Allow RerouteService to reroute at lower priority #44338
Conversation
Today the `BatchedRerouteService` submits its delayed reroute task at `HIGH` priority, but in some cases a lower priority would be more appropriate. This commit adds the facility to submit delayed reroute tasks at different priorities, such that each submitted reroute task runs at a priority no lower than the one requested. It does not change the fact that all delayed reroute tasks are submitted at `HIGH` priority, but at least it makes this explicit.
DaveCTurner
added
>enhancement
:Distributed/Allocation
|
Pinging @elastic/es-distributed |
|
Jenkins run elasticsearch-ci/packaging-sample |
There was a problem hiding this comment.
@DaveCTurner I did a first quick pass and it looks like this makes the listener invocation behave differently (and unsafe).
| @@ -95,7 +123,7 @@ public void onNoLongerMaster(String source) { | |||
| pendingRerouteListeners = null; | |||
| } | |||
| } | |||
| currentListeners.onFailure(new NotMasterException("delayed reroute [" + reason + "] cancelled")); | |||
| currentListeners.forEach(l -> l.onFailure(new NotMasterException("delayed reroute [" + reason + "] cancelled"))); | |||
There was a problem hiding this comment.
Maybe use org.elasticsearch.action.ActionListener#onFailure(java.lang.Iterable<org.elasticsearch.action.ActionListener<Response>>, java.lang.Exception) to make this a little safer?
| @@ -114,22 +142,26 @@ public void onFailure(String source, Exception e) { | |||
| logger.error(() -> new ParameterizedMessage("unexpected failure during [{}], current state version [{}]", | |||
| source, state.version()), e); | |||
| } | |||
| currentListeners.onFailure(new ElasticsearchException("delayed reroute [" + reason + "] failed", e)); | |||
| currentListeners.forEach(l -> | |||
There was a problem hiding this comment.
Same here, use org.elasticsearch.action.ActionListener#onFailure(java.lang.Iterable<org.elasticsearch.action.ActionListener<Response>>, java.lang.Exception)?
| } | ||
|
|
||
| @Override | ||
| public void clusterStateProcessed(String source, ClusterState oldState, ClusterState newState) { | ||
| currentListeners.onResponse(null); | ||
| currentListeners.forEach(l -> l.onResponse(null)); |
There was a problem hiding this comment.
Same issue as with the failures here as well. We probably should use org.elasticsearch.action.ActionListener#onResponse(java.lang.Iterable<org.elasticsearch.action.ActionListener<Response>>, Response) to not fail the loop on the first exception? (Also, this makes the behavior equivalent to what it was where PlainListenableActionFuture would do the same and ensure all the listeners are called).
|
Good point, thanks @original-brownbear. Fixed in 98796c4. |
Today the `BatchedRerouteService` submits its delayed reroute task at `HIGH` priority, but in some cases a lower priority would be more appropriate. This commit adds the facility to submit delayed reroute tasks at different priorities, such that each submitted reroute task runs at a priority no lower than the one requested. It does not change the fact that all delayed reroute tasks are submitted at `HIGH` priority, but at least it makes this explicit.
Today the
BatchedRerouteServicesubmits its delayed reroute task atHIGHpriority, but in some cases a lower priority would be more appropriate. This
commit adds the facility to submit delayed reroute tasks at different
priorities, such that each submitted reroute task runs at a priority no lower
than the one requested. It does not change the fact that all delayed reroute
tasks are submitted at
HIGHpriority, but at least it makes this explicit.