From 687660b27f904422c689e09d2457293e5313d325 Mon Sep 17 00:00:00 2001
From: Alex Auvolat <alex@adnab.me>
Date: Tue, 13 Dec 2022 14:23:45 +0100
Subject: Implement `block list-errors` and `block info`

---
 src/garage/cli/cmd.rs     | 13 +++++++++++
 src/garage/cli/structs.rs | 45 +++++++++++++++++++++++++++++-------
 src/garage/cli/util.rs    | 59 ++++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 108 insertions(+), 9 deletions(-)

(limited to 'src/garage/cli')

diff --git a/src/garage/cli/cmd.rs b/src/garage/cli/cmd.rs
index 6df15a48..6c5598b1 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!(),
 	}
 }
@@ -191,6 +194,16 @@ pub async fn cmd_admin(
 		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 9334564b..6d74b1a4 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)]
@@ -502,14 +506,8 @@ 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 {
@@ -549,3 +547,34 @@ pub enum WorkerSetCmd {
 	#[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
+		blocks: Vec<String>,
+	},
+}
diff --git a/src/garage/cli/util.rs b/src/garage/cli/util.rs
index c1d03b8d..737b54b2 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;
 
@@ -353,3 +356,57 @@ pub fn print_worker_info(tid: usize, info: WorkerInfo) {
 	}
 	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\tPath\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");
+	}
+}
-- 
cgit v1.2.3