Skip to content
Merged
Show file tree
Hide file tree
Changes from 2 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions docs/changelog/127330.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
pr: 127330
summary: Move unpromotable relocations to its own transport action
area: Recovery
type: enhancement
issues: []
Original file line number Diff line number Diff line change
Expand Up @@ -301,18 +301,32 @@ private void doRecovery(final long recoveryId, final StartRecoveryRequest preExi
if (indexShard.routingEntry().isPromotableToPrimary() == false) {
assert preExistingRequest == null;
assert indexShard.indexSettings().getIndexMetadata().isSearchableSnapshot() == false;
ActionListener.run(cleanupOnly.map(v -> {
logger.trace("{} preparing unpromotable shard for recovery", recoveryTarget.shardId());
indexShard.prepareForIndexRecovery();
// Skip unnecessary intermediate stages
recoveryState.setStage(RecoveryState.Stage.VERIFY_INDEX);
recoveryState.setStage(RecoveryState.Stage.TRANSLOG);
indexShard.openEngineAndSkipTranslogRecovery();
recoveryState.getIndex().setFileDetailsComplete();
recoveryState.setStage(RecoveryState.Stage.FINALIZE);
onGoingRecoveries.markRecoveryAsDone(recoveryId);
return null;
}), indexShard::preRecovery);
try (onCompletion) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would think this releases the recovery monitor and the recovery-ref too soon? My intuition would be that it should only be done when the action completes?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My understanding is that the RecoveryTarget would be retained until the recovery is marked as done (since the initial refCount=1 from the AbstractRefCounted corresponds to that decRef). But just to be on the safe side I've reverted to the previous behaviour that would release the RecoveryRef once the action returns.

client.execute(
StatelessUnpromotableRelocationAction.TYPE,
new StatelessUnpromotableRelocationAction.Request(
recoveryId,
indexShard.shardId(),
indexShard.routingEntry().allocationId().getId(),
recoveryTarget.clusterStateVersion()
),
new ActionListener<>() {
@Override
public void onResponse(ActionResponse.Empty empty) {
onGoingRecoveries.markRecoveryAsDone(recoveryId);
}

@Override
public void onFailure(Exception e) {
onGoingRecoveries.failRecovery(
recoveryId,
new RecoveryFailedException(recoveryState, "failed to recover unpromotable shard", e),
true
);
}
}
);
}
return;
}

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -167,7 +167,6 @@ public RecoveryRef getRecoverySafe(long id, ShardId shardId) {
throw new IndexShardClosedException(shardId);
}
assert recoveryRef.target().shardId().equals(shardId);
assert recoveryRef.target().indexShard().routingEntry().isPromotableToPrimary();
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like this was added here, I am also not sure I understand why, perhaps @kingherc remember and can confirm that the assertion is not significant?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not out of the top of my head. But going back to the code, I see we've made a special branch in PeerRecoveryTargetService#doRecovery() with if (indexShard.routingEntry().isPromotableToPrimary() == false) { for unpromotables that basically quick skips all recovery stages, and closes the RecoveryRef as well. So the point of the assertion at the time was that there should be no other coordination needed for unpromotables to justify getting the RecoveryRef.

Seeing though that now this PR introduces some sort of coordination between unpromotables, it probably makes to remove the assertion.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(I did not fully review this PR, but feel free to tell me if I should)

return recoveryRef;
}

Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,95 @@
/*
* Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
* or more contributor license agreements. Licensed under the "Elastic License
* 2.0", the "GNU Affero General Public License v3.0 only", and the "Server Side
* Public License v 1"; you may not use this file except in compliance with, at
* your election, the "Elastic License 2.0", the "GNU Affero General Public
* License v3.0 only", or the "Server Side Public License, v 1".
*/

package org.elasticsearch.indices.recovery;

import org.elasticsearch.action.ActionRequest;
import org.elasticsearch.action.ActionRequestValidationException;
import org.elasticsearch.action.ActionResponse;
import org.elasticsearch.action.ActionType;
import org.elasticsearch.common.io.stream.StreamInput;
import org.elasticsearch.common.io.stream.StreamOutput;
import org.elasticsearch.index.shard.ShardId;

import java.io.IOException;
import java.util.Objects;

public class StatelessUnpromotableRelocationAction {

public static final ActionType<ActionResponse.Empty> TYPE = new ActionType<>(
"internal:index/shard/recovery/stateless_unpromotable_relocation"
);

public static class Request extends ActionRequest {
private final long recoveryId;
private final ShardId shardId;
private final String targetAllocationId;
private final long clusterStateVersion;

public Request(long recoveryId, ShardId shardId, String targetAllocationId, long clusterStateVersion) {
this.recoveryId = recoveryId;
this.shardId = shardId;
this.targetAllocationId = targetAllocationId;
this.clusterStateVersion = clusterStateVersion;
}

public Request(StreamInput in) throws IOException {
super(in);
recoveryId = in.readVLong();
shardId = new ShardId(in);
targetAllocationId = in.readString();
clusterStateVersion = in.readVLong();
}

@Override
public ActionRequestValidationException validate() {
return null;
}

@Override
public void writeTo(StreamOutput out) throws IOException {
super.writeTo(out);
out.writeVLong(recoveryId);
shardId.writeTo(out);
out.writeString(targetAllocationId);
out.writeVLong(clusterStateVersion);
}

public long getRecoveryId() {
return recoveryId;
}

public ShardId getShardId() {
return shardId;
}

public long getClusterStateVersion() {
return clusterStateVersion;
}

public String getTargetAllocationId() {
return targetAllocationId;
}

@Override
public boolean equals(Object o) {
if (o == null || getClass() != o.getClass()) return false;
Request request = (Request) o;
return recoveryId == request.recoveryId
&& clusterStateVersion == request.clusterStateVersion
&& Objects.equals(shardId, request.shardId)
&& Objects.equals(targetAllocationId, request.targetAllocationId);
}

@Override
public int hashCode() {
return Objects.hash(recoveryId, shardId, targetAllocationId, clusterStateVersion);
}
}
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,69 @@
/*
* Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
* or more contributor license agreements. Licensed under the "Elastic License
* 2.0", the "GNU Affero General Public License v3.0 only", and the "Server Side
* Public License v 1"; you may not use this file except in compliance with, at
* your election, the "Elastic License 2.0", the "GNU Affero General Public
* License v3.0 only", or the "Server Side Public License, v 1".
*/

package org.elasticsearch.indices.recovery;

import org.elasticsearch.common.UUIDs;
import org.elasticsearch.common.io.stream.Writeable;
import org.elasticsearch.index.shard.ShardId;
import org.elasticsearch.index.shard.ShardIdTests;
import org.elasticsearch.test.AbstractWireSerializingTestCase;
import org.elasticsearch.test.ESTestCase;

import java.io.IOException;

public class StatelessUnpromotableRelocationActionTests extends AbstractWireSerializingTestCase<
StatelessUnpromotableRelocationAction.Request> {
@Override
protected Writeable.Reader<StatelessUnpromotableRelocationAction.Request> instanceReader() {
return StatelessUnpromotableRelocationAction.Request::new;
}

@Override
protected StatelessUnpromotableRelocationAction.Request createTestInstance() {
return new StatelessUnpromotableRelocationAction.Request(
randomNonNegativeLong(),
new ShardId(randomIdentifier(), UUIDs.randomBase64UUID(), randomIntBetween(0, 99)),
randomUUID(),
randomNonNegativeLong()
);
}

@Override
protected StatelessUnpromotableRelocationAction.Request mutateInstance(StatelessUnpromotableRelocationAction.Request instance)
throws IOException {
return switch (between(0, 3)) {
case 0 -> new StatelessUnpromotableRelocationAction.Request(
randomValueOtherThan(instance.getRecoveryId(), ESTestCase::randomNonNegativeLong),
instance.getShardId(),
instance.getTargetAllocationId(),
instance.getClusterStateVersion()
);
case 1 -> new StatelessUnpromotableRelocationAction.Request(
instance.getRecoveryId(),
ShardIdTests.mutate(instance.getShardId()),
instance.getTargetAllocationId(),
instance.getClusterStateVersion()
);
case 2 -> new StatelessUnpromotableRelocationAction.Request(
instance.getRecoveryId(),
instance.getShardId(),
randomValueOtherThan(instance.getTargetAllocationId(), ESTestCase::randomUUID),
instance.getClusterStateVersion()
);
case 3 -> new StatelessUnpromotableRelocationAction.Request(
instance.getRecoveryId(),
instance.getShardId(),
instance.getTargetAllocationId(),
randomValueOtherThan(instance.getClusterStateVersion(), ESTestCase::randomNonNegativeLong)
);
default -> throw new AssertionError("Illegal randomisation branch");
};
}
}
Loading