stream2000 commented on code in PR #9651: URL: https://github.com/apache/hudi/pull/9651#discussion_r1338115806
########## hudi-timeline-service/src/main/java/org/apache/hudi/timeline/service/handlers/InstantStateHandler.java: ########## @@ -0,0 +1,131 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF 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.apache.hudi.timeline.service.handlers; + +import org.apache.hudi.common.table.timeline.dto.InstantStateDTO; +import org.apache.hudi.common.table.view.FileSystemViewManager; +import org.apache.hudi.exception.HoodieIOException; +import org.apache.hudi.timeline.service.TimelineService; + +import org.apache.hadoop.conf.Configuration; +import org.apache.hadoop.fs.FileSystem; +import org.apache.hadoop.fs.Path; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.io.IOException; +import java.util.Arrays; +import java.util.Collections; +import java.util.List; +import java.util.concurrent.ConcurrentHashMap; +import java.util.concurrent.atomic.AtomicLong; +import java.util.stream.Collectors; + +/** + * REST Handler servicing instant state requests. + * <p> + * The instant states are cached in timeline server and will be refreshed after the instant states in file system were changed. + */ +public class InstantStateHandler extends Handler { + + private static final Logger LOG = LoggerFactory.getLogger(InstantStateHandler.class); + + /** + * Base url for instant state requests. + */ + private static final String BASE_URL = "/v1/hoodie/instantstate"; + + /** + * Param for instant state requests, which contains a uniqueId for different writers. + */ + public static final String INSTANT_STATE_DIR_PATH_PARAM = "instantstatedirpath"; + + /** + * GET requests. Returns all the instant states under instant state path. + */ + public static final String ALL_INSTANT_STATE_URL = String.format("%s/%s", BASE_URL, "all"); + + /** + * POST requests. Refresh the instant state data cached in memory. + */ + public static final String REFRESH_INSTANT_STATE = String.format("%s/%s", BASE_URL, "refresh/"); + + /** + * Cached instant state data, instant state path -> list of instant states in fs. + */ + private final ConcurrentHashMap<String, List<InstantStateDTO>> cachedInstantStates; + + /** + * Number of requests after the last refresh. + */ + private final AtomicLong requestCount; + + public InstantStateHandler(Configuration conf, TimelineService.Config timelineServiceConfig, FileSystem fileSystem, + FileSystemViewManager viewManager) throws IOException { + super(conf, timelineServiceConfig, fileSystem, viewManager); + this.cachedInstantStates = new ConcurrentHashMap<>(); + this.requestCount = new AtomicLong(); + } + + /** + * Read instant states from cache of file system. + * + * @return Instant states under the input instant state path. + */ + public List<InstantStateDTO> getAllInstantStates(String instantStatePath) { + if (requestCount.incrementAndGet() >= timelineServiceConfig.instantStateForceRefreshRequestNumber) { + // Do refresh for every N requests to ensure the writers won't be blocked forever + refresh(instantStatePath); + } + return cachedInstantStates.computeIfAbsent(instantStatePath, k -> scanInstantState(new Path(k))); + } + + /** + * Refresh the checkpoint messages cached. Will be called when coordinator start/commit/abort instant. + * + * @return Whether refreshing is successful. + */ + public boolean refresh(String instantStatePath) { + try { + cachedInstantStates.put(instantStatePath, scanInstantState(new Path(instantStatePath))); + requestCount.set(0); + } catch (Exception e) { + LOG.error("Failed to load instant states, path: " + instantStatePath, e); + return false; + } + return true; + } + + /** + * Scan the instant states from file system. + */ + public List<InstantStateDTO> scanInstantState(Path instantStatePath) { + try { + // Check instantStatePath exists before list status, see HUDI-5915 + if (this.fileSystem.exists(instantStatePath)) { + return Arrays.stream(this.fileSystem.listStatus(instantStatePath)).map(InstantStateDTO::fromFileStatus).collect(Collectors.toList()); Review Comment: You are right, we can implement a complete timeline server memory-based checkpoint metadata since we will delete all previous ckp metadata when bootstrapping(so we do not actually rely on a durable filesystem state). And I think the timeline server is actually stable enough to serve the checkpoint message requests. For now, I think it's OK to keep the filesystem-backed checkpoint messages and make the timeline server a stateless reader for the following reasons: 1. We will have a safe fallback option when the timeline server is not available during runtime. 2. The filesystem-backed checkpoint code is still useful since we need to turn off the timeline server sometimes. And creating ckp messages files is not expensive, we can reuse the code. 3. There will be no conflict if we want to move the start/abort/commit request to the timeline server in the future, and this PR has already implemented the important optimization that reduces the massive file listing. Moving the write part to the timeline server needs extra code and some more careful tests, we can move this PR. We can merge this PR and create a Jira to track any possible optimization, what do you think? And thanks again for your review~ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
