Inline TransportReplAct#createReplicatedOperation#41197
Inline TransportReplAct#createReplicatedOperation#41197DaveCTurner merged 5 commits intoelastic:masterfrom
Conversation
`TransportReplicationAction.AsyncPrimaryAction#createReplicatedOperation` exists so it can be overridden in tests. This commit re-works these tests to use a real `ReplicationOperation` and inlines the now-unnecessary method. Relates elastic#40706.
DaveCTurner
added
>non-issue
:Distributed/CRUD
|
Pinging @elastic/es-distributed |
| PlainActionFuture<TestResponse> listener = new PlainActionFuture<>(); | ||
| ReplicationTask task = maybeTask(); | ||
| int i = randomInt(3); | ||
| final boolean throwExceptionOnCreation = i == 1; |
There was a problem hiding this comment.
This case doesn't seem to be possible in production, so I removed it.
henningandersen
left a comment
henningandersen
left a comment
There was a problem hiding this comment.
LGTM.
Thanks @DaveCTurner, I left 3 comments to consider.
| @Override | ||
| public void onFailure(Exception e) { | ||
| handleException(primaryShardReference, e); | ||
| final ActionListener<Response> referenceClosingListener = ActionListener.wrap(response -> { |
There was a problem hiding this comment.
I find the separation into two listeners artificial and a bit confusing. I suggest something like following instead:
final ActionListener<Response> globalCheckpointSyncingListener = ActionListener.wrap(response -> {
if (syncGlobalCheckpointAfterOperation) {
final IndexShard shard = primaryShardReference.indexShard;
try {
shard.maybeSyncGlobalCheckpoint("post-operation");
} catch (final Exception e) {
// only log non-closed exceptions
if (ExceptionsHelper.unwrap(
e, AlreadyClosedException.class, IndexShardClosedException.class) == null) {
// intentionally swallow, a missed global checkpoint sync should not fail this operation
logger.info(
new ParameterizedMessage(
"{} failed to execute post-operation global checkpoint sync", shard.shardId()), e);
}
}
}
primaryShardReference.close(); // release shard operation lock before responding to caller
setPhase(replicationTask, "finished");
onCompletionListener.onResponse(response);
}, e -> handleException(primaryShardReference, e));
new ReplicationOperation<>(primaryRequest.getRequest(), primaryShardReference,
ActionListener.wrap(result -> result.respond(globalCheckpointSyncingListener),
globalCheckpointSyncingListener::onFailure),
newReplicasProxy(), logger, actionName, primaryRequest.getPrimaryTerm()).execute();There was a problem hiding this comment.
In isolation I agree, but this separation will be important in a followup so I hope it's ok to leave it like it is. The global checkpoint syncing is the responsibility of the primary, whereas the cleanup of the replication task and the primaryShardReference is the responsibility of the reroute/delegation phase.
| super.shardOperationOnPrimary(shardRequest, primary, listener); | ||
| } | ||
| }.run(); | ||
| }.new AsyncPrimaryAction(primaryRequest, ActionListener.wrap(listener::onResponse, throwable -> { |
There was a problem hiding this comment.
Could we instead of using ActionListener.wrap just assert that listener.isDone() and do listener.get() like in the test above?
| } | ||
| } | ||
|
|
||
| if (throwExceptionOnRun || respondWithError) { |
There was a problem hiding this comment.
nit: I think it is more logical to put this inside the try-catch (after listener.get()) and remove the return above.
| new ParameterizedMessage( | ||
| "{} failed to execute post-operation global checkpoint sync", shard.shardId()), e); | ||
| "{} failed to execute post-operation global checkpoint sync", | ||
| primaryShardReference.routingEntry().shardId()), e); |
There was a problem hiding this comment.
Not sure I follow this change, I cannot figure out how this makes a difference. I think using just shard.shardId() is simpler unless there is a reason for this?
There was a problem hiding this comment.
More foreshadowing of changes to come, but I can defer this until later.
This reverts commit a26a986.
`TransportReplicationAction.AsyncPrimaryAction#createReplicatedOperation` exists so it can be overridden in tests. This commit re-works these tests to use a real `ReplicationOperation` and inlines the now-unnecessary method. Relates #40706.
`TransportReplicationAction.AsyncPrimaryAction#createReplicatedOperation` exists so it can be overridden in tests. This commit re-works these tests to use a real `ReplicationOperation` and inlines the now-unnecessary method. Relates elastic#40706.
4 participants
TransportReplicationAction.AsyncPrimaryAction#createReplicatedOperationexists so it can be overridden in tests. This commit re-works these tests to
use a real
ReplicationOperationand inlines the now-unnecessary method.Relates #40706.