-
Notifications
You must be signed in to change notification settings - Fork 24.3k
/
IndexMetaDataUpdater.java
328 lines (297 loc) · 18.1 KB
/
IndexMetaDataUpdater.java
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
/*
* 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.cluster.routing.allocation;
import org.apache.logging.log4j.Logger;
import org.elasticsearch.cluster.ClusterState;
import org.elasticsearch.cluster.metadata.IndexMetaData;
import org.elasticsearch.cluster.metadata.MetaData;
import org.elasticsearch.cluster.routing.IndexShardRoutingTable;
import org.elasticsearch.cluster.routing.RecoverySource;
import org.elasticsearch.cluster.routing.RoutingChangesObserver;
import org.elasticsearch.cluster.routing.RoutingTable;
import org.elasticsearch.cluster.routing.ShardRouting;
import org.elasticsearch.cluster.routing.UnassignedInfo;
import org.elasticsearch.common.util.set.Sets;
import org.elasticsearch.index.Index;
import org.elasticsearch.index.shard.ShardId;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.stream.Collectors;
/**
* Observer that tracks changes made to RoutingNodes in order to update the primary terms and in-sync allocation ids in
* {@link IndexMetaData} once the allocation round has completed.
*
* Primary terms are updated on primary initialization or when an active primary fails.
*
* Allocation ids are added for shards that become active and removed for shards that stop being active.
*/
public class IndexMetaDataUpdater extends RoutingChangesObserver.AbstractRoutingChangesObserver {
private final Map<ShardId, Updates> shardChanges = new HashMap<>();
@Override
public void shardInitialized(ShardRouting unassignedShard, ShardRouting initializedShard) {
assert initializedShard.isRelocationTarget() == false : "shardInitialized is not called on relocation target: " + initializedShard;
if (initializedShard.primary()) {
increasePrimaryTerm(initializedShard.shardId());
Updates updates = changes(initializedShard.shardId());
assert updates.initializedPrimary == null : "Primary cannot be initialized more than once in same allocation round: " +
"(previous: " + updates.initializedPrimary + ", next: " + initializedShard + ")";
updates.initializedPrimary = initializedShard;
}
}
@Override
public void shardStarted(ShardRouting initializingShard, ShardRouting startedShard) {
addAllocationId(startedShard);
if (startedShard.primary()
// started shard has to have null recoverySource; have to pick up recoverySource from its initializing state
&& (initializingShard.recoverySource() instanceof RecoverySource.ExistingStoreRecoverySource
|| initializingShard.recoverySource() instanceof RecoverySource.SnapshotRecoverySource)) {
Updates updates = changes(startedShard.shardId());
updates.removedAllocationIds.add(RecoverySource.ExistingStoreRecoverySource.FORCED_ALLOCATION_ID);
}
}
@Override
public void shardFailed(ShardRouting failedShard, UnassignedInfo unassignedInfo) {
if (failedShard.active() && failedShard.primary()) {
Updates updates = changes(failedShard.shardId());
if (updates.firstFailedPrimary == null) {
// more than one primary can be failed (because of batching, primary can be failed, replica promoted and then failed...)
updates.firstFailedPrimary = failedShard;
}
increasePrimaryTerm(failedShard.shardId());
}
}
@Override
public void relocationCompleted(ShardRouting removedRelocationSource) {
removeAllocationId(removedRelocationSource);
}
/**
* Updates the current {@link MetaData} based on the changes of this RoutingChangesObserver. Specifically
* we update {@link IndexMetaData#getInSyncAllocationIds()} and {@link IndexMetaData#primaryTerm(int)} based on
* the changes made during this allocation.
*
* @param oldMetaData {@link MetaData} object from before the routing nodes was changed.
* @param newRoutingTable {@link RoutingTable} object after routing changes were applied.
* @return adapted {@link MetaData}, potentially the original one if no change was needed.
*/
public MetaData applyChanges(MetaData oldMetaData, RoutingTable newRoutingTable) {
Map<Index, List<Map.Entry<ShardId, Updates>>> changesGroupedByIndex =
shardChanges.entrySet().stream().collect(Collectors.groupingBy(e -> e.getKey().getIndex()));
MetaData.Builder metaDataBuilder = null;
for (Map.Entry<Index, List<Map.Entry<ShardId, Updates>>> indexChanges : changesGroupedByIndex.entrySet()) {
Index index = indexChanges.getKey();
final IndexMetaData oldIndexMetaData = oldMetaData.getIndexSafe(index);
IndexMetaData.Builder indexMetaDataBuilder = null;
for (Map.Entry<ShardId, Updates> shardEntry : indexChanges.getValue()) {
ShardId shardId = shardEntry.getKey();
Updates updates = shardEntry.getValue();
indexMetaDataBuilder = updateInSyncAllocations(newRoutingTable, oldIndexMetaData, indexMetaDataBuilder, shardId, updates);
indexMetaDataBuilder = updatePrimaryTerm(oldIndexMetaData, indexMetaDataBuilder, shardId, updates);
}
if (indexMetaDataBuilder != null) {
if (metaDataBuilder == null) {
metaDataBuilder = MetaData.builder(oldMetaData);
}
metaDataBuilder.put(indexMetaDataBuilder);
}
}
if (metaDataBuilder != null) {
return metaDataBuilder.build();
} else {
return oldMetaData;
}
}
/**
* Updates in-sync allocations with routing changes that were made to the routing table.
*/
private IndexMetaData.Builder updateInSyncAllocations(RoutingTable newRoutingTable, IndexMetaData oldIndexMetaData,
IndexMetaData.Builder indexMetaDataBuilder, ShardId shardId, Updates updates) {
assert Sets.haveEmptyIntersection(updates.addedAllocationIds, updates.removedAllocationIds) :
"allocation ids cannot be both added and removed in the same allocation round, added ids: " +
updates.addedAllocationIds + ", removed ids: " + updates.removedAllocationIds;
Set<String> oldInSyncAllocationIds = oldIndexMetaData.inSyncAllocationIds(shardId.id());
// check if we have been force-initializing an empty primary or a stale primary
if (updates.initializedPrimary != null && oldInSyncAllocationIds.isEmpty() == false &&
oldInSyncAllocationIds.contains(updates.initializedPrimary.allocationId().getId()) == false) {
// we're not reusing an existing in-sync allocation id to initialize a primary, which means that we're either force-allocating
// an empty or a stale primary (see AllocateEmptyPrimaryAllocationCommand or AllocateStalePrimaryAllocationCommand).
RecoverySource recoverySource = updates.initializedPrimary.recoverySource();
RecoverySource.Type recoverySourceType = recoverySource.getType();
boolean emptyPrimary = recoverySourceType == RecoverySource.Type.EMPTY_STORE;
assert updates.addedAllocationIds.isEmpty() : (emptyPrimary ? "empty" : "stale") +
" primary is not force-initialized in same allocation round where shards are started";
if (indexMetaDataBuilder == null) {
indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
}
if (emptyPrimary) {
// forcing an empty primary resets the in-sync allocations to the empty set (ShardRouting.allocatedPostIndexCreate)
indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(), Collections.emptySet());
} else {
assert recoverySource instanceof RecoverySource.ExistingStoreRecoverySource
|| recoverySource instanceof RecoverySource.SnapshotRecoverySource
: recoverySource;
// forcing a stale primary resets the in-sync allocations to the singleton set with the stale id
indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(),
Collections.singleton(RecoverySource.ExistingStoreRecoverySource.FORCED_ALLOCATION_ID));
}
} else {
// standard path for updating in-sync ids
Set<String> inSyncAllocationIds = new HashSet<>(oldInSyncAllocationIds);
inSyncAllocationIds.addAll(updates.addedAllocationIds);
inSyncAllocationIds.removeAll(updates.removedAllocationIds);
assert oldInSyncAllocationIds.contains(RecoverySource.ExistingStoreRecoverySource.FORCED_ALLOCATION_ID) == false
|| inSyncAllocationIds.size() == 1
&& inSyncAllocationIds.contains(RecoverySource.ExistingStoreRecoverySource.FORCED_ALLOCATION_ID) == false :
"fake allocation id has to be removed, inSyncAllocationIds:" + inSyncAllocationIds;
// Prevent set of inSyncAllocationIds to grow unboundedly. This can happen for example if we don't write to a primary
// but repeatedly shut down nodes that have active replicas.
// We use number_of_replicas + 1 (= possible active shard copies) to bound the inSyncAllocationIds set
// Only trim the set of allocation ids when it grows, otherwise we might trim too eagerly when the number
// of replicas was decreased while shards were unassigned.
int maxActiveShards = oldIndexMetaData.getNumberOfReplicas() + 1; // +1 for the primary
IndexShardRoutingTable newShardRoutingTable = newRoutingTable.shardRoutingTable(shardId);
if (inSyncAllocationIds.size() > oldInSyncAllocationIds.size() && inSyncAllocationIds.size() > maxActiveShards) {
// trim entries that have no corresponding shard routing in the cluster state (i.e. trim unavailable copies)
List<ShardRouting> assignedShards = newShardRoutingTable.assignedShards();
assert assignedShards.size() <= maxActiveShards :
"cannot have more assigned shards " + assignedShards + " than maximum possible active shards " + maxActiveShards;
Set<String> assignedAllocations = assignedShards.stream().map(s -> s.allocationId().getId()).collect(Collectors.toSet());
inSyncAllocationIds = inSyncAllocationIds.stream()
.sorted(Comparator.comparing(assignedAllocations::contains).reversed()) // values with routing entries first
.limit(maxActiveShards)
.collect(Collectors.toSet());
}
// only remove allocation id of failed active primary if there is at least one active shard remaining. Assume for example that
// the primary fails but there is no new primary to fail over to. If we were to remove the allocation id of the primary from the
// in-sync set, this could create an empty primary on the next allocation.
if (newShardRoutingTable.activeShards().isEmpty() && updates.firstFailedPrimary != null) {
// add back allocation id of failed primary
inSyncAllocationIds.add(updates.firstFailedPrimary.allocationId().getId());
}
assert inSyncAllocationIds.isEmpty() == false || oldInSyncAllocationIds.isEmpty() :
"in-sync allocations cannot become empty after they have been non-empty: " + oldInSyncAllocationIds;
// be extra safe here and only update in-sync set if it is non-empty
if (inSyncAllocationIds.isEmpty() == false) {
if (indexMetaDataBuilder == null) {
indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
}
indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(), inSyncAllocationIds);
}
}
return indexMetaDataBuilder;
}
/**
* Removes allocation ids from the in-sync set for shard copies for which there is no routing entries in the routing table.
* This method is called in AllocationService before any changes to the routing table are made.
*/
public static ClusterState removeStaleIdsWithoutRoutings(ClusterState clusterState, List<StaleShard> staleShards, Logger logger) {
MetaData oldMetaData = clusterState.metaData();
RoutingTable oldRoutingTable = clusterState.routingTable();
MetaData.Builder metaDataBuilder = null;
// group staleShards entries by index
for (Map.Entry<Index, List<StaleShard>> indexEntry : staleShards.stream().collect(
Collectors.groupingBy(fs -> fs.getShardId().getIndex())).entrySet()) {
final IndexMetaData oldIndexMetaData = oldMetaData.getIndexSafe(indexEntry.getKey());
IndexMetaData.Builder indexMetaDataBuilder = null;
// group staleShards entries by shard id
for (Map.Entry<ShardId, List<StaleShard>> shardEntry : indexEntry.getValue().stream().collect(
Collectors.groupingBy(staleShard -> staleShard.getShardId())).entrySet()) {
int shardNumber = shardEntry.getKey().getId();
Set<String> oldInSyncAllocations = oldIndexMetaData.inSyncAllocationIds(shardNumber);
Set<String> idsToRemove = shardEntry.getValue().stream().map(e -> e.getAllocationId()).collect(Collectors.toSet());
assert idsToRemove.stream().allMatch(id -> oldRoutingTable.getByAllocationId(shardEntry.getKey(), id) == null) :
"removing stale ids: " + idsToRemove + ", some of which have still a routing entry: " + oldRoutingTable;
Set<String> remainingInSyncAllocations = Sets.difference(oldInSyncAllocations, idsToRemove);
assert remainingInSyncAllocations.isEmpty() == false : "Set of in-sync ids cannot become empty for shard " +
shardEntry.getKey() + " (before: " + oldInSyncAllocations + ", ids to remove: " + idsToRemove + ")";
// be extra safe here: if the in-sync set were to become empty, this would create an empty primary on the next allocation
// (see ShardRouting#allocatedPostIndexCreate)
if (remainingInSyncAllocations.isEmpty() == false) {
if (indexMetaDataBuilder == null) {
indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
}
indexMetaDataBuilder.putInSyncAllocationIds(shardNumber, remainingInSyncAllocations);
}
logger.warn("{} marking unavailable shards as stale: {}", shardEntry.getKey(), idsToRemove);
}
if (indexMetaDataBuilder != null) {
if (metaDataBuilder == null) {
metaDataBuilder = MetaData.builder(oldMetaData);
}
metaDataBuilder.put(indexMetaDataBuilder);
}
}
if (metaDataBuilder != null) {
return ClusterState.builder(clusterState).metaData(metaDataBuilder).build();
} else {
return clusterState;
}
}
/**
* Increases the primary term if {@link #increasePrimaryTerm} was called for this shard id.
*/
private IndexMetaData.Builder updatePrimaryTerm(IndexMetaData oldIndexMetaData, IndexMetaData.Builder indexMetaDataBuilder,
ShardId shardId, Updates updates) {
if (updates.increaseTerm) {
if (indexMetaDataBuilder == null) {
indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
}
indexMetaDataBuilder.primaryTerm(shardId.id(), oldIndexMetaData.primaryTerm(shardId.id()) + 1);
}
return indexMetaDataBuilder;
}
/**
* Helper method that creates update entry for the given shard id if such an entry does not exist yet.
*/
private Updates changes(ShardId shardId) {
return shardChanges.computeIfAbsent(shardId, k -> new Updates());
}
/**
* Remove allocation id of this shard from the set of in-sync shard copies
*/
void removeAllocationId(ShardRouting shardRouting) {
if (shardRouting.active()) {
changes(shardRouting.shardId()).removedAllocationIds.add(shardRouting.allocationId().getId());
}
}
/**
* Add allocation id of this shard to the set of in-sync shard copies
*/
private void addAllocationId(ShardRouting shardRouting) {
final Updates changes = changes(shardRouting.shardId());
changes.addedAllocationIds.add(shardRouting.allocationId().getId());
}
/**
* Increase primary term for this shard id
*/
private void increasePrimaryTerm(ShardId shardId) {
changes(shardId).increaseTerm = true;
}
private static class Updates {
private boolean increaseTerm; // whether primary term should be increased
private Set<String> addedAllocationIds = new HashSet<>(); // allocation ids that should be added to the in-sync set
private Set<String> removedAllocationIds = new HashSet<>(); // allocation ids that should be removed from the in-sync set
private ShardRouting initializedPrimary = null; // primary that was initialized from unassigned
private ShardRouting firstFailedPrimary = null; // first active primary that was failed
}
}