cmccabe commented on code in PR #12776: URL: https://github.com/apache/kafka/pull/12776#discussion_r1011337211
########## metadata/src/main/java/org/apache/kafka/image/loader/MetadataLoader.java: ########## @@ -0,0 +1,504 @@ +/* + * 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.kafka.image.loader; + +import org.apache.kafka.common.utils.LogContext; +import org.apache.kafka.common.utils.Time; +import org.apache.kafka.image.MetadataDelta; +import org.apache.kafka.image.MetadataImage; +import org.apache.kafka.image.MetadataProvenance; +import org.apache.kafka.image.MetadataVersionChange; +import org.apache.kafka.image.MetadataVersionChangeException; +import org.apache.kafka.image.publisher.MetadataPublisher; +import org.apache.kafka.image.writer.ImageReWriter; +import org.apache.kafka.image.writer.ImageWriterOptions; +import org.apache.kafka.queue.KafkaEventQueue; +import org.apache.kafka.raft.Batch; +import org.apache.kafka.raft.BatchReader; +import org.apache.kafka.raft.LeaderAndEpoch; +import org.apache.kafka.raft.RaftClient; +import org.apache.kafka.server.common.ApiMessageAndVersion; +import org.apache.kafka.server.fault.FaultHandler; +import org.apache.kafka.snapshot.SnapshotReader; +import org.slf4j.Logger; + +import java.util.ArrayList; +import java.util.List; +import java.util.concurrent.CompletableFuture; + + +/** + * The MetadataLoader follows changes provided by a RaftClient, and packages them into metadata + * deltas and images that can be consumed by publishers. + * + * The Loader maintains its own thread, which is used to make all callbacks into publishers. If a + * publisher A is installed before B, A will receive all callbacks before B. This is also true if + * A and B are installed as part of a list [A, B]. + * + * Publishers should not modify any data structures passed to them. + * + * It is possible to change the list of publishers dynamically over time. Whenever a new publisher is + * added, it receives a catch-up delta which contains the full state. Any publisher installed when the + * loader is closed will itself be closed. + */ +public class MetadataLoader implements RaftClient.Listener<ApiMessageAndVersion>, AutoCloseable { + public static class Builder { + private int nodeId = -1; + private Time time = Time.SYSTEM; + private LogContext logContext = null; + private String threadNamePrefix = ""; + private FaultHandler faultHandler = null; + private MetadataLoaderMetrics metrics = new MetadataLoaderMetrics() { + private volatile long lastAppliedOffset = -1L; + + @Override + public void updateBatchProcessingTime(long elapsedNs) { } + + @Override + public void updateBatchSize(int size) { } + + @Override + public void updateLastAppliedImageProvenance(MetadataProvenance provenance) { + this.lastAppliedOffset = provenance.offset(); + } + + @Override + public long lastAppliedOffset() { + return lastAppliedOffset; + } + + @Override + public void close() throws Exception { } + }; + + public Builder setNodeId(int nodeId) { + this.nodeId = nodeId; + return this; + } + + public Builder setTime(Time time) { + this.time = time; + return this; + } + + public Builder setThreadNamePrefix(String threadNamePrefix) { + this.threadNamePrefix = threadNamePrefix; + return this; + } + + public Builder setFaultHandler(FaultHandler faultHandler) { + this.faultHandler = faultHandler; + return this; + } + + public Builder setMetadataLoaderMetrics(MetadataLoaderMetrics metrics) { + this.metrics = metrics; + return this; + } + + public MetadataLoader build() { + if (logContext == null) { + logContext = new LogContext("[MetadataLoader " + nodeId + "] "); + } + if (faultHandler == null) throw new RuntimeException("You must set a fault handler."); + return new MetadataLoader( + time, + logContext, + threadNamePrefix, + faultHandler, + metrics); + } + } + + /** + * The log4j logger for this loader. + */ + private final Logger log; + + /** + * The clock used by this loader. + */ + private final Time time; + + /** + * The fault handler to use if metadata loading fails. + */ + private final FaultHandler faultHandler; + + /** + * Callbacks for updating metrics. + */ + private final MetadataLoaderMetrics metrics; + + /** + * The publishers which should receive cluster metadata updates. + */ + private final List<MetadataPublisher> publishers; + + /** + * The current leader and epoch. + */ + private LeaderAndEpoch currentLeaderAndEpoch = LeaderAndEpoch.UNKNOWN; + + /** + * The current metadata image. Accessed only from the event queue thread. + */ + private MetadataImage image; + + /** + * The event queue which runs this loader. + */ + private final KafkaEventQueue eventQueue; + + private MetadataLoader( + Time time, + LogContext logContext, + String threadNamePrefix, + FaultHandler faultHandler, + MetadataLoaderMetrics metrics + ) { + this.log = logContext.logger(MetadataLoader.class); + this.time = time; + this.faultHandler = faultHandler; + this.metrics = metrics; + this.publishers = new ArrayList<>(); + this.image = MetadataImage.EMPTY; + this.eventQueue = new KafkaEventQueue(time, logContext, threadNamePrefix); + } + + @Override + public void handleCommit(BatchReader<ApiMessageAndVersion> reader) { + eventQueue.append(() -> { + try { + MetadataDelta delta = new MetadataDelta.Builder(). + setImage(image). + build(); + LogDeltaManifest manifest = loadLogDelta(delta, reader); + try { + image = delta.apply(manifest.provenance()); + } catch (Throwable e) { + faultHandler.handleFault("Error generating new metadata image from " + + "metadata delta between offset " + image.offset() + + " and " + manifest.provenance().offset(), e); + return; + } + for (MetadataPublisher publisher : publishers) { + try { + publisher.publishLogDelta(delta, image, manifest); + } catch (Throwable e) { + faultHandler.handleFault("Unhandled error publishing the new metadata " + + "image ending at " + manifest.provenance().offset() + + " with publisher " + publisher.name(), e); + } + } + metrics.updateLastAppliedImageProvenance(image.provenance()); + } catch (Throwable e) { + // This is a general catch-all block where we don't expect to end up; + // failure-prone operations should have individual try/catch blocks around them. + faultHandler.handleFault("Unhandled fault in MetadataLoader#handleCommit. " + + "Last image offset was " + image.offset(), e); + } finally { + reader.close(); + } + }); + } + + /** + * Load some batches of records from the log. We have to do some bookkeeping here to + * translate between batch offsets and record offsets, and track the number of bytes we + * have read. Additionally, there is the chance that one of the records is a metadata + * version change which needs to be handled differently. + * + * @param delta The metadata delta we are preparing. + * @param reader The reader which yields the batches. + * @return A manifest of what was loaded. + */ + LogDeltaManifest loadLogDelta( + MetadataDelta delta, + BatchReader<ApiMessageAndVersion> reader + ) { + long startNs = time.nanoseconds(); + int numBatches = 0; + long numBytes = 0L; + long lastOffset = image.provenance().offset(); + int lastEpoch = image.provenance().epoch(); + long lastContainedLogTimeMs = image.provenance().lastContainedLogTimeMs(); + + while (reader.hasNext()) { + Batch<ApiMessageAndVersion> batch = reader.next(); + int indexWithinBatch = 0; + for (ApiMessageAndVersion record : batch.records()) { + try { + delta.replay(record.message()); + } catch (MetadataVersionChangeException e) { + handleMetadataVersionChange(delta, + e.change(), + batch.baseOffset() + indexWithinBatch, + lastEpoch, + lastContainedLogTimeMs); + } catch (Throwable e) { + faultHandler.handleFault("Error loading metadata log record from offset " + + batch.baseOffset() + indexWithinBatch, e); + } + lastEpoch = batch.epoch(); + lastContainedLogTimeMs = batch.appendTimestamp(); + indexWithinBatch++; + } + metrics.updateBatchSize(batch.records().size()); + lastOffset = batch.lastOffset(); + numBytes += batch.sizeInBytes(); + numBatches++; + } + MetadataProvenance provenance = + new MetadataProvenance(lastOffset, lastEpoch, lastContainedLogTimeMs); + long elapsedNs = time.nanoseconds() - startNs; + // TODO: this metric should be renamed something like "delta processing time" + metrics.updateBatchProcessingTime(elapsedNs); + return new LogDeltaManifest(provenance, + numBatches, + elapsedNs, + numBytes); + } + + /** + * Handle a change in the metadata version. + * + * @param delta The metadata delta we're working with. + * @param change The change we're handling. + * @param changeOffset The offset of the change. + * @param preChangeEpoch The log epoch BEFORE the change. + */ + void handleMetadataVersionChange( + MetadataDelta delta, + MetadataVersionChange change, + long changeOffset, + int preChangeEpoch, + long lastContainedLogTimeMs + ) { + // First, we materialize the current metadata image and send it to all the publishers that Review Comment: The issue is that different publishers may be installed at different times. For example, let's say you have some hypothetical cluster balancing service that can be dynamically enabled. (not included in this PR, of course). When it's dynamically enabled, it registers as a metadata publisher, to get all the updates. If you enable this thing well after startup, you will need to send it a MetadataDelta that contains all the updates (since all the updates are new to it, although not to the existing publishers). How do you get such a MetadataDelta? Write out the existing image to a delta. Basically this PR adds dynamic installation / removal to the system, which it didn't previously have. That's why the old code locks it to a single publisher. But that isn't very dynamic and we have some use cases for installing things at runtime. -- 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]
