) struct TableGc<F: TableSchema + 'static, R: TableReplication + 'static> { +pub(crate) struct TableGc<F: TableSchema, R: TableReplication> { system: Arc<System>, data: Arc<TableData<F, R>>, @@ -49,29 +49,26 @@ impl Rpc for GcRpc { type Response = Result<GcRpc, Error>; } -impl<F, R> TableGc<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ - pub(crate) fn launch(system: Arc<System>, data: Arc<TableData<F, R>>) -> Arc<Self> { +impl<F: TableSchema, R: TableReplication> TableGc<F, R> { + pub(crate) fn new(system: Arc<System>, data: Arc<TableData<F, R>>) -> Arc<Self> { let endpoint = system .netapp .endpoint(format!("garage_table/gc.rs/Rpc:{}", F::TABLE_NAME)); let gc = Arc::new(Self { - system: system.clone(), + system, data, endpoint, }); - gc.endpoint.set_handler(gc.clone()); - system.background.spawn_worker(GcWorker::new(gc.clone())); - gc } + pub(crate) fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) { + bg.spawn_worker(GcWorker::new(self.clone())); + } + async fn gc_loop_iter(&self) -> Result<Option<Duration>, Error> { let now = now_msec(); @@ -276,11 +273,7 @@ where } #[async_trait] -impl<F, R> EndpointHandler<GcRpc> for TableGc<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> EndpointHandler<GcRpc> for TableGc<F, R> { async fn handle(self: &Arc<Self>, message: &GcRpc, _from: NodeID) -> Result<GcRpc, Error> { match message { GcRpc::Update(items) => { @@ -298,20 +291,12 @@ where } } -struct GcWorker<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +struct GcWorker<F: TableSchema, R: TableReplication> { gc: Arc<TableGc<F, R>>, wait_delay: Duration, } -impl<F, R> GcWorker<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> GcWorker<F, R> { fn new(gc: Arc<TableGc<F, R>>) -> Self { Self { gc, @@ -321,21 +306,15 @@ where } #[async_trait] -impl<F, R> Worker for GcWorker<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> Worker for GcWorker<F, R> { fn name(&self) -> String { format!("{} GC", F::TABLE_NAME) } - fn info(&self) -> Option<String> { - let l = self.gc.data.gc_todo_len().unwrap_or(0); - if l > 0 { - Some(format!("{} items in queue", l)) - } else { - None + fn status(&self) -> WorkerStatus { + WorkerStatus { + queue_length: Some(self.gc.data.gc_todo_len().unwrap_or(0) as u64), + ..Default::default() } } @@ -349,10 +328,7 @@ where } } - async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState { - if *must_exit.borrow() { - return WorkerState::Done; - } + async fn wait_for_work(&mut self) -> WorkerState { tokio::time::sleep(self.wait_delay).await; WorkerState::Busy } diff --git a/src/table/lib.rs b/src/table/lib.rs index b0153e9a..fdf114a6 100644 --- a/src/table/lib.rs +++ b/src/table/lib.rs @@ -4,16 +4,18 @@ #[macro_use] extern crate tracing; -mod metrics; pub mod schema; pub mod util; pub mod data; +pub mod replication; +pub mod table; + mod gc; mod merkle; -pub mod replication; +mod metrics; +mod queue; mod sync; -pub mod table; pub use schema::*; pub use table::*; diff --git a/src/table/merkle.rs b/src/table/merkle.rs index a5c29723..e86d0251 100644 --- a/src/table/merkle.rs +++ b/src/table/merkle.rs @@ -3,12 +3,14 @@ use std::time::Duration; use async_trait::async_trait; use serde::{Deserialize, Serialize}; +use tokio::select; use tokio::sync::watch; use garage_db as db; use garage_util::background::*; use garage_util::data::*; +use garage_util::encode::{nonversioned_decode, nonversioned_encode}; use garage_util::error::Error; use garage_rpc::ring::*; @@ -64,22 +66,18 @@ pub enum MerkleNode { Leaf(Vec<u8>, Hash), } -impl<F, R> MerkleUpdater<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ - pub(crate) fn launch(background: &BackgroundRunner, data: Arc<TableData<F, R>>) -> Arc<Self> { - let empty_node_hash = blake2sum(&rmp_to_vec_all_named(&MerkleNode::Empty).unwrap()[..]); +impl<F: TableSchema, R: TableReplication> MerkleUpdater<F, R> { + pub(crate) fn new(data: Arc<TableData<F, R>>) -> Arc<Self> { + let empty_node_hash = blake2sum(&nonversioned_encode(&MerkleNode::Empty).unwrap()[..]); - let ret = Arc::new(Self { + Arc::new(Self { data, empty_node_hash, - }); - - background.spawn_worker(MerkleWorker(ret.clone())); + }) + } - ret + pub(crate) fn spawn_workers(self: &Arc<Self>, background: &BackgroundRunner) { + background.spawn_worker(MerkleWorker(self.clone())); } fn updater_loop_iter(&self) -> Result<WorkerState, Error> { @@ -276,7 +274,7 @@ where tx.remove(&self.data.merkle_tree, k.encode())?; Ok(self.empty_node_hash) } else { - let vby = rmp_to_vec_all_named(v).map_err(|e| db::TxError::Abort(e.into()))?; + let vby = nonversioned_encode(v).map_err(|e| db::TxError::Abort(e.into()))?; let rethash = blake2sum(&vby[..]); tx.insert(&self.data.merkle_tree, k.encode(), vby)?; Ok(rethash) @@ -293,32 +291,27 @@ where Ok(self.data.merkle_tree.len()?) } + pub fn merkle_tree_fast_len(&self) -> Result<Option<usize>, Error> { + Ok(self.data.merkle_tree.fast_len()?) + } + pub fn todo_len(&self) -> Result<usize, Error> { Ok(self.data.merkle_todo.len()?) } } -struct MerkleWorker<F, R>(Arc<MerkleUpdater<F, R>>) -where - F: TableSchema + 'static, - R: TableReplication + 'static; +struct MerkleWorker<F: TableSchema, R: TableReplication>(Arc<MerkleUpdater<F, R>>); #[async_trait] -impl<F, R> Worker for MerkleWorker<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> Worker for MerkleWorker<F, R> { fn name(&self) -> String { - format!("{} Merkle tree updater", F::TABLE_NAME) + format!("{} Merkle", F::TABLE_NAME) } - fn info(&self) -> Option<String> { - let l = self.0.todo_len().unwrap_or(0); - if l > 0 { - Some(format!("{} items in queue", l)) - } else { - None + fn status(&self) -> WorkerStatus { + WorkerStatus { + queue_length: Some(self.0.todo_len().unwrap_or(0) as u64), + ..Default::default() } } @@ -337,11 +330,11 @@ where .unwrap() } - async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState { - if *must_exit.borrow() { - return WorkerState::Done; + async fn wait_for_work(&mut self) -> WorkerState { + select! { + _ = tokio::time::sleep(Duration::from_secs(60)) => (), + _ = self.0.data.merkle_todo_notify.notified() => (), } - tokio::time::sleep(Duration::from_secs(10)).await; WorkerState::Busy } } @@ -372,7 +365,7 @@ impl MerkleNode { fn decode_opt(ent: &Option<db::Value>) -> Result<Self, Error> { match ent { None => Ok(MerkleNode::Empty), - Some(v) => Ok(rmp_serde::decode::from_read_ref::<_, MerkleNode>(&v[..])?), + Some(v) => Ok(nonversioned_decode::<MerkleNode>(&v[..])?), } } diff --git a/src/table/metrics.rs b/src/table/metrics.rs index 3a1783e0..8318a84f 100644 --- a/src/table/metrics.rs +++ b/src/table/metrics.rs @@ -5,6 +5,8 @@ use garage_db::counted_tree_hack::CountedTree; /// TableMetrics reference all counter used for metrics pub struct TableMetrics { + pub(crate) _table_size: ValueObserver<u64>, + pub(crate) _merkle_tree_size: ValueObserver<u64>, pub(crate) _merkle_todo_len: ValueObserver<u64>, pub(crate) _gc_todo_len: ValueObserver<u64>, @@ -20,9 +22,43 @@ pub struct TableMetrics { pub(crate) sync_items_received: Counter<u64>, } impl TableMetrics { - pub fn new(table_name: &'static str, merkle_todo: db::Tree, gc_todo: CountedTree) -> Self { + pub fn new( + table_name: &'static str, + store: db::Tree, + merkle_tree: db::Tree, + merkle_todo: db::Tree, + gc_todo: CountedTree, + ) -> Self { let meter = global::meter(table_name); TableMetrics { + _table_size: meter + .u64_value_observer( + "table.size", + move |observer| { + if let Ok(Some(v)) = store.fast_len() { + observer.observe( + v as u64, + &[KeyValue::new("table_name", table_name)], + ); + } + }, + ) + .with_description("Number of items in table") + .init(), + _merkle_tree_size: meter + .u64_value_observer( + "table.merkle_tree_size", + move |observer| { + if let Ok(Some(v)) = merkle_tree.fast_len() { + observer.observe( + v as u64, + &[KeyValue::new("table_name", table_name)], + ); + } + }, + ) + .with_description("Number of nodes in table's Merkle tree") + .init(), _merkle_todo_len: meter .u64_value_observer( "table.merkle_updater_todo_queue_length", diff --git a/src/table/queue.rs b/src/table/queue.rs new file mode 100644 index 00000000..0857209b --- /dev/null +++ b/src/table/queue.rs @@ -0,0 +1,77 @@ +use std::sync::Arc; +use std::time::Duration; + +use async_trait::async_trait; +use tokio::select; +use tokio::sync::watch; + +use garage_util::background::*; +use garage_util::error::Error; + +use crate::replication::*; +use crate::schema::*; +use crate::table::*; + +const BATCH_SIZE: usize = 100; + +pub(crate) struct InsertQueueWorker<F, R>(pub(crate) Arc<Table<F, R>>) +where + F: TableSchema, + R: TableReplication; + +#[async_trait] +impl<F: TableSchema, R: TableReplication> Worker for InsertQueueWorker<F, R> { + fn name(&self) -> String { + format!("{} queue", F::TABLE_NAME) + } + + fn status(&self) -> WorkerStatus { + WorkerStatus { + queue_length: Some(self.0.data.insert_queue.len().unwrap_or(0) as u64), + ..Default::default() + } + } + + async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { + let mut kv_pairs = vec![]; + let mut values = vec![]; + + for entry_kv in self.0.data.insert_queue.iter()? { + let (k, v) = entry_kv?; + + values.push(self.0.data.decode_entry(&v)?); + kv_pairs.push((k, v)); + + if kv_pairs.len() > BATCH_SIZE { + break; + } + } + + if kv_pairs.is_empty() { + return Ok(WorkerState::Idle); + } + + self.0.insert_many(values).await?; + + self.0.data.insert_queue.db().transaction(|mut tx| { + for (k, v) in kv_pairs.iter() { + if let Some(v2) = tx.get(&self.0.data.insert_queue, k)? { + if &v2 == v { + tx.remove(&self.0.data.insert_queue, k)?; + } + } + } + Ok(()) + })?; + + Ok(WorkerState::Busy) + } + + async fn wait_for_work(&mut self) -> WorkerState { + select! { + _ = tokio::time::sleep(Duration::from_secs(600)) => (), + _ = self.0.data.insert_queue_notify.notified() => (), + } + WorkerState::Busy + } +} diff --git a/src/table/replication/parameters.rs b/src/table/replication/parameters.rs index 3740d947..f00815a2 100644 --- a/src/table/replication/parameters.rs +++ b/src/table/replication/parameters.rs @@ -2,7 +2,7 @@ use garage_rpc::ring::*; use garage_util::data::*; /// Trait to describe how a table shall be replicated -pub trait TableReplication: Send + Sync { +pub trait TableReplication: Send + Sync + 'static { // See examples in table_sharded.rs and table_fullcopy.rs // To understand various replication methods diff --git a/src/table/schema.rs b/src/table/schema.rs index f37e98d8..5cbf6c95 100644 --- a/src/table/schema.rs +++ b/src/table/schema.rs @@ -2,11 +2,14 @@ use serde::{Deserialize, Serialize}; use garage_db as db; use garage_util::data::*; +use garage_util::migrate::Migrate; use crate::crdt::Crdt; /// Trait for field used to partition data -pub trait PartitionKey { +pub trait PartitionKey: + Clone + PartialEq + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static +{ /// Get the key used to partition fn hash(&self) -> Hash; } @@ -27,7 +30,7 @@ impl PartitionKey for FixedBytes32 { } /// Trait for field used to sort data -pub trait SortKey { +pub trait SortKey: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static { /// Get the key used to sort fn sort_key(&self) -> &[u8]; } @@ -46,7 +49,7 @@ impl SortKey for FixedBytes32 { /// Trait for an entry in a table. It must be sortable and partitionnable. pub trait Entry<P: PartitionKey, S: SortKey>: - Crdt + PartialEq + Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + Crdt + PartialEq + Clone + Migrate + Send + Sync + 'static { /// Get the key used to partition fn partition_key(&self) -> &P; @@ -65,23 +68,16 @@ pub trait TableSchema: Send + Sync + 'static { const TABLE_NAME: &'static str; /// The partition key used in that table - type P: PartitionKey + Clone + PartialEq + Serialize + for<'de> Deserialize<'de> + Send + Sync; + type P: PartitionKey; /// The sort key used int that table - type S: SortKey + Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync; + type S: SortKey; /// They type for an entry in that table type E: Entry<Self::P, Self::S>; /// The type for a filter that can be applied to select entries /// (e.g. filter out deleted entries) - type Filter: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync; - - // Action to take if not able to decode current version: - // try loading from an older version - /// Try migrating an entry from an older version - fn try_migrate(_bytes: &[u8]) -> Option<Self::E> { - None - } + type Filter: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static; /// Actions triggered by data changing in a table. If such actions /// include updates to the local database that should be applied diff --git a/src/table/sync.rs b/src/table/sync.rs index 9d79d856..92a353c6 100644 --- a/src/table/sync.rs +++ b/src/table/sync.rs @@ -2,6 +2,7 @@ use std::collections::VecDeque; use std::sync::Arc; use std::time::{Duration, Instant}; +use arc_swap::ArcSwapOption; use async_trait::async_trait; use futures_util::stream::*; use opentelemetry::KeyValue; @@ -13,7 +14,8 @@ use tokio::sync::{mpsc, watch}; use garage_util::background::*; use garage_util::data::*; -use garage_util::error::Error; +use garage_util::encode::{debug_serialize, nonversioned_encode}; +use garage_util::error::{Error, OkOrMessage}; use garage_rpc::ring::*; use garage_rpc::system::System; @@ -27,12 +29,12 @@ use crate::*; // Do anti-entropy every 10 minutes const ANTI_ENTROPY_INTERVAL: Duration = Duration::from_secs(10 * 60); -pub struct TableSyncer<F: TableSchema + 'static, R: TableReplication + 'static> { +pub struct TableSyncer<F: TableSchema, R: TableReplication> { system: Arc<System>, data: Arc<TableData<F, R>>, merkle: Arc<MerkleUpdater<F, R>>, - add_full_sync_tx: mpsc::UnboundedSender<()>, + add_full_sync_tx: ArcSwapOption<mpsc::UnboundedSender<()>>, endpoint: Arc<Endpoint<SyncRpc, Self>>, } @@ -60,12 +62,8 @@ struct TodoPartition { retain: bool, } -impl<F, R> TableSyncer<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ - pub(crate) fn launch( +impl<F: TableSchema, R: TableReplication> TableSyncer<F, R> { + pub(crate) fn new( system: Arc<System>, data: Arc<TableData<F, R>>, merkle: Arc<MerkleUpdater<F, R>>, @@ -74,34 +72,40 @@ where .netapp .endpoint(format!("garage_table/sync.rs/Rpc:{}", F::TABLE_NAME)); - let (add_full_sync_tx, add_full_sync_rx) = mpsc::unbounded_channel(); - let syncer = Arc::new(Self { - system: system.clone(), + system, data, merkle, - add_full_sync_tx, + add_full_sync_tx: ArcSwapOption::new(None), endpoint, }); - syncer.endpoint.set_handler(syncer.clone()); - system.background.spawn_worker(SyncWorker { - syncer: syncer.clone(), - ring_recv: system.ring.clone(), - ring: system.ring.borrow().clone(), + syncer + } + + pub(crate) fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) { + let (add_full_sync_tx, add_full_sync_rx) = mpsc::unbounded_channel(); + self.add_full_sync_tx + .store(Some(Arc::new(add_full_sync_tx))); + + bg.spawn_worker(SyncWorker { + syncer: self.clone(), + ring_recv: self.system.ring.clone(), + ring: self.system.ring.borrow().clone(), add_full_sync_rx, todo: vec![], next_full_sync: Instant::now() + Duration::from_secs(20), }); - - syncer } - pub fn add_full_sync(&self) { - if self.add_full_sync_tx.send(()).is_err() { - error!("({}) Could not add full sync", F::TABLE_NAME); - } + pub fn add_full_sync(&self) -> Result<(), Error> { + let tx = self.add_full_sync_tx.load(); + let tx = tx + .as_ref() + .ok_or_message("table sync worker is not running")?; + tx.send(()).ok_or_message("send error")?; + Ok(()) } // ---- @@ -295,7 +299,7 @@ where ); return Ok(()); } - let root_ck_hash = hash_of::<MerkleNode>(&root_ck)?; + let root_ck_hash = hash_of_merkle_node(&root_ck)?; // Check if they have the same root checksum // If so, do nothing. @@ -452,16 +456,12 @@ where // ======= SYNCHRONIZATION PROCEDURE -- RECEIVER SIDE ====== #[async_trait] -impl<F, R> EndpointHandler<SyncRpc> for TableSyncer<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> EndpointHandler<SyncRpc> for TableSyncer<F, R> { async fn handle(self: &Arc<Self>, message: &SyncRpc, from: NodeID) -> Result<SyncRpc, Error> { match message { SyncRpc::RootCkHash(range, h) => { let (_root_ck_key, root_ck) = self.get_root_ck(*range)?; - let hash = hash_of::<MerkleNode>(&root_ck)?; + let hash = hash_of_merkle_node(&root_ck)?; Ok(SyncRpc::RootCkDifferent(hash != *h)) } SyncRpc::GetNode(k) => { @@ -490,7 +490,7 @@ where // -------- Sync Worker --------- -struct SyncWorker<F: TableSchema + 'static, R: TableReplication + 'static> { +struct SyncWorker<F: TableSchema, R: TableReplication> { syncer: Arc<TableSyncer<F, R>>, ring_recv: watch::Receiver<Arc<Ring>>, ring: Arc<Ring>, @@ -499,7 +499,7 @@ struct SyncWorker<F: TableSchema + 'static, R: TableReplication + 'static> { next_full_sync: Instant, } -impl<F: TableSchema + 'static, R: TableReplication + 'static> SyncWorker<F, R> { +impl<F: TableSchema, R: TableReplication> SyncWorker<F, R> { fn add_full_sync(&mut self) { let system = &self.syncer.system; let data = &self.syncer.data; @@ -565,17 +565,15 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> SyncWorker<F, R> { } #[async_trait] -impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWorker<F, R> { +impl<F: TableSchema, R: TableReplication> Worker for SyncWorker<F, R> { fn name(&self) -> String { format!("{} sync", F::TABLE_NAME) } - fn info(&self) -> Option<String> { - let l = self.todo.len(); - if l > 0 { - Some(format!("{} partitions remaining", l)) - } else { - None + fn status(&self) -> WorkerStatus { + WorkerStatus { + queue_length: Some(self.todo.len() as u64), + ..Default::default() } } @@ -588,10 +586,7 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWor } } - async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState { - if *must_exit.borrow() { - return WorkerState::Done; - } + async fn wait_for_work(&mut self) -> WorkerState { select! { s = self.add_full_sync_rx.recv() => { if let Some(()) = s { @@ -620,8 +615,8 @@ impl<F: TableSchema + 'static, R: TableReplication + 'static> Worker for SyncWor // ---- UTIL ---- -fn hash_of<T: Serialize>(x: &T) -> Result<Hash, Error> { - Ok(blake2sum(&rmp_to_vec_all_named(x)?[..])) +fn hash_of_merkle_node(x: &MerkleNode) -> Result<Hash, Error> { + Ok(blake2sum(&nonversioned_encode(x)?[..])) } fn join_ordered<'a, K: Ord + Eq, V1, V2>( diff --git a/src/table/table.rs b/src/table/table.rs index 8a66c420..7ad79677 100644 --- a/src/table/table.rs +++ b/src/table/table.rs @@ -14,9 +14,11 @@ use opentelemetry::{ use garage_db as db; +use garage_util::background::BackgroundRunner; use garage_util::data::*; use garage_util::error::Error; use garage_util::metrics::RecordDuration; +use garage_util::migrate::Migrate; use garage_rpc::system::System; use garage_rpc::*; @@ -25,16 +27,18 @@ use crate::crdt::Crdt; use crate::data::*; use crate::gc::*; use crate::merkle::*; +use crate::queue::InsertQueueWorker; use crate::replication::*; use crate::schema::*; use crate::sync::*; use crate::util::*; -pub struct Table<F: TableSchema + 'static, R: TableReplication + 'static> { +pub struct Table<F: TableSchema, R: TableReplication> { pub system: Arc<System>, pub data: Arc<TableData<F, R>>, pub merkle_updater: Arc<MerkleUpdater<F, R>>, pub syncer: Arc<TableSyncer<F, R>>, + gc: Arc<TableGc<F, R>>, endpoint: Arc<Endpoint<TableRpc<F>, Self>>, } @@ -61,11 +65,7 @@ impl<F: TableSchema> Rpc for TableRpc<F> { type Response = Result<TableRpc<F>, Error>; } -impl<F, R> Table<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> Table<F, R> { // =============== PUBLIC INTERFACE FUNCTIONS (new, insert, get, etc) =============== pub fn new(instance: F, replication: R, system: Arc<System>, db: &db::Db) -> Arc<Self> { @@ -75,15 +75,16 @@ where let data = TableData::new(system.clone(), instance, replication, db); - let merkle_updater = MerkleUpdater::launch(&system.background, data.clone()); + let merkle_updater = MerkleUpdater::new(data.clone()); - let syncer = TableSyncer::launch(system.clone(), data.clone(), merkle_updater.clone()); - TableGc::launch(system.clone(), data.clone()); + let syncer = TableSyncer::new(system.clone(), data.clone(), merkle_updater.clone()); + let gc = TableGc::new(system.clone(), data.clone()); let table = Arc::new(Self { system, data, merkle_updater, + gc, syncer, endpoint, }); @@ -93,6 +94,13 @@ where table } + pub fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) { + self.merkle_updater.spawn_workers(bg); + self.syncer.spawn_workers(bg); + self.gc.spawn_workers(bg); + bg.spawn_worker(InsertQueueWorker(self.clone())); + } + pub async fn insert(&self, e: &F::E) -> Result<(), Error> { let tracer = opentelemetry::global::tracer("garage_table"); let span = tracer.start(format!("{} insert", F::TABLE_NAME)); @@ -111,7 +119,7 @@ where let hash = e.partition_key().hash(); let who = self.data.replication.write_nodes(&hash); - let e_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(e)?)); + let e_enc = Arc::new(ByteBuf::from(e.encode()?)); let rpc = TableRpc::<F>::Update(vec![e_enc]); self.system @@ -128,6 +136,11 @@ where Ok(()) } + /// Insert item locally + pub fn queue_insert(&self, tx: &mut db::Transaction, e: &F::E) -> db::TxResult<(), Error> { + self.data.queue_insert(tx, e) + } + pub async fn insert_many<I, IE>(&self, entries: I) -> Result<(), Error> where I: IntoIterator<Item = IE> + Send + Sync, @@ -157,7 +170,7 @@ where let entry = entry.borrow(); let hash = entry.partition_key().hash(); let who = self.data.replication.write_nodes(&hash); - let e_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(entry)?)); + let e_enc = Arc::new(ByteBuf::from(entry.encode()?)); for node in who { call_list.entry(node).or_default().push(e_enc.clone()); } @@ -259,9 +272,11 @@ where if not_all_same { let self2 = self.clone(); let ent2 = ret_entry.clone(); - self.system - .background - .spawn_cancellable(async move { self2.repair_on_read(&who[..], ent2).await }); + tokio::spawn(async move { + if let Err(e) = self2.repair_on_read(&who[..], ent2).await { + warn!("Error doing repair on read: {}", e); + } + }); } } @@ -358,11 +373,12 @@ where .into_iter() .map(|k| ret.get(&k).unwrap().clone()) .collect::<Vec<_>>(); - self.system.background.spawn_cancellable(async move { + tokio::spawn(async move { for v in to_repair { - self2.repair_on_read(&who[..], v).await?; + if let Err(e) = self2.repair_on_read(&who[..], v).await { + warn!("Error doing repair on read: {}", e); + } } - Ok(()) }); } @@ -393,7 +409,7 @@ where // =============== UTILITY FUNCTION FOR CLIENT OPERATIONS =============== async fn repair_on_read(&self, who: &[Uuid], what: F::E) -> Result<(), Error> { - let what_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(&what)?)); + let what_enc = Arc::new(ByteBuf::from(what.encode()?)); self.system .rpc .try_call_many( @@ -408,11 +424,7 @@ where } #[async_trait] -impl<F, R> EndpointHandler<TableRpc<F>> for Table<F, R> -where - F: TableSchema + 'static, - R: TableReplication + 'static, -{ +impl<F: TableSchema, R: TableReplication> EndpointHandler<TableRpc<F>> for Table<F, R> { async fn handle( self: &Arc<Self>, msg: &TableRpc<F>, diff --git a/src/table/util.rs b/src/table/util.rs index 20595a94..0b10cf3f 100644 --- a/src/table/util.rs +++ b/src/table/util.rs @@ -49,3 +49,9 @@ impl EnumerationOrder { } } } + +impl Default for EnumerationOrder { + fn default() -> Self { + EnumerationOrder::Forward + } +} diff --git a/src/util/Cargo.toml b/src/util/Cargo.toml index 8e978fc2..32e9c851 100644 --- a/src/util/Cargo.toml +++ b/src/util/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_util" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>"] edition = "2018" license = "AGPL-3.0" @@ -14,7 +14,7 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_db = { version = "0.8.0", path = "../db" } +garage_db = { version = "0.8.1", path = "../db" } arc-swap = "1.0" async-trait = "0.1" @@ -23,6 +23,7 @@ bytes = "1.0" digest = "0.10" err-derive = "0.3" git-version = "0.3.4" +hexdump = "0.1" xxhash-rust = { version = "0.8", default-features = false, features = ["xxh3"] } hex = "0.4" lazy_static = "1.4" diff --git a/src/util/background/job_worker.rs b/src/util/background/job_worker.rs deleted file mode 100644 index 2568ea11..00000000 --- a/src/util/background/job_worker.rs +++ /dev/null @@ -1,48 +0,0 @@ -//! Job worker: a generic worker that just processes incoming -//! jobs one by one - -use std::sync::Arc; - -use async_trait::async_trait; -use tokio::sync::{mpsc, Mutex}; - -use crate::background::worker::*; -use crate::background::*; - -pub(crate) struct JobWorker { - pub(crate) index: usize, - pub(crate) job_chan: Arc<Mutex<mpsc::UnboundedReceiver<(Job, bool)>>>, - pub(crate) next_job: Option<Job>, -} - -#[async_trait] -impl Worker for JobWorker { - fn name(&self) -> String { - format!("Job worker #{}", self.index) - } - - async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> { - match self.next_job.take() { - None => return Ok(WorkerState::Idle), - Some(job) => { - job.await?; - Ok(WorkerState::Busy) - } - } - } - - async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState { - loop { - match self.job_chan.lock().await.recv().await { - Some((job, cancellable)) => { - if cancellable && *must_exit.borrow() { - continue; - } - self.next_job = Some(job); - return WorkerState::Busy; - } - None => return WorkerState::Done, - } - } - } -} diff --git a/src/util/background/mod.rs b/src/util/background/mod.rs index 619f5068..41b48e93 100644 --- a/src/util/background/mod.rs +++ b/src/util/background/mod.rs @@ -1,27 +1,18 @@ //! Job runner for futures and async functions -pub mod job_worker; pub mod worker; -use core::future::Future; - use std::collections::HashMap; -use std::pin::Pin; use std::sync::Arc; use serde::{Deserialize, Serialize}; -use tokio::sync::{mpsc, watch, Mutex}; +use tokio::sync::{mpsc, watch}; -use crate::error::Error; use worker::WorkerProcessor; pub use worker::{Worker, WorkerState}; -pub(crate) type JobOutput = Result<(), Error>; -pub(crate) type Job = Pin<Box<dyn Future<Output = JobOutput> + Send>>; - /// Job runner for futures and async functions pub struct BackgroundRunner { - send_job: mpsc::UnboundedSender<(Job, bool)>, send_worker: mpsc::UnboundedSender<Box<dyn Worker>>, worker_info: Arc<std::sync::Mutex<HashMap<usize, WorkerInfo>>>, } @@ -29,19 +20,27 @@ pub struct BackgroundRunner { #[derive(Clone, Serialize, Deserialize, Debug)] pub struct WorkerInfo { pub name: String, - pub info: Option<String>, + pub status: WorkerStatus, pub state: WorkerState, pub errors: usize, pub consecutive_errors: usize, pub last_error: Option<(String, u64)>, } +/// WorkerStatus is a struct returned by the worker with a bunch of canonical +/// fields to indicate their status to CLI users. All fields are optional. +#[derive(Clone, Serialize, Deserialize, Debug, Default)] +pub struct WorkerStatus { + pub tranquility: Option<u32>, + pub progress: Option<String>, + pub queue_length: Option<u64>, + pub persistent_errors: Option<u64>, + pub freeform: Vec<String>, +} + impl BackgroundRunner { /// Create a new BackgroundRunner - pub fn new( - n_runners: usize, - stop_signal: watch::Receiver<bool>, - ) -> (Arc<Self>, tokio::task::JoinHandle<()>) { + pub fn new(stop_signal: watch::Receiver<bool>) -> (Arc<Self>, tokio::task::JoinHandle<()>) { let (send_worker, worker_out) = mpsc::unbounded_channel::<Box<dyn Worker>>(); let worker_info = Arc::new(std::sync::Mutex::new(HashMap::new())); @@ -52,24 +51,7 @@ impl BackgroundRunner { worker_processor.run().await; }); - let (send_job, queue_out) = mpsc::unbounded_channel(); - let queue_out = Arc::new(Mutex::new(queue_out)); - - for i in 0..n_runners { - let queue_out = queue_out.clone(); - - send_worker - .send(Box::new(job_worker::JobWorker { - index: i, - job_chan: queue_out.clone(), - next_job: None, - })) - .ok() - .unwrap(); - } - let bgrunner = Arc::new(Self { - send_job, send_worker, worker_info, }); @@ -80,31 +62,6 @@ impl BackgroundRunner { self.worker_info.lock().unwrap().clone() } - /// Spawn a task to be run in background - pub fn spawn<T>(&self, job: T) - where - T: Future<Output = JobOutput> + Send + 'static, - { - let boxed: Job = Box::pin(job); - self.send_job - .send((boxed, false)) - .ok() - .expect("Could not put job in queue"); - } - - /// Spawn a task to be run in background. It may get discarded before running if spawned while - /// the runner is stopping - pub fn spawn_cancellable<T>(&self, job: T) - where - T: Future<Output = JobOutput> + Send + 'static, - { - let boxed: Job = Box::pin(job); - self.send_job - .send((boxed, true)) - .ok() - .expect("Could not put job in queue"); - } - pub fn spawn_worker<W>(&self, worker: W) where W: Worker + 'static, diff --git a/src/util/background/worker.rs b/src/util/background/worker.rs index f5e3addb..8165e2cb 100644 --- a/src/util/background/worker.rs +++ b/src/util/background/worker.rs @@ -1,6 +1,6 @@ use std::collections::HashMap; use std::sync::Arc; -use std::time::{Duration, Instant}; +use std::time::Duration; use async_trait::async_trait; use futures::future::*; @@ -10,10 +10,14 @@ use serde::{Deserialize, Serialize}; use tokio::select; use tokio::sync::{mpsc, watch}; -use crate::background::WorkerInfo; +use crate::background::{WorkerInfo, WorkerStatus}; use crate::error::Error; use crate::time::now_msec; +// All workers that haven't exited for this time after an exit signal was recieved +// will be interrupted in the middle of whatever they are doing. +const EXIT_DEADLINE: Duration = Duration::from_secs(8); + #[derive(PartialEq, Copy, Clone, Serialize, Deserialize, Debug)] pub enum WorkerState { Busy, @@ -26,7 +30,7 @@ impl std::fmt::Display for WorkerState { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { match self { WorkerState::Busy => write!(f, "Busy"), - WorkerState::Throttled(t) => write!(f, "Thr:{:.3}", t), + WorkerState::Throttled(_) => write!(f, "Busy*"), WorkerState::Idle => write!(f, "Idle"), WorkerState::Done => write!(f, "Done"), } @@ -37,8 +41,8 @@ impl std::fmt::Display for WorkerState { pub trait Worker: Send { fn name(&self) -> String; - fn info(&self) -> Option<String> { - None + fn status(&self) -> WorkerStatus { + Default::default() } /// Work: do a basic unit of work, if one is available (otherwise, should return @@ -50,10 +54,8 @@ pub trait Worker: Send { async fn work(&mut self, must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error>; /// Wait for work: await for some task to become available. This future can be interrupted in - /// the middle for any reason. This future doesn't have to await on must_exit.changed(), we - /// are doing it for you. Therefore it only receives a read refernce to must_exit which allows - /// it to check if we are exiting. - async fn wait_for_work(&mut self, must_exit: &watch::Receiver<bool>) -> WorkerState; + /// the middle for any reason, for example if an interrupt signal was recieved. + async fn wait_for_work(&mut self) -> WorkerState; } pub(crate) struct WorkerProcessor { @@ -93,11 +95,9 @@ impl WorkerProcessor { let task_id = next_task_id; next_task_id += 1; let stop_signal = self.stop_signal.clone(); - let stop_signal_worker = self.stop_signal.clone(); let mut worker = WorkerHandler { task_id, stop_signal, - stop_signal_worker, worker: new_worker, state: WorkerState::Busy, errors: 0, @@ -119,7 +119,7 @@ impl WorkerProcessor { match wi.get_mut(&worker.task_id) { Some(i) => { i.state = worker.state; - i.info = worker.worker.info(); + i.status = worker.worker.status(); i.errors = worker.errors; i.consecutive_errors = worker.consecutive_errors; if worker.last_error.is_some() { @@ -130,7 +130,7 @@ impl WorkerProcessor { wi.insert(worker.task_id, WorkerInfo { name: worker.worker.name(), state: worker.state, - info: worker.worker.info(), + status: worker.worker.status(), errors: worker.errors, consecutive_errors: worker.consecutive_errors, last_error: worker.last_error.take(), @@ -153,26 +153,14 @@ impl WorkerProcessor { } // We are exiting, drain everything - let drain_half_time = Instant::now() + Duration::from_secs(5); let drain_everything = async move { - while let Some(mut worker) = workers.next().await { - if worker.state == WorkerState::Done { - info!( - "Worker {} (TID {}) exited", - worker.worker.name(), - worker.task_id - ); - } else if Instant::now() > drain_half_time { - warn!("Worker {} (TID {}) interrupted between two iterations in state {:?} (this should be fine)", worker.worker.name(), worker.task_id, worker.state); - } else { - workers.push( - async move { - worker.step().await; - worker - } - .boxed(), - ); - } + while let Some(worker) = workers.next().await { + info!( + "Worker {} (TID {}) exited (last state: {:?})", + worker.worker.name(), + worker.task_id, + worker.state + ); } }; @@ -180,7 +168,7 @@ impl WorkerProcessor { _ = drain_everything => { info!("All workers exited peacefully \\o/"); } - _ = tokio::time::sleep(Duration::from_secs(9)) => { + _ = tokio::time::sleep(EXIT_DEADLINE) => { error!("Some workers could not exit in time, we are cancelling some things in the middle"); } } @@ -190,7 +178,6 @@ impl WorkerProcessor { struct WorkerHandler { task_id: usize, stop_signal: watch::Receiver<bool>, - stop_signal_worker: watch::Receiver<bool>, worker: Box<dyn Worker>, state: WorkerState, errors: usize, @@ -225,33 +212,19 @@ impl WorkerHandler { }, WorkerState::Throttled(delay) => { // Sleep for given delay and go back to busy state - if !*self.stop_signal.borrow() { - select! { - _ = tokio::time::sleep(Duration::from_secs_f32(delay)) => (), - _ = self.stop_signal.changed() => (), + select! { + _ = tokio::time::sleep(Duration::from_secs_f32(delay)) => { + self.state = WorkerState::Busy; } + _ = self.stop_signal.changed() => (), } - self.state = WorkerState::Busy; } WorkerState::Idle => { - if *self.stop_signal.borrow() { - select! { - new_st = self.worker.wait_for_work(&self.stop_signal_worker) => { - self.state = new_st; - } - _ = tokio::time::sleep(Duration::from_secs(1)) => { - // stay in Idle state - } - } - } else { - select! { - new_st = self.worker.wait_for_work(&self.stop_signal_worker) => { - self.state = new_st; - } - _ = self.stop_signal.changed() => { - // stay in Idle state - } + select! { + new_st = self.worker.wait_for_work() => { + self.state = new_st; } + _ = self.stop_signal.changed() => (), } } WorkerState::Done => unreachable!(), diff --git a/src/util/data.rs b/src/util/data.rs index 7715c2cc..3f61e301 100644 --- a/src/util/data.rs +++ b/src/util/data.rs @@ -140,34 +140,3 @@ pub fn fasthash(data: &[u8]) -> FastHash { pub fn gen_uuid() -> Uuid { rand::thread_rng().gen::<[u8; 32]>().into() } - -// RMP serialization with names of fields and variants - -/// Serialize to MessagePack -pub fn rmp_to_vec_all_named<T>(val: &T) -> Result<Vec<u8>, rmp_serde::encode::Error> -where - T: Serialize + ?Sized, -{ - let mut wr = Vec::with_capacity(128); - let mut se = rmp_serde::Serializer::new(&mut wr) - .with_struct_map() - .with_string_variants(); - val.serialize(&mut se)?; - Ok(wr) -} - -/// Serialize to JSON, truncating long result -pub fn debug_serialize<T: Serialize>(x: T) -> String { - match serde_json::to_string(&x) { - Ok(ss) => { - if ss.len() > 100 { - // TODO this can panic if 100 is not a codepoint boundary, but inside a 2 Bytes - // (or more) codepoint - ss[..100].to_string() - } else { - ss - } - } - Err(e) => format!("<JSON serialization error: {}>", e), - } -} diff --git a/src/util/encode.rs b/src/util/encode.rs new file mode 100644 index 00000000..1cd3198f --- /dev/null +++ b/src/util/encode.rs @@ -0,0 +1,42 @@ +use serde::{Deserialize, Serialize}; + +/// Serialize to MessagePacki, without versionning +/// (see garage_util::migrate for functions that manage versionned +/// data formats) +pub fn nonversioned_encode<T>(val: &T) -> Result<Vec<u8>, rmp_serde::encode::Error> +where + T: Serialize + ?Sized, +{ + let mut wr = Vec::with_capacity(128); + let mut se = rmp_serde::Serializer::new(&mut wr) + .with_struct_map() + .with_string_variants(); + val.serialize(&mut se)?; + Ok(wr) +} + +/// Deserialize from MessagePacki, without versionning +/// (see garage_util::migrate for functions that manage versionned +/// data formats) +pub fn nonversioned_decode<T>(bytes: &[u8]) -> Result<T, rmp_serde::decode::Error> +where + T: for<'de> Deserialize<'de> + ?Sized, +{ + rmp_serde::decode::from_read_ref::<_, T>(bytes) +} + +/// Serialize to JSON, truncating long result +pub fn debug_serialize<T: Serialize>(x: T) -> String { + match serde_json::to_string(&x) { + Ok(ss) => { + if ss.len() > 100 { + // TODO this can panic if 100 is not a codepoint boundary, but inside a 2 Bytes + // (or more) codepoint + ss[..100].to_string() + } else { + ss + } + } + Err(e) => format!("<JSON serialization error: {}>", e), + } +} diff --git a/src/util/error.rs b/src/util/error.rs index 9995c746..3fcee71d 100644 --- a/src/util/error.rs +++ b/src/util/error.rs @@ -7,6 +7,7 @@ use err_derive::Error; use serde::{de::Visitor, Deserialize, Deserializer, Serialize, Serializer}; use crate::data::*; +use crate::encode::debug_serialize; /// Regroup all Garage errors #[derive(Debug, Error)] diff --git a/src/util/formater.rs b/src/util/formater.rs index 95324f9a..2ea53ebb 100644 --- a/src/util/formater.rs +++ b/src/util/formater.rs @@ -1,4 +1,4 @@ -pub fn format_table(data: Vec<String>) { +pub fn format_table_to_string(data: Vec<String>) -> String { let data = data .iter() .map(|s| s.split('\t').collect::<Vec<_>>()) @@ -24,5 +24,9 @@ pub fn format_table(data: Vec<String>) { out.push('\n'); } - print!("{}", out); + out +} + +pub fn format_table(data: Vec<String>) { + print!("{}", format_table_to_string(data)); } diff --git a/src/util/lib.rs b/src/util/lib.rs index 264cc192..be82061f 100644 --- a/src/util/lib.rs +++ b/src/util/lib.rs @@ -8,9 +8,11 @@ pub mod background; pub mod config; pub mod crdt; pub mod data; +pub mod encode; pub mod error; pub mod formater; pub mod metrics; +pub mod migrate; pub mod persister; pub mod time; pub mod token_bucket; diff --git a/src/util/migrate.rs b/src/util/migrate.rs new file mode 100644 index 00000000..1229fd9c --- /dev/null +++ b/src/util/migrate.rs @@ -0,0 +1,159 @@ +use serde::{Deserialize, Serialize}; + +/// Indicates that this type has an encoding that can be migrated from +/// a previous version upon upgrades of Garage. +pub trait Migrate: Serialize + for<'de> Deserialize<'de> + 'static { + /// A sequence of bytes to add at the beginning of the serialized + /// string, to identify that the data is of this version. + const VERSION_MARKER: &'static [u8] = b""; + + /// The previous version of this data type, from which items of this version + /// can be migrated. + type Previous: Migrate; + + /// The migration function that transforms a value decoded in the old format + /// to an up-to-date value. + fn migrate(previous: Self::Previous) -> Self; + + /// Decode an encoded version of this type, going through a migration if necessary. + fn decode(bytes: &[u8]) -> Option<Self> { + let marker_len = Self::VERSION_MARKER.len(); + if bytes.get(..marker_len) == Some(Self::VERSION_MARKER) { + if let Ok(value) = rmp_serde::decode::from_read_ref::<_, Self>(&bytes[marker_len..]) { + return Some(value); + } + } + + Self::Previous::decode(bytes).map(Self::migrate) + } + + /// Encode this type with optionnal version marker + fn encode(&self) -> Result<Vec<u8>, rmp_serde::encode::Error> { + let mut wr = Vec::with_capacity(128); + wr.extend_from_slice(Self::VERSION_MARKER); + let mut se = rmp_serde::Serializer::new(&mut wr) + .with_struct_map() + .with_string_variants(); + self.serialize(&mut se)?; + Ok(wr) + } +} + +/// Indicates that this type has no previous encoding version to be migrated from. +pub trait InitialFormat: Serialize + for<'de> Deserialize<'de> + 'static { + /// A sequence of bytes to add at the beginning of the serialized + /// string, to identify that the data is of this version. + const VERSION_MARKER: &'static [u8] = b""; +} + +impl<T: InitialFormat> Migrate for T { + const VERSION_MARKER: &'static [u8] = <T as InitialFormat>::VERSION_MARKER; + + type Previous = NoPrevious; + + fn migrate(_previous: Self::Previous) -> Self { + unreachable!(); + } +} + +/// Internal type used by InitialFormat, not meant for general use. +#[derive(Serialize, Deserialize)] +pub enum NoPrevious {} + +impl Migrate for NoPrevious { + type Previous = NoPrevious; + + fn migrate(_previous: Self::Previous) -> Self { + unreachable!(); + } + + fn decode(_bytes: &[u8]) -> Option<Self> { + None + } + + fn encode(&self) -> Result<Vec<u8>, rmp_serde::encode::Error> { + unreachable!() + } +} + +#[cfg(test)] +mod test { + use super::*; + + #[derive(Serialize, Deserialize, PartialEq, Eq, Debug)] + struct V1 { + a: usize, + b: String, + } + impl InitialFormat for V1 {} + + #[derive(Serialize, Deserialize, PartialEq, Eq, Debug)] + struct V2 { + a: usize, + b: Vec<String>, + c: String, + } + impl Migrate for V2 { + const VERSION_MARKER: &'static [u8] = b"GtestV2"; + type Previous = V1; + fn migrate(prev: V1) -> V2 { + V2 { + a: prev.a, + b: vec![prev.b], + c: String::new(), + } + } + } + + #[test] + fn test_v1() { + let x = V1 { + a: 12, + b: "hello".into(), + }; + let x_enc = x.encode().unwrap(); + let y = V1::decode(&x_enc).unwrap(); + assert_eq!(x, y); + } + + #[test] + fn test_v2() { + let x = V2 { + a: 12, + b: vec!["hello".into(), "world".into()], + c: "plop".into(), + }; + let x_enc = x.encode().unwrap(); + assert_eq!(&x_enc[..V2::VERSION_MARKER.len()], V2::VERSION_MARKER); + let y = V2::decode(&x_enc).unwrap(); + assert_eq!(x, y); + } + + #[test] + fn test_migrate() { + let x = V1 { + a: 12, + b: "hello".into(), + }; + let x_enc = x.encode().unwrap(); + + let xx = V1::decode(&x_enc).unwrap(); + assert_eq!(x, xx); + + let y = V2::decode(&x_enc).unwrap(); + assert_eq!( + y, + V2 { + a: 12, + b: vec!["hello".into()], + c: "".into(), + } + ); + + let y_enc = y.encode().unwrap(); + assert_eq!(&y_enc[..V2::VERSION_MARKER.len()], V2::VERSION_MARKER); + + let z = V2::decode(&y_enc).unwrap(); + assert_eq!(y, z); + } +} diff --git a/src/util/persister.rs b/src/util/persister.rs index 9e1a1910..4b9adf51 100644 --- a/src/util/persister.rs +++ b/src/util/persister.rs @@ -3,21 +3,16 @@ use std::path::{Path, PathBuf}; use tokio::io::{AsyncReadExt, AsyncWriteExt}; -use serde::{Deserialize, Serialize}; - -use crate::data::*; use crate::error::Error; +use crate::migrate::Migrate; -pub struct Persister<T: Serialize + for<'de> Deserialize<'de>> { +pub struct Persister<T: Migrate> { path: PathBuf, _marker: std::marker::PhantomData<T>, } -impl<T> Persister<T> -where - T: Serialize + for<'de> Deserialize<'de>, -{ +impl<T: Migrate> Persister<T> { pub fn new(base_dir: &Path, file_name: &str) -> Self { let mut path = base_dir.to_path_buf(); path.push(file_name); @@ -27,18 +22,37 @@ where } } + fn decode(&self, bytes: &[u8]) -> Result<T, Error> { + match T::decode(bytes) { + Some(v) => Ok(v), + None => { + error!( + "Unable to decode persisted data file {}", + self.path.display() + ); + for line in hexdump::hexdump_iter(bytes) { + debug!("{}", line); + } + Err(Error::Message(format!( + "Unable to decode persisted data file {}", + self.path.display() + ))) + } + } + } + pub fn load(&self) -> Result<T, Error> { let mut file = std::fs::OpenOptions::new().read(true).open(&self.path)?; let mut bytes = vec![]; file.read_to_end(&mut bytes)?; - let value = rmp_serde::decode::from_read_ref(&bytes[..])?; + let value = self.decode(&bytes[..])?; Ok(value) } pub fn save(&self, t: &T) -> Result<(), Error> { - let bytes = rmp_to_vec_all_named(t)?; + let bytes = t.encode()?; let mut file = std::fs::OpenOptions::new() .write(true) @@ -57,12 +71,12 @@ where let mut bytes = vec![]; file.read_to_end(&mut bytes).await?; - let value = rmp_serde::decode::from_read_ref(&bytes[..])?; + let value = self.decode(&bytes[..])?; Ok(value) } pub async fn save_async(&self, t: &T) -> Result<(), Error> { - let bytes = rmp_to_vec_all_named(t)?; + let bytes = t.encode()?; let mut file = tokio::fs::File::create(&self.path).await?; file.write_all(&bytes[..]).await?; diff --git a/src/web/Cargo.toml b/src/web/Cargo.toml index 7bf70c55..dbc5e5fb 100644 --- a/src/web/Cargo.toml +++ b/src/web/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "garage_web" -version = "0.8.0" +version = "0.8.1" authors = ["Alex Auvolat <alex@adnab.me>", "Quentin Dufour <quentin@dufour.io>"] edition = "2018" license = "AGPL-3.0" @@ -14,10 +14,10 @@ path = "lib.rs" # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html [dependencies] -garage_api = { version = "0.8.0", path = "../api" } -garage_model = { version = "0.8.0", path = "../model" } -garage_util = { version = "0.8.0", path = "../util" } -garage_table = { version = "0.8.0", path = "../table" } +garage_api = { version = "0.8.1", path = "../api" } +garage_model = { version = "0.8.1", path = "../model" } +garage_util = { version = "0.8.1", path = "../util" } +garage_table = { version = "0.8.1", path = "../table" } err-derive = "0.3" tracing = "0.1.30" |