-
Notifications
You must be signed in to change notification settings - Fork 24.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Wait for all Rec. to Stop on Node Close #46178
Changes from all commits
2da67c0
d846de5
8c79b03
dbb15d7
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -24,10 +24,13 @@ | |
import org.elasticsearch.ExceptionsHelper; | ||
import org.elasticsearch.action.ActionListener; | ||
import org.elasticsearch.action.support.ChannelActionListener; | ||
import org.elasticsearch.action.support.PlainActionFuture; | ||
import org.elasticsearch.cluster.routing.ShardRouting; | ||
import org.elasticsearch.common.Nullable; | ||
import org.elasticsearch.common.component.AbstractLifecycleComponent; | ||
import org.elasticsearch.common.inject.Inject; | ||
import org.elasticsearch.common.settings.Settings; | ||
import org.elasticsearch.common.util.concurrent.FutureUtils; | ||
import org.elasticsearch.index.IndexService; | ||
import org.elasticsearch.index.shard.IndexEventListener; | ||
import org.elasticsearch.index.shard.IndexShard; | ||
|
@@ -50,7 +53,7 @@ | |
* The source recovery accepts recovery requests from other peer shards and start the recovery process from this | ||
* source shard to the target shard. | ||
*/ | ||
public class PeerRecoverySourceService implements IndexEventListener { | ||
public class PeerRecoverySourceService extends AbstractLifecycleComponent implements IndexEventListener { | ||
|
||
private static final Logger logger = LogManager.getLogger(PeerRecoverySourceService.class); | ||
|
||
|
@@ -74,6 +77,19 @@ public PeerRecoverySourceService(TransportService transportService, IndicesServi | |
new StartRecoveryTransportRequestHandler()); | ||
} | ||
|
||
@Override | ||
protected void doStart() { | ||
} | ||
|
||
@Override | ||
protected void doStop() { | ||
ongoingRecoveries.awaitEmpty(); | ||
} | ||
|
||
@Override | ||
protected void doClose() { | ||
} | ||
|
||
@Override | ||
public void beforeIndexShardClosed(ShardId shardId, @Nullable IndexShard indexShard, | ||
Settings indexSettings) { | ||
|
@@ -118,9 +134,14 @@ final int numberOfOngoingRecoveries() { | |
} | ||
|
||
final class OngoingRecoveries { | ||
|
||
private final Map<IndexShard, ShardRecoveryContext> ongoingRecoveries = new HashMap<>(); | ||
|
||
@Nullable | ||
private List<ActionListener<Void>> emptyListeners; | ||
|
||
synchronized RecoverySourceHandler addNewRecovery(StartRecoveryRequest request, IndexShard shard) { | ||
assert lifecycle.started(); | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. 👍 There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. When such an assertion fails, we don't have context about why the assertion failed (only that |
||
final ShardRecoveryContext shardContext = ongoingRecoveries.computeIfAbsent(shard, s -> new ShardRecoveryContext()); | ||
RecoverySourceHandler handler = shardContext.addNewRecovery(request, shard); | ||
shard.recoveryStats().incCurrentAsSource(); | ||
|
@@ -138,6 +159,13 @@ synchronized void remove(IndexShard shard, RecoverySourceHandler handler) { | |
if (shardRecoveryContext.recoveryHandlers.isEmpty()) { | ||
ongoingRecoveries.remove(shard); | ||
} | ||
if (ongoingRecoveries.isEmpty()) { | ||
if (emptyListeners != null) { | ||
final List<ActionListener<Void>> onEmptyListeners = emptyListeners; | ||
emptyListeners = null; | ||
ActionListener.onResponse(onEmptyListeners, null); | ||
} | ||
} | ||
} | ||
|
||
synchronized void cancel(IndexShard shard, String reason) { | ||
|
@@ -157,6 +185,22 @@ synchronized void cancel(IndexShard shard, String reason) { | |
} | ||
} | ||
|
||
void awaitEmpty() { | ||
assert lifecycle.stoppedOrClosed(); | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Can we be sure that no new recoveries may start after this method has returned? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. The indices services close before this one so it should be fine. I added an assertion about the state of the peer recovery source service before adding a recovery though to make sure :) There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Similarly to my earlier comment, which is a general comment about code assertions, it would be nice to have more context here in case this assertion ever fails. |
||
final PlainActionFuture<Void> future; | ||
synchronized (this) { | ||
if (ongoingRecoveries.isEmpty()) { | ||
return; | ||
} | ||
future = new PlainActionFuture<>(); | ||
if (emptyListeners == null) { | ||
emptyListeners = new ArrayList<>(); | ||
} | ||
emptyListeners.add(future); | ||
} | ||
FutureUtils.get(future); | ||
} | ||
|
||
private final class ShardRecoveryContext { | ||
final Set<RecoverySourceHandler> recoveryHandlers = new HashSet<>(); | ||
|
||
|
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -402,13 +402,14 @@ private Releasable acquireStore(Store store) { | |
store.incRef(); | ||
return Releasables.releaseOnce(() -> { | ||
final PlainActionFuture<Void> future = new PlainActionFuture<>(); | ||
threadPool.generic().execute(new ActionRunnable<>(future) { | ||
@Override | ||
protected void doRun() { | ||
store.decRef(); | ||
listener.onResponse(null); | ||
} | ||
}); | ||
assert threadPool.generic().isShutdown() == false; | ||
// TODO: We shouldn't use the generic thread pool here as we already execute this from the generic pool. | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. @henningandersen Maybe we need to revisit the original solution (in #45409) to avoid forking here. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Yes, I agree to that. I think the best path is to make closing the resources async (except the one in the outer exception handler) such that we notify the recovery-response listener after completion of closing resources (we sometimes notify first and then close). |
||
// While practically unlikely at a min pool size of 128 we could technically block the whole pool by waiting on futures | ||
// below and thus make it impossible for the store release to execute which in turn would block the futures forever | ||
threadPool.generic().execute(ActionRunnable.wrap(future, l -> { | ||
store.decRef(); | ||
l.onResponse(null); | ||
})); | ||
FutureUtils.get(future); | ||
}); | ||
} | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Does this awaitEmpty block a node from closing? Should we just cancel all ongoing recoveries?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@dnhatn it was my understanding that we already do cancel all ongoing recoveries when shutting down the indices service (that's what happens in practice at the moment it seems) => I didn't think I need to take any action here.
Yes, but see above. I think this is fine. In the end, if the recoveries aren't properly stopped we will eventually interrupt this wait in the test cluster code and throw an exception so this won't silently cause trouble in the future either.