From aa59059a910eb6e1e824b84413a66909d697ef8a Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Thu, 7 Dec 2023 11:50:00 +0100 Subject: [PATCH] layout cli: safer skip-dead-nodes command --- src/garage/cli/cmd.rs | 23 ++++++++++++++++------- src/garage/cli/layout.rs | 35 +++++++++++++++++++++++++---------- src/garage/cli/structs.rs | 12 ++++++++---- 3 files changed, 49 insertions(+), 21 deletions(-) diff --git a/src/garage/cli/cmd.rs b/src/garage/cli/cmd.rs index c7f0ad2b..196c0cb3 100644 --- a/src/garage/cli/cmd.rs +++ b/src/garage/cli/cmd.rs @@ -49,13 +49,7 @@ pub async fn cli_command_dispatch( } pub async fn cmd_status(rpc_cli: &Endpoint, 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 ===="); @@ -268,3 +262,18 @@ pub async fn cmd_admin( } Ok(()) } + +// ---- utility ---- + +pub async fn fetch_status( + rpc_cli: &Endpoint, + rpc_host: NodeID, +) -> Result, Error> { + match rpc_cli + .call(&rpc_host, SystemRpc::GetKnownNodes, PRIO_NORMAL) + .await?? + { + SystemRpc::ReturnKnownNodes(nodes) => Ok(nodes), + resp => Err(Error::Message(format!("Invalid RPC response: {:?}", resp))), + } +} diff --git a/src/garage/cli/layout.rs b/src/garage/cli/layout.rs index 3c7843bd..cdf77c04 100644 --- a/src/garage/cli/layout.rs +++ b/src/garage/cli/layout.rs @@ -33,8 +33,8 @@ pub async fn cli_layout_command_dispatch( cmd_config_layout(system_rpc_endpoint, rpc_host, config_opt).await } LayoutOperation::History => cmd_layout_history(system_rpc_endpoint, rpc_host).await, - LayoutOperation::AssumeSync(assume_sync_opt) => { - cmd_layout_assume_sync(system_rpc_endpoint, rpc_host, assume_sync_opt).await + LayoutOperation::SkipDeadNodes(assume_sync_opt) => { + cmd_layout_skip_dead_nodes(system_rpc_endpoint, rpc_host, assume_sync_opt).await } } } @@ -388,13 +388,21 @@ pub async fn cmd_layout_history( Ok(()) } -pub async fn cmd_layout_assume_sync( +pub async fn cmd_layout_skip_dead_nodes( rpc_cli: &Endpoint, rpc_host: NodeID, - opt: AssumeSyncOpt, + 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!( @@ -408,12 +416,19 @@ pub async fn cmd_layout_assume_sync( let all_nodes = layout.get_all_nodes(); for node in all_nodes.iter() { - layout.update_trackers.ack_map.set_max(*node, opt.version); - layout.update_trackers.sync_map.set_max(*node, opt.version); - layout - .update_trackers - .sync_ack_map - .set_max(*node, opt.version); + 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); + } + + if opt.allow_missing_data { + if layout.update_trackers.sync_map.set_max(*node, opt.version) { + println!("Increased the SYNC tracker for node {:?}", node); + } + } } send_layout(rpc_cli, rpc_host, layout).await?; diff --git a/src/garage/cli/structs.rs b/src/garage/cli/structs.rs index c4b400f4..6bc3da22 100644 --- a/src/garage/cli/structs.rs +++ b/src/garage/cli/structs.rs @@ -117,9 +117,9 @@ pub enum LayoutOperation { #[structopt(name = "history", version = garage_version())] History, - /// Assume all nodes are synchronized up to a certain layout version - #[structopt(name = "assume-sync", version = garage_version())] - AssumeSync(AssumeSyncOpt), + /// 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)] @@ -178,11 +178,15 @@ pub struct RevertLayoutOpt { } #[derive(StructOpt, Debug)] -pub struct AssumeSyncOpt { +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: 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)]