ES5.6.4原始碼分析----索引的建立過程
阿新 • • 發佈：2018-12-12
入口
TransportCreateIndexAction#masterOperation
解析請求中的索引名稱
final String indexName = indexNameExpressionResolver.resolveDateMathExpression(request.index());
由於建立索引請求中的索引名稱可能是日期數學表示式表示的，所以要將其解析成具體的名稱。
規範化+校驗請求中的setting引數
updatedSettingsBuilder.put(request.settings()).normalizePrefix(IndexMetaData.INDEX_SETTING_PREFIX); indexScopedSettings.validate(updatedSettingsBuilder); request.settings(updatedSettingsBuilder.build());
請求體例子： { “settings”:{ “number_of_shards”:3, “number_of_replicas”:0 } }
- 檢查settings中的引數是否以”index.”開頭，如果不是將這個字首加上 number_of_shards => index.number_of_shards
- 檢查ES是否支援這些配置
將請求封裝成任務提交
clusterService.submitStateUpdateTask("create-index [" + request.index() + "], cause [" + request.cause() + "]", new AckedClusterStateUpdateTask<ClusterStateUpdateResponse>(Priority.URGENT, request, wrapPreservingContext(listener, threadPool.getThreadContext())) { @Override protected ClusterStateUpdateResponse newResponse(boolean acknowledged) { return new ClusterStateUpdateResponse(acknowledged); } @Override public ClusterState execute(ClusterState currentState) throws Exception { Index createdIndex = null; String removalExtraInfo = null; IndexRemovalReason removalReason = IndexRemovalReason.FAILURE; try { validate(request, currentState); for (Alias alias : request.aliases()) { aliasValidator.validateAlias(alias, request.index(), currentState.metaData()); } // we only find a template when its an API call (a new index) // find templates, highest order are better matching List<IndexTemplateMetaData> templates = findTemplates(request, currentState); Map<String, Custom> customs = new HashMap<>(); // add the request mapping Map<String, Map<String, Object>> mappings = new HashMap<>(); Map<String, AliasMetaData> templatesAliases = new HashMap<>(); List<String> templateNames = new ArrayList<>(); for (Map.Entry<String, String> entry : request.mappings().entrySet()) { mappings.put(entry.getKey(), MapperService.parseMapping(xContentRegistry, entry.getValue())); } for (Map.Entry<String, Custom> entry : request.customs().entrySet()) { customs.put(entry.getKey(), entry.getValue()); } final Index shrinkFromIndex = request.shrinkFrom(); if (shrinkFromIndex == null) { // apply templates, merging the mappings into the request mapping if exists for (IndexTemplateMetaData template : templates) { templateNames.add(template.getName()); for (ObjectObjectCursor<String, CompressedXContent> cursor : template.mappings()) { String mappingString = cursor.value.string(); if (mappings.containsKey(cursor.key)) { XContentHelper.mergeDefaults(mappings.get(cursor.key), MapperService.parseMapping(xContentRegistry, mappingString)); } else { mappings.put(cursor.key, MapperService.parseMapping(xContentRegistry, mappingString)); } } // handle custom for (ObjectObjectCursor<String, Custom> cursor : template.customs()) { String type = cursor.key; IndexMetaData.Custom custom = cursor.value; IndexMetaData.Custom existing = customs.get(type); if (existing == null) { customs.put(type, custom); } else { IndexMetaData.Custom merged = existing.mergeWith(custom); customs.put(type, merged); } } //handle aliases for (ObjectObjectCursor<String, AliasMetaData> cursor : template.aliases()) { AliasMetaData aliasMetaData = cursor.value; //if an alias with same name came with the create index request itself, // ignore this one taken from the index template if (request.aliases().contains(new Alias(aliasMetaData.alias()))) { continue; } //if an alias with same name was already processed, ignore this one if (templatesAliases.containsKey(cursor.key)) { continue; } //Allow templatesAliases to be templated by replacing a token with the name of the index that we are applying it to if (aliasMetaData.alias().contains("{index}")) { String templatedAlias = aliasMetaData.alias().replace("{index}", request.index()); aliasMetaData = AliasMetaData.newAliasMetaData(aliasMetaData, templatedAlias); } aliasValidator.validateAliasMetaData(aliasMetaData, request.index(), currentState.metaData()); templatesAliases.put(aliasMetaData.alias(), aliasMetaData); } } } Settings.Builder indexSettingsBuilder = Settings.builder(); if (shrinkFromIndex == null) { // apply templates, here, in reverse order, since first ones are better matching for (int i = templates.size() - 1; i >= 0; i--) { indexSettingsBuilder.put(templates.get(i).settings()); } } // now, put the request settings, so they override templates indexSettingsBuilder.put(request.settings()); if (indexSettingsBuilder.get(SETTING_NUMBER_OF_SHARDS) == null) { indexSettingsBuilder.put(SETTING_NUMBER_OF_SHARDS, settings.getAsInt(SETTING_NUMBER_OF_SHARDS, 5)); } if (indexSettingsBuilder.get(SETTING_NUMBER_OF_REPLICAS) == null) { indexSettingsBuilder.put(SETTING_NUMBER_OF_REPLICAS, settings.getAsInt(SETTING_NUMBER_OF_REPLICAS, 1)); } if (settings.get(SETTING_AUTO_EXPAND_REPLICAS) != null && indexSettingsBuilder.get(SETTING_AUTO_EXPAND_REPLICAS) == null) { indexSettingsBuilder.put(SETTING_AUTO_EXPAND_REPLICAS, settings.get(SETTING_AUTO_EXPAND_REPLICAS)); } if (indexSettingsBuilder.get(SETTING_VERSION_CREATED) == null) { DiscoveryNodes nodes = currentState.nodes(); final Version createdVersion = Version.min(Version.CURRENT, nodes.getSmallestNonClientNodeVersion()); indexSettingsBuilder.put(SETTING_VERSION_CREATED, createdVersion); } if (indexSettingsBuilder.get(SETTING_CREATION_DATE) == null) { indexSettingsBuilder.put(SETTING_CREATION_DATE, new DateTime(DateTimeZone.UTC).getMillis()); } indexSettingsBuilder.put(IndexMetaData.SETTING_INDEX_PROVIDED_NAME, request.getProvidedName()); indexSettingsBuilder.put(SETTING_INDEX_UUID, UUIDs.randomBase64UUID()); final IndexMetaData.Builder tmpImdBuilder = IndexMetaData.builder(request.index()); final int routingNumShards; if (shrinkFromIndex == null) { routingNumShards = IndexMetaData.INDEX_NUMBER_OF_SHARDS_SETTING.get(indexSettingsBuilder.build()); } else { final IndexMetaData sourceMetaData = currentState.metaData().getIndexSafe(shrinkFromIndex); routingNumShards = sourceMetaData.getRoutingNumShards(); } tmpImdBuilder.setRoutingNumShards(routingNumShards); if (shrinkFromIndex != null) { prepareShrinkIndexSettings( currentState, mappings.keySet(), indexSettingsBuilder, shrinkFromIndex, request.index()); } final Settings actualIndexSettings = indexSettingsBuilder.build(); tmpImdBuilder.settings(actualIndexSettings); // Set up everything, now locally create the index to see that things are ok, and apply final IndexMetaData tmpImd = tmpImdBuilder.build(); ActiveShardCount waitForActiveShards = request.waitForActiveShards(); if (waitForActiveShards == ActiveShardCount.DEFAULT) { waitForActiveShards = tmpImd.getWaitForActiveShards(); } if (waitForActiveShards.validate(tmpImd.getNumberOfReplicas()) == false) { throw new IllegalArgumentException("invalid wait_for_active_shards[" + request.waitForActiveShards() + "]: cannot be greater than number of shard copies [" + (tmpImd.getNumberOfReplicas() + 1) + "]"); } // create the index here (on the master) to validate it can be created, as well as adding the mapping final IndexService indexService = indicesService.createIndex(tmpImd, Collections.emptyList()); createdIndex = indexService.index(); // now add the mappings MapperService mapperService = indexService.mapperService(); try { mapperService.merge(mappings, MergeReason.MAPPING_UPDATE, request.updateAllTypes()); } catch (Exception e) { removalExtraInfo = "failed on parsing default mapping/mappings on index creation"; throw e; } // the context is only used for validation so it's fine to pass fake values for the shard id and the current // timestamp final QueryShardContext queryShardContext = indexService.newQueryShardContext(0, null, () -> 0L); for (Alias alias : request.aliases()) { if (Strings.hasLength(alias.filter())) { aliasValidator.validateAliasFilter(alias.name(), alias.filter(), queryShardContext, xContentRegistry); } } for (AliasMetaData aliasMetaData : templatesAliases.values()) { if (aliasMetaData.filter() != null) { aliasValidator.validateAliasFilter(aliasMetaData.alias(), aliasMetaData.filter().uncompressed(), queryShardContext, xContentRegistry); } } // now, update the mappings with the actual source Map<String, MappingMetaData> mappingsMetaData = new HashMap<>(); for (DocumentMapper mapper : mapperService.docMappers(true)) { MappingMetaData mappingMd = new MappingMetaData(mapper); mappingsMetaData.put(mapper.type(), mappingMd); } final IndexMetaData.Builder indexMetaDataBuilder = IndexMetaData.builder(request.index()) .settings(actualIndexSettings) .setRoutingNumShards(routingNumShards); for (int shardId = 0; shardId < tmpImd.getNumberOfShards(); shardId++) { indexMetaDataBuilder.primaryTerm(shardId, tmpImd.primaryTerm(shardId)); } for (MappingMetaData mappingMd : mappingsMetaData.values()) { indexMetaDataBuilder.putMapping(mappingMd); } for (AliasMetaData aliasMetaData : templatesAliases.values()) { indexMetaDataBuilder.putAlias(aliasMetaData); } for (Alias alias : request.aliases()) { AliasMetaData aliasMetaData = AliasMetaData.builder(alias.name()).filter(alias.filter()) .indexRouting(alias.indexRouting()).searchRouting(alias.searchRouting()).build(); indexMetaDataBuilder.putAlias(aliasMetaData); } for (Map.Entry<String, Custom> customEntry : customs.entrySet()) { indexMetaDataBuilder.putCustom(customEntry.getKey(), customEntry.getValue()); } indexMetaDataBuilder.state(request.state()); final IndexMetaData indexMetaData; try { indexMetaData = indexMetaDataBuilder.build(); } catch (Exception e) { removalExtraInfo = "failed to build index metadata"; throw e; } indexService.getIndexEventListener().beforeIndexAddedToCluster(indexMetaData.getIndex(), indexMetaData.getSettings()); MetaData newMetaData = MetaData.builder(currentState.metaData()) .put(indexMetaData, false) .build(); String maybeShadowIndicator = IndexMetaData.isIndexUsingShadowReplicas(indexMetaData.getSettings()) ? "s" : ""; logger.info("[{}] creating index, cause [{}], templates {}, shards [{}]/[{}{}], mappings {}", request.index(), request.cause(), templateNames, indexMetaData.getNumberOfShards(), indexMetaData.getNumberOfReplicas(), maybeShadowIndicator, mappings.keySet()); ClusterBlocks.Builder blocks = ClusterBlocks.builder().blocks(currentState.blocks()); if (!request.blocks().isEmpty()) { for (ClusterBlock block : request.blocks()) { blocks.addIndexBlock(request.index(), block); } } blocks.updateBlocks(indexMetaData); ClusterState updatedState = ClusterState.builder(currentState).blocks(blocks).metaData(newMetaData).build(); if (request.state() == State.OPEN) { RoutingTable.Builder routingTableBuilder = RoutingTable.builder(updatedState.routingTable()) .addAsNew(updatedState.metaData().index(request.index())); updatedState = allocationService.reroute( ClusterState.builder(updatedState).routingTable(routingTableBuilder.build()).build(), "index [" + request.index() + "] created"); } removalExtraInfo = "cleaning up after validating index on master"; removalReason = IndexRemovalReason.NO_LONGER_ASSIGNED; return updatedState; } finally { if (createdIndex != null) { // Index was already partially created - need to clean up indicesService.removeIndex(createdIndex, removalReason, removalExtraInfo); } } } @Override public void onFailure(String source, Exception e) { if (e instanceof ResourceAlreadyExistsException) { logger.trace((Supplier<?>) () -> new ParameterizedMessage("[{}] failed to create", request.index()), e); } else { logger.debug((Supplier<?>) () -> new ParameterizedMessage("[{}] failed to create", request.index()), e); } super.onFailure(source, e); } });
這個任務的執行內容在public ClusterState execute(ClusterState currentState) throws Exception中。其內容如下：
索引名稱校驗
- 名稱不能包含#
- 不能以_、-、+開頭
- 位元組長度不大於255
- 不能是.或…
- 要求小寫
- 是否已經存在
- 不能存在同名的別名
裝載索引的setting、mapping配置
將主分片ID載入到索引元資料中
for (int shardId = 0; shardId < tmpImd.getNumberOfShards(); shardId++) { indexMetaDataBuilder.primaryTerm(shardId, tmpImd.primaryTerm(shardId)); }
將mapping資訊也載入到索引元資料中
for (MappingMetaData mappingMd : mappingsMetaData.values()) {
indexMetaDataBuilder.putMapping(mappingMd);
}
將別名資訊載入到元資料中
for (AliasMetaData aliasMetaData : templatesAliases.values()) {
indexMetaDataBuilder.putAlias(aliasMetaData);
}
for (Alias alias : request.aliases()) {
AliasMetaData aliasMetaData = AliasMetaData.builder(alias.name()).filter(alias.filter())
.indexRouting(alias.indexRouting()).searchRouting(alias.searchRouting()).build();
indexMetaDataBuilder.putAlias(aliasMetaData);
}
將所有元資料資訊更新到叢集元資料資訊中
MetaData newMetaData = MetaData.builder(currentState.metaData())
.put(indexMetaData, false)
.build();
為新分片分配節點
protected ClusterState reroute(final ClusterState clusterState, String reason, boolean debug) {
RoutingNodes routingNodes = getMutableRoutingNodes(clusterState);
// shuffle the unassigned nodes, just so we won't have things like poison failed shards
routingNodes.unassigned().shuffle();
RoutingAllocation allocation = new RoutingAllocation(allocationDeciders, routingNodes, clusterState,
clusterInfoService.getClusterInfo(), currentNanoTime(), false);
allocation.debugDecision(debug);
reroute(allocation);
if (allocation.routingNodesChanged() == false) {
return clusterState;
}
return buildResultAndLogHealthChange(clusterState, allocation, reason);
}
分配演算法：
weight=0.45*（當前節點擁有的分片數+1-平均每個節點的分片數（包括未分配的分片數））+0.55*（當前節點下該索引擁有的分片數+1-平均每個節點下該索引的分片數（包括未分配的分片數））
取weight最小的節點作為該分片分配的節點
將元資料的變化生效
前一步僅僅是在元資料中增加新索引的資訊，而這些資訊只保留在記憶體中，並沒有在磁碟中建立對應的資料。這一步比較重要的就是分片在磁碟的分配，請參考ES5.6.4原始碼解析----分片在磁碟間的分配策略