GET _ilm/status returns STOPPED before it's safe to run POST _ilm/migrate_to_data_tiers #140885
Description
GET _ilm/status returns its response according to the responding node's last-applied cluster state, but POST _ilm/migrate_to_data_tiers requires the ILM status to be STOPPED in the last-applied cluster state on the elected master node.
Prior to #129367 these would usually be the same node, but ever since then it is more likely that the node which responds to GET _ilm/status might have applied a cluster state newer than the one on the elected master, responding with STOPPED before the point at which POST _ilm/migrate_to_data_tiers can be called.
It should be enough to move the STOPPED check inside the cluster state update task, but this needs some testing to verify that it works and I don't currently have capacity to add that testing.
diff --git a/x-pack/plugin/ilm/src/main/java/org/elasticsearch/xpack/ilm/action/TransportMigrateToDataTiersAction.java b/x-pack/plugin/ilm/src/main/java/org/elasticsearch/xpack/ilm/action/TransportMigrateToDataTiersAction.java
index 2ef2260891f0..0b95ecc28961 100644
--- a/x-pack/plugin/ilm/src/main/java/org/elasticsearch/xpack/ilm/action/TransportMigrateToDataTiersAction.java
+++ b/x-pack/plugin/ilm/src/main/java/org/elasticsearch/xpack/ilm/action/TransportMigrateToDataTiersAction.java
@@ -115,21 +115,18 @@ public class TransportMigrateToDataTiersAction extends TransportMasterNodeAction
return;
}
- ProjectMetadata projectMetadata = projectResolver.getProjectMetadata(state);
- IndexLifecycleMetadata currentMetadata = projectMetadata.custom(IndexLifecycleMetadata.TYPE);
- if (currentMetadata != null && currentILMMode(projectMetadata) != STOPPED) {
- listener.onFailure(
- new IllegalStateException(
- "stop ILM before migrating to data tiers, current state is [" + currentILMMode(projectMetadata) + "]"
- )
- );
- return;
- }
-
final SetOnce<MigratedEntities> migratedEntities = new SetOnce<>();
submitUnbatchedTask("migrate-to-data-tiers []", new ClusterStateUpdateTask(Priority.HIGH) {
@Override
public ClusterState execute(ClusterState currentState) {
+ ProjectMetadata projectMetadata = projectResolver.getProjectMetadata(currentState);
+ IndexLifecycleMetadata currentMetadata = projectMetadata.custom(IndexLifecycleMetadata.TYPE);
+ if (currentMetadata != null && currentILMMode(projectMetadata) != STOPPED) {
+ throw new IllegalStateException(
+ "stop ILM before migrating to data tiers, current state is [" + currentILMMode(projectMetadata) + "]"
+ );
+ }
+
Tuple<ClusterState, MigratedEntities> migratedEntitiesTuple = migrateToDataTiersRouting(
currentState.projectState(projectId),
request.getNodeAttributeName(),