From 12012916b7eeb8e2d62db4d48a30a2b50d9d884d Mon Sep 17 00:00:00 2001
From: Alex Auvolat <lx@deuxfleurs.fr>
Date: Fri, 1 May 2026 20:51:49 +0200
Subject: [PATCH] simplify the `garage health` subcommand

---
 src/garage/cli/remote/cluster.rs | 50 ++++++++++++++------------------
 src/garage/cli/remote/mod.rs     |  7 ++---
 src/garage/cli/structs.rs        | 12 ++++----
 src/garage/main.rs               |  1 -
 4 files changed, 30 insertions(+), 40 deletions(-)

diff --git a/src/garage/cli/remote/cluster.rs b/src/garage/cli/remote/cluster.rs
index fe54a0e4..ff533289 100644
--- a/src/garage/cli/remote/cluster.rs
+++ b/src/garage/cli/remote/cluster.rs
@@ -9,37 +9,29 @@ use crate::cli::remote::*;
 use crate::cli::structs::*;
 
 impl Cli {
-	pub async fn cmd_health_check(&self, quiet: bool) -> Result<(), Error> {
-		let status = self.api_request(GetClusterStatusRequest).await?;
+	pub async fn cmd_health(&self, quiet: bool) -> Result<(), Error> {
+		let health = self.api_request(GetClusterHealthRequest).await?;
 
-		let mut result = Err(Error::Message(
-			"Failed to find node id in configuration".into(),
-		));
-
-		if let Some(config) = &self.config {
-			let local_node_id = hex::encode(
-					garage_rpc::system::read_node_id(&config.metadata_dir)
-						.err_context(crate::cli::local::init::READ_KEY_ERROR)?
-				).to_string();
-			result = Err(Error::Message("Cluster is not healthy".into()));
-
-			for node in status.nodes.iter() {
-				if node.id == local_node_id {
-					result = Ok(())
-				}
-			}
-			if !quiet {
-				match result {
-					Ok(_) => {
-						println!("Healthy");
-					}
-					Err(_) => {
-						println!("Not healthy");
-					}
-				}
-			}
+		if !quiet {
+			let table = vec![
+				format!("Cluster health:\t{}", health.status.to_uppercase()),
+				format!("Known nodes:\t{}", health.known_nodes),
+				format!("Connected nodes:\t{}", health.connected_nodes),
+				format!("Storage nodes:\t{}", health.storage_nodes),
+				format!("Storage nodes up:\t{}", health.storage_nodes_up),
+				format!("Partitions:\t{}", health.partitions),
+				format!("Partitions with quorum:\t{}", health.partitions_quorum),
+				format!("Fully healthy partitions:\t{}", health.partitions_all_ok),
+			];
+			format_table(table);
+		}
+
+		match health.status.as_str() {
+			"unavailable" => Err(Error::Message(
+				"Cluster is currently unavailable".to_string(),
+			)),
+			_ => Ok(()),
 		}
-		result
 	}
 
 	pub async fn cmd_status(&self) -> Result<(), Error> {
diff --git a/src/garage/cli/remote/mod.rs b/src/garage/cli/remote/mod.rs
index 56077c23..a30b5c1a 100644
--- a/src/garage/cli/remote/mod.rs
+++ b/src/garage/cli/remote/mod.rs
@@ -18,23 +18,22 @@ use garage_util::error::*;
 
 use garage_rpc::*;
 
-use crate::cli::structs::*;
 use garage_api_admin::api::*;
 use garage_api_admin::api_server::{AdminRpc as ProxyRpc, AdminRpcResponse as ProxyRpcResponse};
 use garage_api_admin::RequestHandler;
-use garage_util::config::Config;
+
+use crate::cli::structs::*;
 
 pub struct Cli {
 	pub proxy_rpc_endpoint: Arc<Endpoint<ProxyRpc, ()>>,
 	pub rpc_host: NodeID,
-	pub config: Option<Config>,
 }
 
 impl Cli {
 	pub async fn handle(&self, cmd: Command) -> Result<(), Error> {
 		match cmd {
 			Command::Status => self.cmd_status().await,
-			Command::HealthCheck(opt) => self.cmd_health_check(opt.quiet).await,
+			Command::Health(opt) => self.cmd_health(opt.quiet).await,
 			Command::Node(NodeOperation::Connect(connect_opt)) => {
 				self.cmd_connect(connect_opt).await
 			}
diff --git a/src/garage/cli/structs.rs b/src/garage/cli/structs.rs
index bde0cc13..368581cb 100644
--- a/src/garage/cli/structs.rs
+++ b/src/garage/cli/structs.rs
@@ -10,14 +10,14 @@ pub enum Command {
 	#[structopt(name = "server", version = garage_version())]
 	Server(ServerOpt),
 
+	/// Check the cluster health and set the exit code to 1 if it is unavailable
+	#[structopt(name = "health", version = garage_version())]
+	Health(HealthOpt),
+
 	/// Get network status
 	#[structopt(name = "status", version = garage_version())]
 	Status,
 
-	/// Check the local node health and set the exit code to 1 if it is unhealthy.
-	#[structopt(name = "health-check", version = garage_version())]
-	HealthCheck(HealthCheckOpt),
-
 	/// Operations on individual Garage nodes
 	#[structopt(name = "node", version = garage_version())]
 	Node(NodeOperation),
@@ -108,11 +108,11 @@ pub struct ServerOpt {
 }
 
 // -------------------------
-// ---- garage health-check ... ----
+// ---- garage health ----
 // -------------------------
 
 #[derive(StructOpt, Debug)]
-pub struct HealthCheckOpt {
+pub struct HealthOpt {
 	/// Do not print healthyness to stdout
 	#[structopt(short = "q", long = "quiet")]
 	pub(crate) quiet: bool,
diff --git a/src/garage/main.rs b/src/garage/main.rs
index f3f1422c..62d10c82 100644
--- a/src/garage/main.rs
+++ b/src/garage/main.rs
@@ -351,7 +351,6 @@ async fn cli_command(opt: Opt) -> Result<(), Error> {
 	let cli = cli::remote::Cli {
 		proxy_rpc_endpoint,
 		rpc_host: id,
-		config: config,
 	};
 
 	cli.handle(opt.cmd).await