Skip to content

Commit

Permalink
Refactor TransportShardBulkAction to better support retries (elastic#…
Browse files Browse the repository at this point in the history
…31821)

Processing bulk request goes item by item. Sometimes during processing, we need to stop execution and wait for a new mapping update to be processed by the node. This is currently achieved by throwing a `RetryOnPrimaryException`, which is caught higher up. When the exception is caught, we wait for the next cluster state to arrive and process the request again. Sadly this is a problem because all operations that were already done until the mapping change was required are applied again and get new sequence numbers. This in turn means that the previously issued sequence numbers are never replicated to the replicas. That causes the local checkpoint of those shards to be stuck and with it all the seq# based infrastructure.

This commit refactors how we deal with retries with the goal of removing  `RetryOnPrimaryException` and `RetryOnReplicaException` (not done yet). It achieves so by introducing a class `BulkPrimaryExecutionContext` that is used the capture the execution state and allows continuing from where the execution stopped. The class also formalizes the steps each item has to go through:
1) A translation phase for updates
2) Execution phase (always index/delete)
3) Waiting for a mapping update to come in, if needed
4) Requires a retry (for updates and cases where the mapping are still not available after the put mapping call returns)
5) A finalization phase which allows updates to the index/delete result to an update result.
  • Loading branch information
bleskes authored Aug 10, 2018
1 parent af8c23e commit f58ed21
Show file tree
Hide file tree
Showing 9 changed files with 1,113 additions and 748 deletions.

This file was deleted.

Original file line number Diff line number Diff line change
@@ -0,0 +1,345 @@
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch licenses this file to you under
* the Apache License, Version 2.0 (the "License"); you may
* not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/

package org.elasticsearch.action.bulk;

import org.elasticsearch.action.DocWriteRequest;
import org.elasticsearch.action.DocWriteResponse;
import org.elasticsearch.action.delete.DeleteResponse;
import org.elasticsearch.action.index.IndexResponse;
import org.elasticsearch.action.support.replication.ReplicationResponse;
import org.elasticsearch.action.support.replication.TransportWriteAction;
import org.elasticsearch.index.engine.Engine;
import org.elasticsearch.index.shard.IndexShard;
import org.elasticsearch.index.translog.Translog;

import java.util.Arrays;

/**
* This is a utility class that holds the per request state needed to perform bulk operations on the primary.
* More specifically, it maintains an index to the current executing bulk item, which allows execution
* to stop and wait for external events such as mapping updates.
*/
class BulkPrimaryExecutionContext {

enum ItemProcessingState {
/** Item execution is ready to start, no operations have been performed yet */
INITIAL,
/**
* The incoming request has been translated to a request that can be executed on the shard.
* This is used to convert update requests to a fully specified index or delete requests.
*/
TRANSLATED,
/**
* the request can not execute with the current mapping and should wait for a new mapping
* to arrive from the master. A mapping request for the needed changes has already been
* submitted
*/
WAIT_FOR_MAPPING_UPDATE,
/**
* The request should be executed again, but there is no need to wait for an external event.
* This is needed to support retry on conflicts during updates.
*/
IMMEDIATE_RETRY,
/** The request has been executed on the primary shard (successfully or not) */
EXECUTED,
/**
* No further handling of current request is needed. The result has been converted to a user response
* and execution can continue to the next item (if available).
*/
COMPLETED
}

private final BulkShardRequest request;
private final IndexShard primary;
private Translog.Location locationToSync = null;
private int currentIndex = -1;

private ItemProcessingState currentItemState;
private DocWriteRequest requestToExecute;
private BulkItemResponse executionResult;
private int retryCounter;


BulkPrimaryExecutionContext(BulkShardRequest request, IndexShard primary) {
this.request = request;
this.primary = primary;
advance();
}


private int findNextNonAborted(int startIndex) {
final int length = request.items().length;
while (startIndex < length && isAborted(request.items()[startIndex].getPrimaryResponse())) {
startIndex++;
}
return startIndex;
}

private static boolean isAborted(BulkItemResponse response) {
return response != null && response.isFailed() && response.getFailure().isAborted();
}

/** move to the next item to execute */
private void advance() {
assert currentItemState == ItemProcessingState.COMPLETED || currentIndex == -1 :
"moving to next but current item wasn't completed (state: " + currentItemState + ")";
currentItemState = ItemProcessingState.INITIAL;
currentIndex = findNextNonAborted(currentIndex + 1);
retryCounter = 0;
requestToExecute = null;
executionResult = null;
assert assertInvariants(ItemProcessingState.INITIAL);
}

/** gets the current, untranslated item request */
public DocWriteRequest<?> getCurrent() {
return getCurrentItem().request();
}

public BulkShardRequest getBulkShardRequest() {
return request;
}

/** returns the result of the request that has been executed on the shard */
public BulkItemResponse getExecutionResult() {
assert assertInvariants(ItemProcessingState.EXECUTED);
return executionResult;
}

/** returns the number of times the current operation has been retried */
public int getRetryCounter() {
return retryCounter;
}

/** returns true if the current request has been executed on the primary */
public boolean isOperationExecuted() {
return currentItemState == ItemProcessingState.EXECUTED;
}

/** returns true if the request needs to wait for a mapping update to arrive from the master */
public boolean requiresWaitingForMappingUpdate() {
return currentItemState == ItemProcessingState.WAIT_FOR_MAPPING_UPDATE;
}

/** returns true if the current request should be retried without waiting for an external event */
public boolean requiresImmediateRetry() {
return currentItemState == ItemProcessingState.IMMEDIATE_RETRY;
}

/**
* returns true if the current request has been completed and it's result translated to a user
* facing response
*/
public boolean isCompleted() {
return currentItemState == ItemProcessingState.COMPLETED;
}

/**
* returns true if the current request is in INITIAL state
*/
public boolean isInitial() {
return currentItemState == ItemProcessingState.INITIAL;
}

/**
* returns true if {@link #advance()} has moved the current item beyond the
* end of the {@link BulkShardRequest#items()} array.
*/
public boolean hasMoreOperationsToExecute() {
return currentIndex < request.items().length;
}


/** returns the name of the index the current request used */
public String getConcreteIndex() {
return getCurrentItem().index();
}

/** returns any primary response that was set by a previous primary */
public BulkItemResponse getPreviousPrimaryResponse() {
return getCurrentItem().getPrimaryResponse();
}

/** returns a translog location that is needed to be synced in order to persist all operations executed so far */
public Translog.Location getLocationToSync() {
assert hasMoreOperationsToExecute() == false;
// we always get to the end of the list by using advance, which in turn sets the state to INITIAL
assert assertInvariants(ItemProcessingState.INITIAL);
return locationToSync;
}

private BulkItemRequest getCurrentItem() {
return request.items()[currentIndex];
}

/** returns the primary shard */
public IndexShard getPrimary() {
return primary;
}

/**
* sets the request that should actually be executed on the primary. This can be different then the request
* received from the user (specifically, an update request is translated to an indexing or delete request).
*/
public void setRequestToExecute(DocWriteRequest writeRequest) {
assert assertInvariants(ItemProcessingState.INITIAL);
requestToExecute = writeRequest;
currentItemState = ItemProcessingState.TRANSLATED;
assert assertInvariants(ItemProcessingState.TRANSLATED);
}

/** returns the request that should be executed on the shard. */
public <T extends DocWriteRequest<T>> T getRequestToExecute() {
assert assertInvariants(ItemProcessingState.TRANSLATED);
return (T) requestToExecute;
}

/** indicates that the current operation can not be completed and needs to wait for a new mapping from the master */
public void markAsRequiringMappingUpdate() {
assert assertInvariants(ItemProcessingState.TRANSLATED);
currentItemState = ItemProcessingState.WAIT_FOR_MAPPING_UPDATE;
requestToExecute = null;
assert assertInvariants(ItemProcessingState.WAIT_FOR_MAPPING_UPDATE);
}

/** resets the current item state, prepare for a new execution */
public void resetForExecutionForRetry() {
assertInvariants(ItemProcessingState.WAIT_FOR_MAPPING_UPDATE, ItemProcessingState.EXECUTED);
currentItemState = ItemProcessingState.INITIAL;
requestToExecute = null;
executionResult = null;
assertInvariants(ItemProcessingState.INITIAL);
}

/** completes the operation without doing anything on the primary */
public void markOperationAsNoOp(DocWriteResponse response) {
assertInvariants(ItemProcessingState.INITIAL);
executionResult = new BulkItemResponse(getCurrentItem().id(), getCurrentItem().request().opType(), response);
currentItemState = ItemProcessingState.EXECUTED;
assertInvariants(ItemProcessingState.EXECUTED);
}

/** indicates that the operation needs to be failed as the required mapping didn't arrive in time */
public void failOnMappingUpdate(Exception cause) {
assert assertInvariants(ItemProcessingState.WAIT_FOR_MAPPING_UPDATE);
currentItemState = ItemProcessingState.EXECUTED;
final DocWriteRequest docWriteRequest = getCurrentItem().request();
executionResult = new BulkItemResponse(getCurrentItem().id(), docWriteRequest.opType(),
// Make sure to use getCurrentItem().index() here, if you use docWriteRequest.index() it will use the
// concrete index instead of an alias if used!
new BulkItemResponse.Failure(getCurrentItem().index(), docWriteRequest.type(), docWriteRequest.id(), cause));
markAsCompleted(executionResult);
}

/** the current operation has been executed on the primary with the specified result */
public void markOperationAsExecuted(Engine.Result result) {
assertInvariants(ItemProcessingState.TRANSLATED);
final BulkItemRequest current = getCurrentItem();
DocWriteRequest docWriteRequest = getRequestToExecute();
switch (result.getResultType()) {
case SUCCESS:
final DocWriteResponse response;
if (result.getOperationType() == Engine.Operation.TYPE.INDEX) {
Engine.IndexResult indexResult = (Engine.IndexResult) result;
response = new IndexResponse(primary.shardId(), requestToExecute.type(), requestToExecute.id(),
result.getSeqNo(), result.getTerm(), indexResult.getVersion(), indexResult.isCreated());
} else if (result.getOperationType() == Engine.Operation.TYPE.DELETE) {
Engine.DeleteResult deleteResult = (Engine.DeleteResult) result;
response = new DeleteResponse(primary.shardId(), requestToExecute.type(), requestToExecute.id(),
deleteResult.getSeqNo(), result.getTerm(), deleteResult.getVersion(), deleteResult.isFound());

} else {
throw new AssertionError("unknown result type :" + result.getResultType());
}
executionResult = new BulkItemResponse(current.id(), current.request().opType(), response);
// set a blank ShardInfo so we can safely send it to the replicas. We won't use it in the real response though.
executionResult.getResponse().setShardInfo(new ReplicationResponse.ShardInfo());
locationToSync = TransportWriteAction.locationToSync(locationToSync, result.getTranslogLocation());
break;
case FAILURE:
executionResult = new BulkItemResponse(current.id(), docWriteRequest.opType(),
// Make sure to use request.index() here, if you
// use docWriteRequest.index() it will use the
// concrete index instead of an alias if used!
new BulkItemResponse.Failure(request.index(), docWriteRequest.type(), docWriteRequest.id(),
result.getFailure(), result.getSeqNo()));
break;
default:
throw new AssertionError("unknown result type for " + getCurrentItem() + ": " + result.getResultType());
}
currentItemState = ItemProcessingState.EXECUTED;
}

/** finishes the execution of the current request, with the response that should be returned to the user */
public void markAsCompleted(BulkItemResponse translatedResponse) {
assertInvariants(ItemProcessingState.EXECUTED);
assert executionResult == null || translatedResponse.getItemId() == executionResult.getItemId();
assert translatedResponse.getItemId() == getCurrentItem().id();

if (translatedResponse.isFailed() == false && requestToExecute != getCurrent()) {
request.items()[currentIndex] = new BulkItemRequest(request.items()[currentIndex].id(), requestToExecute);
}
getCurrentItem().setPrimaryResponse(translatedResponse);
currentItemState = ItemProcessingState.COMPLETED;
advance();
}

/** builds the bulk shard response to return to the user */
public BulkShardResponse buildShardResponse() {
assert hasMoreOperationsToExecute() == false;
return new BulkShardResponse(request.shardId(),
Arrays.stream(request.items()).map(BulkItemRequest::getPrimaryResponse).toArray(BulkItemResponse[]::new));
}

private boolean assertInvariants(ItemProcessingState... expectedCurrentState) {
assert Arrays.asList(expectedCurrentState).contains(currentItemState):
"expected current state [" + currentItemState + "] to be one of " + Arrays.toString(expectedCurrentState);
assert currentIndex >= 0 : currentIndex;
assert retryCounter >= 0 : retryCounter;
switch (currentItemState) {
case INITIAL:
assert requestToExecute == null : requestToExecute;
assert executionResult == null : executionResult;
break;
case TRANSLATED:
assert requestToExecute != null;
assert executionResult == null : executionResult;
break;
case WAIT_FOR_MAPPING_UPDATE:
assert requestToExecute == null;
assert executionResult == null : executionResult;
break;
case IMMEDIATE_RETRY:
assert requestToExecute != null;
assert executionResult == null : executionResult;
break;
case EXECUTED:
// requestToExecute can be null if the update ended up as NOOP
assert executionResult != null;
break;
case COMPLETED:
assert requestToExecute != null;
assert executionResult != null;
assert getCurrentItem().getPrimaryResponse() != null;
break;
}
return true;
}
}
Loading

0 comments on commit f58ed21

Please sign in to comment.