5 files changed, 345 insertions, 138 deletions

diff --git a/src/garage/cli/cmd.rs b/src/garage/cli/cmd.rs
index 4c0a5322..a84061a7 100644
--- a/src/garage/cli/cmd.rs
+++ b/src/garage/cli/cmd.rs
@@ -1,4 +1,4 @@
-use std::collections::HashSet;
+use std::collections::{HashMap, HashSet};
 use std::time::Duration;
 
 use format_table::format_table;
@@ -33,9 +33,6 @@ pub async fn cli_command_dispatch(
 		Command::Key(ko) => {
 			cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::KeyOperation(ko)).await
 		}
-		Command::Migrate(mo) => {
-			cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::Migrate(mo)).await
-		}
 		Command::Repair(ro) => {
 			cmd_admin(admin_rpc_endpoint, rpc_host, AdminRpc::LaunchRepair(ro)).await
 		}
@@ -52,50 +49,61 @@ pub async fn cli_command_dispatch(
 }
 
 pub async fn cmd_status(rpc_cli: &Endpoint<SystemRpc, ()>, rpc_host: NodeID) -> Result<(), Error> {
-	let status = match rpc_cli
-		.call(&rpc_host, SystemRpc::GetKnownNodes, PRIO_NORMAL)
-		.await??
-	{
-		SystemRpc::ReturnKnownNodes(nodes) => nodes,
-		resp => return Err(Error::Message(format!("Invalid RPC response: {:?}", resp))),
-	};
+	let status = fetch_status(rpc_cli, rpc_host).await?;
 	let layout = fetch_layout(rpc_cli, rpc_host).await?;
 
 	println!("==== HEALTHY NODES ====");
 	let mut healthy_nodes =
 		vec!["ID\tHostname\tAddress\tTags\tZone\tCapacity\tDataAvail".to_string()];
 	for adv in status.iter().filter(|adv| adv.is_up) {
-		match layout.roles.get(&adv.id) {
-			Some(NodeRoleV(Some(cfg))) => {
-				let data_avail = match &adv.status.data_disk_avail {
-					_ if cfg.capacity.is_none() => "N/A".into(),
-					Some((avail, total)) => {
-						let pct = (*avail as f64) / (*total as f64) * 100.;
-						let avail = bytesize::ByteSize::b(*avail);
-						format!("{} ({:.1}%)", avail, pct)
-					}
-					None => "?".into(),
-				};
+		let host = adv.status.hostname.as_deref().unwrap_or("?");
+		if let Some(NodeRoleV(Some(cfg))) = layout.current().roles.get(&adv.id) {
+			let data_avail = match &adv.status.data_disk_avail {
+				_ if cfg.capacity.is_none() => "N/A".into(),
+				Some((avail, total)) => {
+					let pct = (*avail as f64) / (*total as f64) * 100.;
+					let avail = bytesize::ByteSize::b(*avail);
+					format!("{} ({:.1}%)", avail, pct)
+				}
+				None => "?".into(),
+			};
+			healthy_nodes.push(format!(
+				"{id:?}\t{host}\t{addr}\t[{tags}]\t{zone}\t{capacity}\t{data_avail}",
+				id = adv.id,
+				host = host,
+				addr = adv.addr,
+				tags = cfg.tags.join(","),
+				zone = cfg.zone,
+				capacity = cfg.capacity_string(),
+				data_avail = data_avail,
+			));
+		} else {
+			let prev_role = layout
+				.versions
+				.iter()
+				.rev()
+				.find_map(|x| match x.roles.get(&adv.id) {
+					Some(NodeRoleV(Some(cfg))) => Some(cfg),
+					_ => None,
+				});
+			if let Some(cfg) = prev_role {
 				healthy_nodes.push(format!(
-					"{id:?}\t{host}\t{addr}\t[{tags}]\t{zone}\t{capacity}\t{data_avail}",
+					"{id:?}\t{host}\t{addr}\t[{tags}]\t{zone}\tdraining metadata...",
 					id = adv.id,
-					host = adv.status.hostname,
+					host = host,
 					addr = adv.addr,
 					tags = cfg.tags.join(","),
 					zone = cfg.zone,
-					capacity = cfg.capacity_string(),
-					data_avail = data_avail,
 				));
-			}
-			_ => {
-				let new_role = match layout.staging_roles.get(&adv.id) {
-					Some(NodeRoleV(Some(_))) => "(pending)",
+			} else {
+				let new_role = match layout.staging.get().roles.get(&adv.id) {
+					Some(NodeRoleV(Some(_))) => "pending...",
 					_ => "NO ROLE ASSIGNED",
 				};
 				healthy_nodes.push(format!(
-					"{id:?}\t{h}\t{addr}\t{new_role}",
+					"{id:?}\t{h}\t{addr}\t\t\t{new_role}",
 					id = adv.id,
-					h = adv.status.hostname,
+					h = host,
 					addr = adv.addr,
 					new_role = new_role,
 				));
@@ -104,51 +112,76 @@ pub async fn cmd_status(rpc_cli: &Endpoint<SystemRpc, ()>, rpc_host: NodeID) ->
 	}
 	format_table(healthy_nodes);
 
-	let status_keys = status.iter().map(|adv| adv.id).collect::<HashSet<_>>();
-	let failure_case_1 = status
-		.iter()
-		.any(|adv| !adv.is_up && matches!(layout.roles.get(&adv.id), Some(NodeRoleV(Some(_)))));
-	let failure_case_2 = layout
-		.roles
-		.items()
+	// Determine which nodes are unhealthy and print that to stdout
+	let status_map = status
 		.iter()
-		.any(|(id, _, v)| !status_keys.contains(id) && v.0.is_some());
-	if failure_case_1 || failure_case_2 {
-		println!("\n==== FAILED NODES ====");
-		let mut failed_nodes =
-			vec!["ID\tHostname\tAddress\tTags\tZone\tCapacity\tLast seen".to_string()];
-		for adv in status.iter().filter(|adv| !adv.is_up) {
-			if let Some(NodeRoleV(Some(cfg))) = layout.roles.get(&adv.id) {
-				let tf = timeago::Formatter::new();
-				failed_nodes.push(format!(
-					"{id:?}\t{host}\t{addr}\t[{tags}]\t{zone}\t{capacity}\t{last_seen}",
-					id = adv.id,
-					host = adv.status.hostname,
-					addr = adv.addr,
-					tags = cfg.tags.join(","),
-					zone = cfg.zone,
-					capacity = cfg.capacity_string(),
-					last_seen = adv
-						.last_seen_secs_ago
-						.map(|s| tf.convert(Duration::from_secs(s)))
-						.unwrap_or_else(|| "never seen".into()),
-				));
+		.map(|adv| (adv.id, adv))
+		.collect::<HashMap<_, _>>();
+
+	let tf = timeago::Formatter::new();
+	let mut drain_msg = false;
+	let mut failed_nodes =
+		vec!["ID\tHostname\tAddress\tTags\tZone\tCapacity\tLast seen".to_string()];
+	let mut listed = HashSet::new();
+	for ver in layout.versions.iter().rev() {
+		for (node, _, role) in ver.roles.items().iter() {
+			let cfg = match role {
+				NodeRoleV(Some(role)) if role.capacity.is_some() => role,
+				_ => continue,
+			};
+
+			if listed.contains(node) {
+				continue;
 			}
-		}
-		for (id, _, role_v) in layout.roles.items().iter() {
-			if let NodeRoleV(Some(cfg)) = role_v {
-				if !status_keys.contains(id) {
-					failed_nodes.push(format!(
-						"{id:?}\t??\t??\t[{tags}]\t{zone}\t{capacity}\tnever seen",
-						id = id,
-						tags = cfg.tags.join(","),
-						zone = cfg.zone,
-						capacity = cfg.capacity_string(),
-					));
-				}
+			listed.insert(*node);
+
+			let adv = status_map.get(node);
+			if adv.map(|x| x.is_up).unwrap_or(false) {
+				continue;
 			}
+
+			// Node is in a layout version, is not a gateway node, and is not up:
+			// it is in a failed state, add proper line to the output
+			let (host, addr, last_seen) = match adv {
+				Some(adv) => (
+					adv.status.hostname.as_deref().unwrap_or("?"),
+					adv.addr.to_string(),
+					adv.last_seen_secs_ago
+						.map(|s| tf.convert(Duration::from_secs(s)))
+						.unwrap_or_else(|| "never seen".into()),
+				),
+				None => ("??", "??".into(), "never seen".into()),
+			};
+			let capacity = if ver.version == layout.current().version {
+				cfg.capacity_string()
+			} else {
+				drain_msg = true;
+				"draining metadata...".to_string()
+			};
+			failed_nodes.push(format!(
+				"{id:?}\t{host}\t{addr}\t[{tags}]\t{zone}\t{capacity}\t{last_seen}",
+				id = node,
+				host = host,
+				addr = addr,
+				tags = cfg.tags.join(","),
+				zone = cfg.zone,
+				capacity = capacity,
+				last_seen = last_seen,
+			));
 		}
+	}
+
+	if failed_nodes.len() > 1 {
+		println!("\n==== FAILED NODES ====");
 		format_table(failed_nodes);
+		if drain_msg {
+			println!();
+			println!("Your cluster is expecting to drain data from nodes that are currently unavailable.");
+			println!("If these nodes are definitely dead, please review the layout history with");
+			println!(
+				"`garage layout history` and use `garage layout skip-dead-nodes` to force progress."
+			);
+		}
 	}
 
 	if print_staging_role_changes(&layout) {
@@ -229,3 +262,18 @@ pub async fn cmd_admin(
 	}
 	Ok(())
 }
+
+// ---- utility ----
+
+pub async fn fetch_status(
+	rpc_cli: &Endpoint<SystemRpc, ()>,
+	rpc_host: NodeID,
+) -> Result<Vec<KnownNodeInfo>, Error> {
+	match rpc_cli
+		.call(&rpc_host, SystemRpc::GetKnownNodes, PRIO_NORMAL)
+		.await??
+	{
+		SystemRpc::ReturnKnownNodes(nodes) => Ok(nodes),
+		resp => Err(Error::unexpected_rpc_message(resp)),
+	}
+}
diff --git a/src/garage/cli/convert_db.rs b/src/garage/cli/convert_db.rs
index 2aadb1d6..5346d55a 100644
--- a/src/garage/cli/convert_db.rs
+++ b/src/garage/cli/convert_db.rs
@@ -11,7 +11,7 @@ pub struct ConvertDbOpt {
 	/// https://garagehq.deuxfleurs.fr/documentation/reference-manual/configuration/#db-engine-since-v0-8-0)
 	#[structopt(short = "i")]
 	input_path: PathBuf,
-	/// Input database engine (sled, lmdb or sqlite; limited by db engines
+	/// Input database engine (lmdb or sqlite; limited by db engines
 	/// enabled in this build)
 	#[structopt(short = "a")]
 	input_engine: Engine,
diff --git a/src/garage/cli/layout.rs b/src/garage/cli/layout.rs
index ce2b11e0..f76e33c5 100644
--- a/src/garage/cli/layout.rs
+++ b/src/garage/cli/layout.rs
@@ -32,6 +32,10 @@ pub async fn cli_layout_command_dispatch(
 		LayoutOperation::Config(config_opt) => {
 			cmd_config_layout(system_rpc_endpoint, rpc_host, config_opt).await
 		}
+		LayoutOperation::History => cmd_layout_history(system_rpc_endpoint, rpc_host).await,
+		LayoutOperation::SkipDeadNodes(assume_sync_opt) => {
+			cmd_layout_skip_dead_nodes(system_rpc_endpoint, rpc_host, assume_sync_opt).await
+		}
 	}
 }
 
@@ -49,6 +53,7 @@ pub async fn cmd_assign_role(
 	};
 
 	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;
+	let all_nodes = layout.get_all_nodes();
 
 	let added_nodes = args
 		.node_ids
@@ -58,21 +63,23 @@ pub async fn cmd_assign_role(
 				status
 					.iter()
 					.map(|adv| adv.id)
-					.chain(layout.node_ids().iter().cloned()),
+					.chain(all_nodes.iter().cloned()),
 				node_id,
 			)
 		})
 		.collect::<Result<Vec<_>, _>>()?;
 
-	let mut roles = layout.roles.clone();
-	roles.merge(&layout.staging_roles);
+	let mut roles = layout.current().roles.clone();
+	roles.merge(&layout.staging.get().roles);
 
 	for replaced in args.replace.iter() {
-		let replaced_node = find_matching_node(layout.node_ids().iter().cloned(), replaced)?;
+		let replaced_node = find_matching_node(all_nodes.iter().cloned(), replaced)?;
 		match roles.get(&replaced_node) {
 			Some(NodeRoleV(Some(_))) => {
 				layout
-					.staging_roles
+					.staging
+					.get_mut()
+					.roles
 					.merge(&roles.update_mutator(replaced_node, NodeRoleV(None)));
 			}
 			_ => {
@@ -130,7 +137,9 @@ pub async fn cmd_assign_role(
 		};
 
 		layout
-			.staging_roles
+			.staging
+			.get_mut()
+			.roles
 			.merge(&roles.update_mutator(added_node, NodeRoleV(Some(new_entry))));
 	}
 
@@ -149,14 +158,16 @@ pub async fn cmd_remove_role(
 ) -> Result<(), Error> {
 	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;
 
-	let mut roles = layout.roles.clone();
-	roles.merge(&layout.staging_roles);
+	let mut roles = layout.current().roles.clone();
+	roles.merge(&layout.staging.get().roles);
 
 	let deleted_node =
 		find_matching_node(roles.items().iter().map(|(id, _, _)| *id), &args.node_id)?;
 
 	layout
-		.staging_roles
+		.staging
+		.get_mut()
+		.roles
 		.merge(&roles.update_mutator(deleted_node, NodeRoleV(None)));
 
 	send_layout(rpc_cli, rpc_host, layout).await?;
@@ -174,13 +185,16 @@ pub async fn cmd_show_layout(
 	let layout = fetch_layout(rpc_cli, rpc_host).await?;
 
 	println!("==== CURRENT CLUSTER LAYOUT ====");
-	print_cluster_layout(&layout, "No nodes currently have a role in the cluster.\nSee `garage status` to view available nodes.");
+	print_cluster_layout(layout.current(), "No nodes currently have a role in the cluster.\nSee `garage status` to view available nodes.");
 	println!();
-	println!("Current cluster layout version: {}", layout.version);
+	println!(
+		"Current cluster layout version: {}",
+		layout.current().version
+	);
 
 	let has_role_changes = print_staging_role_changes(&layout);
 	if has_role_changes {
-		let v = layout.version;
+		let v = layout.current().version;
 		let res_apply = layout.apply_staged_changes(Some(v + 1));
 
 		// this will print the stats of what partitions
@@ -189,7 +203,7 @@ pub async fn cmd_show_layout(
 			Ok((layout, msg)) => {
 				println!();
 				println!("==== NEW CLUSTER LAYOUT AFTER APPLYING CHANGES ====");
-				print_cluster_layout(&layout, "No nodes have a role in the new layout.");
+				print_cluster_layout(layout.current(), "No nodes have a role in the new layout.");
 				println!();
 
 				for line in msg.iter() {
@@ -199,16 +213,12 @@ pub async fn cmd_show_layout(
 				println!();
 				println!("    garage layout apply --version {}", v + 1);
 				println!();
-				println!(
-                    "You can also revert all proposed changes with: garage layout revert --version {}",
-                    v + 1)
+				println!("You can also revert all proposed changes with: garage layout revert");
 			}
 			Err(e) => {
 				println!("Error while trying to compute the assignment: {}", e);
 				println!("This new layout cannot yet be applied.");
-				println!(
-                    "You can also revert all proposed changes with: garage layout revert --version {}",
-                    v + 1)
+				println!("You can also revert all proposed changes with: garage layout revert");
 			}
 		}
 	}
@@ -241,9 +251,15 @@ pub async fn cmd_revert_layout(
 	rpc_host: NodeID,
 	revert_opt: RevertLayoutOpt,
 ) -> Result<(), Error> {
+	if !revert_opt.yes {
+		return Err(Error::Message(
+			"Please add the --yes flag to run the layout revert operation".into(),
+		));
+	}
+
 	let layout = fetch_layout(rpc_cli, rpc_host).await?;
 
-	let layout = layout.revert_staged_changes(revert_opt.version)?;
+	let layout = layout.revert_staged_changes()?;
 
 	send_layout(rpc_cli, rpc_host, layout).await?;
 
@@ -266,11 +282,11 @@ pub async fn cmd_config_layout(
 				.parse::<ZoneRedundancy>()
 				.ok_or_message("invalid zone redundancy value")?;
 			if let ZoneRedundancy::AtLeast(r_int) = r {
-				if r_int > layout.replication_factor {
+				if r_int > layout.current().replication_factor {
 					return Err(Error::Message(format!(
 						"The zone redundancy must be smaller or equal to the \
                     replication factor ({}).",
-						layout.replication_factor
+						layout.current().replication_factor
 					)));
 				} else if r_int < 1 {
 					return Err(Error::Message(
@@ -280,7 +296,9 @@ pub async fn cmd_config_layout(
 			}
 
 			layout
-				.staging_parameters
+				.staging
+				.get_mut()
+				.parameters
 				.update(LayoutParameters { zone_redundancy: r });
 			println!("The zone redundancy parameter has been set to '{}'.", r);
 			did_something = true;
@@ -297,25 +315,166 @@ pub async fn cmd_config_layout(
 	Ok(())
 }
 
+pub async fn cmd_layout_history(
+	rpc_cli: &Endpoint<SystemRpc, ()>,
+	rpc_host: NodeID,
+) -> Result<(), Error> {
+	let layout = fetch_layout(rpc_cli, rpc_host).await?;
+	let min_stored = layout.min_stored();
+
+	println!("==== LAYOUT HISTORY ====");
+	let mut table = vec!["Version\tStatus\tStorage nodes\tGateway nodes".to_string()];
+	for ver in layout
+		.versions
+		.iter()
+		.rev()
+		.chain(layout.old_versions.iter().rev())
+	{
+		let status = if ver.version == layout.current().version {
+			"current"
+		} else if ver.version >= min_stored {
+			"draining"
+		} else {
+			"historical"
+		};
+		table.push(format!(
+			"#{}\t{}\t{}\t{}",
+			ver.version,
+			status,
+			ver.roles
+				.items()
+				.iter()
+				.filter(|(_, _, x)| matches!(x, NodeRoleV(Some(c)) if c.capacity.is_some()))
+				.count(),
+			ver.roles
+				.items()
+				.iter()
+				.filter(|(_, _, x)| matches!(x, NodeRoleV(Some(c)) if c.capacity.is_none()))
+				.count(),
+		));
+	}
+	format_table(table);
+	println!();
+
+	if layout.versions.len() > 1 {
+		println!("==== UPDATE TRACKERS ====");
+		println!("Several layout versions are currently live in the version, and data is being migrated.");
+		println!(
+			"This is the internal data that Garage stores to know which nodes have what data."
+		);
+		println!();
+		let mut table = vec!["Node\tAck\tSync\tSync_ack".to_string()];
+		let all_nodes = layout.get_all_nodes();
+		for node in all_nodes.iter() {
+			table.push(format!(
+				"{:?}\t#{}\t#{}\t#{}",
+				node,
+				layout.update_trackers.ack_map.get(node, min_stored),
+				layout.update_trackers.sync_map.get(node, min_stored),
+				layout.update_trackers.sync_ack_map.get(node, min_stored),
+			));
+		}
+		table[1..].sort();
+		format_table(table);
+
+		println!();
+		println!(
+			"If some nodes are not catching up to the latest layout version in the update trackers,"
+		);
+		println!("it might be because they are offline or unable to complete a sync successfully.");
+		println!(
+			"You may force progress using `garage layout skip-dead-nodes --version {}`",
+			layout.current().version
+		);
+	} else {
+		println!("Your cluster is currently in a stable state with a single live layout version.");
+		println!("No metadata migration is in progress. Note that the migration of data blocks is not tracked,");
+		println!(
+			"so you might want to keep old nodes online until their data directories become empty."
+		);
+	}
+
+	Ok(())
+}
+
+pub async fn cmd_layout_skip_dead_nodes(
+	rpc_cli: &Endpoint<SystemRpc, ()>,
+	rpc_host: NodeID,
+	opt: SkipDeadNodesOpt,
+) -> Result<(), Error> {
+	let status = fetch_status(rpc_cli, rpc_host).await?;
+	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;
+
+	if layout.versions.len() == 1 {
+		return Err(Error::Message(
+			"This command cannot be called when there is only one live cluster layout version"
+				.into(),
+		));
+	}
+
+	let min_v = layout.min_stored();
+	if opt.version <= min_v || opt.version > layout.current().version {
+		return Err(Error::Message(format!(
+			"Invalid version, you may use the following version numbers: {}",
+			(min_v + 1..=layout.current().version)
+				.map(|x| x.to_string())
+				.collect::<Vec<_>>()
+				.join(" ")
+		)));
+	}
+
+	let all_nodes = layout.get_all_nodes();
+	let mut did_something = false;
+	for node in all_nodes.iter() {
+		if status.iter().any(|x| x.id == *node && x.is_up) {
+			continue;
+		}
+
+		if layout.update_trackers.ack_map.set_max(*node, opt.version) {
+			println!("Increased the ACK tracker for node {:?}", node);
+			did_something = true;
+		}
+
+		if opt.allow_missing_data {
+			if layout.update_trackers.sync_map.set_max(*node, opt.version) {
+				println!("Increased the SYNC tracker for node {:?}", node);
+				did_something = true;
+			}
+		}
+	}
+
+	if did_something {
+		send_layout(rpc_cli, rpc_host, layout).await?;
+		println!("Success.");
+		Ok(())
+	} else if !opt.allow_missing_data {
+		Err(Error::Message("Nothing was done, try passing the `--allow-missing-data` flag to force progress even when not enough nodes can complete a metadata sync.".into()))
+	} else {
+		Err(Error::Message(
+			"Sorry, there is nothing I can do for you. Please wait patiently. If you ask for help, please send the output of the `garage layout history` command.".into(),
+		))
+	}
+}
+
 // --- utility ---
 
 pub async fn fetch_layout(
 	rpc_cli: &Endpoint<SystemRpc, ()>,
 	rpc_host: NodeID,
-) -> Result<ClusterLayout, Error> {
+) -> Result<LayoutHistory, Error> {
 	match rpc_cli
 		.call(&rpc_host, SystemRpc::PullClusterLayout, PRIO_NORMAL)
 		.await??
 	{
 		SystemRpc::AdvertiseClusterLayout(t) => Ok(t),
-		resp => Err(Error::Message(format!("Invalid RPC response: {:?}", resp))),
+		resp => Err(Error::unexpected_rpc_message(resp)),
 	}
 }
 
 pub async fn send_layout(
 	rpc_cli: &Endpoint<SystemRpc, ()>,
 	rpc_host: NodeID,
-	layout: ClusterLayout,
+	layout: LayoutHistory,
 ) -> Result<(), Error> {
 	rpc_cli
 		.call(
@@ -327,7 +486,7 @@ pub async fn send_layout(
 	Ok(())
 }
 
-pub fn print_cluster_layout(layout: &ClusterLayout, empty_msg: &str) {
+pub fn print_cluster_layout(layout: &LayoutVersion, empty_msg: &str) {
 	let mut table = vec!["ID\tTags\tZone\tCapacity\tUsable capacity".to_string()];
 	for (id, _, role) in layout.roles.items().iter() {
 		let role = match &role.0 {
@@ -366,21 +525,22 @@ pub fn print_cluster_layout(layout: &ClusterLayout, empty_msg: &str) {
 	}
 }
 
-pub fn print_staging_role_changes(layout: &ClusterLayout) -> bool {
-	let has_role_changes = layout
-		.staging_roles
+pub fn print_staging_role_changes(layout: &LayoutHistory) -> bool {
+	let staging = layout.staging.get();
+	let has_role_changes = staging
+		.roles
 		.items()
 		.iter()
-		.any(|(k, _, v)| layout.roles.get(k) != Some(v));
-	let has_layout_changes = *layout.staging_parameters.get() != layout.parameters;
+		.any(|(k, _, v)| layout.current().roles.get(k) != Some(v));
+	let has_layout_changes = *staging.parameters.get() != layout.current().parameters;
 
 	if has_role_changes || has_layout_changes {
 		println!();
 		println!("==== STAGED ROLE CHANGES ====");
 		if has_role_changes {
 			let mut table = vec!["ID\tTags\tZone\tCapacity".to_string()];
-			for (id, _, role) in layout.staging_roles.items().iter() {
-				if layout.roles.get(id) == Some(role) {
+			for (id, _, role) in staging.roles.items().iter() {
+				if layout.current().roles.get(id) == Some(role) {
 					continue;
 				}
 				if let Some(role) = &role.0 {
@@ -402,7 +562,7 @@ pub fn print_staging_role_changes(layout: &ClusterLayout) -> bool {
 		if has_layout_changes {
 			println!(
 				"Zone redundancy: {}",
-				layout.staging_parameters.get().zone_redundancy
+				staging.parameters.get().zone_redundancy
 			);
 		}
 		true
diff --git a/src/garage/cli/structs.rs b/src/garage/cli/structs.rs
index 51d2bed3..8380b5e2 100644
--- a/src/garage/cli/structs.rs
+++ b/src/garage/cli/structs.rs
@@ -31,11 +31,6 @@ pub enum Command {
 	#[structopt(name = "key", version = garage_version())]
 	Key(KeyOperation),
 
-	/// Run migrations from previous Garage version
-	/// (DO NOT USE WITHOUT READING FULL DOCUMENTATION)
-	#[structopt(name = "migrate", version = garage_version())]
-	Migrate(MigrateOpt),
-
 	/// Start repair of node data on remote node
 	#[structopt(name = "repair", version = garage_version())]
 	Repair(RepairOpt),
@@ -118,6 +113,14 @@ pub enum LayoutOperation {
 	/// Revert staged changes to cluster layout
 	#[structopt(name = "revert", version = garage_version())]
 	Revert(RevertLayoutOpt),
+
+	/// View the history of layouts in the cluster
+	#[structopt(name = "history", version = garage_version())]
+	History,
+
+	/// Skip dead nodes when awaiting for a new layout version to be synchronized
+	#[structopt(name = "skip-dead-nodes", version = garage_version())]
+	SkipDeadNodes(SkipDeadNodesOpt),
 }
 
 #[derive(StructOpt, Debug)]
@@ -170,9 +173,21 @@ pub struct ApplyLayoutOpt {
 
 #[derive(StructOpt, Debug)]
 pub struct RevertLayoutOpt {
-	/// Version number of old configuration to which to revert
+	/// The revert operation will not be ran unless this flag is added
+	#[structopt(long = "yes")]
+	pub(crate) yes: bool,
+}
+
+#[derive(StructOpt, Debug)]
+pub struct SkipDeadNodesOpt {
+	/// Version number of the layout to assume is currently up-to-date.
+	/// This will generally be the current layout version.
 	#[structopt(long = "version")]
-	pub(crate) version: Option<u64>,
+	pub(crate) version: u64,
+	/// Allow the skip even if a quorum of ndoes could not be found for
+	/// the data among the remaining nodes
+	#[structopt(long = "allow-missing-data")]
+	pub(crate) allow_missing_data: bool,
 }
 
 #[derive(Serialize, Deserialize, StructOpt, Debug)]
@@ -430,23 +445,6 @@ pub struct KeyImportOpt {
 }
 
 #[derive(Serialize, Deserialize, StructOpt, Debug, Clone)]
-pub struct MigrateOpt {
-	/// Confirm the launch of the migrate operation
-	#[structopt(long = "yes")]
-	pub yes: bool,
-
-	#[structopt(subcommand)]
-	pub what: MigrateWhat,
-}
-
-#[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
-pub enum MigrateWhat {
-	/// Migrate buckets and permissions from v0.5.0
-	#[structopt(name = "buckets050", version = garage_version())]
-	Buckets050,
-}
-
-#[derive(Serialize, Deserialize, StructOpt, Debug, Clone)]
 pub struct RepairOpt {
 	/// Launch repair operation on all nodes
 	#[structopt(short = "a", long = "all-nodes")]
@@ -475,8 +473,11 @@ pub enum RepairWhat {
 	#[structopt(name = "mpu", version = garage_version())]
 	MultipartUploads,
 	/// Repropagate version deletions to the block ref table
-	#[structopt(name = "block_refs", version = garage_version())]
+	#[structopt(name = "block-refs", version = garage_version())]
 	BlockRefs,
+	/// Recalculate block reference counters
+	#[structopt(name = "block-rc", version = garage_version())]
+	BlockRc,
 	/// Verify integrity of all blocks on disc
 	#[structopt(name = "scrub", version = garage_version())]
 	Scrub {
@@ -537,10 +538,6 @@ pub struct StatsOpt {
 	#[structopt(short = "a", long = "all-nodes")]
 	pub all_nodes: bool,
 
-	/// Gather detailed statistics (this can be long)
-	#[structopt(short = "d", long = "detailed")]
-	pub detailed: bool,
-
 	/// Don't show global cluster stats (internal use in RPC)
 	#[structopt(skip)]
 	#[serde(default)]
diff --git a/src/garage/cli/util.rs b/src/garage/cli/util.rs
index 2232d395..21c14f42 100644
--- a/src/garage/cli/util.rs
+++ b/src/garage/cli/util.rs
@@ -450,6 +450,8 @@ pub fn print_block_info(
 
 	if refcount != nondeleted_count {
 		println!();
-		println!("Warning: refcount does not match number of non-deleted versions");
+		println!(
+			"Warning: refcount does not match number of non-deleted versions, you should try `garage repair block-rc`."
+		);
 	}
 }