fix(backend): Only remove dead nodes from the registry by default

dfinity · Jan 4, 2024 · 938d675 · 938d675
1 parent 44e0a7f
commit 938d675
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 2 deletions.
diff --git a/rs/cli/src/cli.rs b/rs/cli/src/cli.rs
@@ -339,6 +339,10 @@ pub(crate) mod nodes {
             #[clap(long)]
             no_auto: bool,
 
+            /// Remove also degraded nodes; by default only dead (offline) nodes are automatically removed
+            #[clap(long)]
+            remove_degraded: bool,
+
             /// Specifies the filter used to remove extra nodes
             extra_nodes_filter: Vec<String>,
 

diff --git a/rs/cli/src/main.rs b/rs/cli/src/main.rs
@@ -243,7 +243,7 @@ async fn main() -> Result<(), anyhow::Error> {
             },
             cli::Commands::Nodes(nodes) => {
                 match &nodes.subcommand {
-                    cli::nodes::Commands::Remove { extra_nodes_filter, no_auto, exclude, motivation } => {
+                    cli::nodes::Commands::Remove { extra_nodes_filter, no_auto, remove_degraded, exclude, motivation } => {
                         if motivation.is_none() && !extra_nodes_filter.is_empty() {
                             cmd.error(
                                 ErrorKind::MissingRequiredArgument,
@@ -255,6 +255,7 @@ async fn main() -> Result<(), anyhow::Error> {
                         runner.remove_nodes(NodesRemoveRequest {
                             extra_nodes_filter: extra_nodes_filter.clone(),
                             no_auto: *no_auto,
+                            remove_degraded: *remove_degraded,
                             exclude: Some(exclude.clone()),
                             motivation: motivation.clone().unwrap_or_default(),
                         }, simulate).await

diff --git a/rs/ic-management-backend/src/endpoints/nodes_ops.rs b/rs/ic-management-backend/src/endpoints/nodes_ops.rs
@@ -65,7 +65,13 @@ async fn remove(
                                     reason: NodeRemovalReason::Duplicates(principal),
                                 });
                             }
-                            if !matches!(status, ic_management_types::Status::Healthy) {
+                            let should_remove_node = if request.remove_degraded {
+                                matches!(status, ic_management_types::Status::Dead)
+                                    || matches!(status, ic_management_types::Status::Degraded)
+                            } else {
+                                matches!(status, ic_management_types::Status::Dead)
+                            };
+                            if should_remove_node {
                                 return Some(NodeRemoval {
                                     node: n,
                                     reason: NodeRemovalReason::Unhealthy(status),

diff --git a/rs/ic-management-types/src/requests.rs b/rs/ic-management-types/src/requests.rs
@@ -130,6 +130,7 @@ impl Display for HostosRolloutReason {
 #[derive(Serialize, Deserialize)]
 pub struct NodesRemoveRequest {
     pub no_auto: bool,
+    pub remove_degraded: bool,
     pub extra_nodes_filter: Vec<String>,
     pub exclude: Option<Vec<String>>,
     pub motivation: String,