dlg99 commented on a change in pull request #9927: URL: https://github.com/apache/pulsar/pull/9927#discussion_r597269231
########## File path: pulsar-io/kafka-connect-adaptor/src/main/java/org/apache/pulsar/io/kafka/connect/KafkaConnectSink.java ########## @@ -0,0 +1,277 @@ +/** + * 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.pulsar.io.kafka.connect; + +import com.google.common.annotations.VisibleForTesting; +import com.google.common.collect.ImmutableMap; +import com.google.common.collect.Lists; +import com.google.common.collect.Maps; +import com.google.common.util.concurrent.ThreadFactoryBuilder; +import lombok.extern.slf4j.Slf4j; +import org.apache.kafka.clients.consumer.OffsetAndMetadata; +import org.apache.kafka.common.TopicPartition; +import org.apache.kafka.common.record.TimestampType; +import org.apache.kafka.connect.connector.Task; +import org.apache.kafka.connect.data.Schema; +import org.apache.kafka.connect.sink.SinkConnector; +import org.apache.kafka.connect.sink.SinkRecord; +import org.apache.kafka.connect.sink.SinkTask; +import org.apache.pulsar.functions.api.Record; +import org.apache.pulsar.io.core.KeyValue; +import org.apache.pulsar.io.core.Sink; +import org.apache.pulsar.io.core.SinkContext; + +import java.util.List; +import java.util.Map; +import java.util.Objects; +import java.util.Properties; +import java.util.concurrent.CompletableFuture; +import java.util.concurrent.Executors; +import java.util.concurrent.ScheduledExecutorService; +import java.util.concurrent.TimeUnit; +import java.util.concurrent.atomic.AtomicInteger; + +import static org.apache.pulsar.io.kafka.connect.PulsarKafkaWorkerConfig.OFFSET_STORAGE_TOPIC_CONFIG; +import static org.apache.pulsar.io.kafka.connect.PulsarKafkaWorkerConfig.PULSAR_SERVICE_URL_CONFIG; + +@Slf4j +public class KafkaConnectSink implements Sink<Object> { + + private boolean unwrapKeyValueIfAvailable; + + private final static ImmutableMap<Class<?>, Schema> primitiveTypeToSchema; + static { + primitiveTypeToSchema = ImmutableMap.<Class<?>, Schema>builder() + .put(Boolean.class, Schema.BOOLEAN_SCHEMA) + .put(Byte.class, Schema.INT8_SCHEMA) + .put(Short.class, Schema.INT16_SCHEMA) + .put(Integer.class, Schema.INT32_SCHEMA) + .put(Long.class, Schema.INT64_SCHEMA) + .put(Float.class, Schema.FLOAT32_SCHEMA) + .put(Double.class, Schema.FLOAT64_SCHEMA) + .put(String.class, Schema.STRING_SCHEMA) + .put(byte[].class, Schema.BYTES_SCHEMA) + .build(); + } + + private PulsarKafkaSinkContext sinkContext; + private PulsarKafkaSinkTaskContext taskContext; + private SinkConnector connector; + private SinkTask task; + + private Schema defaultKeySchema; + private Schema defaultValueSchema; + + + private int batchSize; + private long lingerMs; + private final ScheduledExecutorService scheduledExecutor = + Executors.newSingleThreadScheduledExecutor(new ThreadFactoryBuilder() + .setNameFormat("pulsar-io-kafka-adaptor-sink-flush-%d") + .build()); + private final AtomicInteger numPendingRecords = new AtomicInteger(0); + + private volatile CompletableFuture<Void> pendingFlush = new CompletableFuture<>(); + private volatile boolean isRunning = false; + + private Properties props = new Properties(); + private PulsarKafkaConnectSinkConfig kafkaSinkConfig; + + protected String topicName; + + @Override + public void write(Record<Object> sourceRecord) { + if (log.isDebugEnabled()) { + log.debug("Record sending to kafka, record={}.", sourceRecord); + } + + if (!isRunning) { + log.error("Sink is stopped. Cannot send the record {}", sourceRecord); + sourceRecord.fail(); + return; + } + + try { + SinkRecord record = toSinkRecord(sourceRecord); + task.put(Lists.newArrayList(record)); + } catch (Exception ex) { + log.error("Error sending the record {}", sourceRecord, ex); + sourceRecord.fail(); + return; + } + pendingFlush.whenComplete((ignore, ex) -> { + if (ex == null) { + sourceRecord.ack(); + } else { + log.error("Error sending the record {}", sourceRecord, ex); + sourceRecord.fail(); + } + throw new IllegalArgumentException(); + }); + numPendingRecords.incrementAndGet(); + flushIfNeeded(false); + } + + @Override + public void close() throws Exception { + isRunning = false; + flushIfNeeded(true); + scheduledExecutor.shutdown(); + if (!scheduledExecutor.awaitTermination(10 * lingerMs, TimeUnit.MILLISECONDS)) { + log.error("scheduledExecutor did not terminate in {} ms", 10 * lingerMs); + } + + task.stop(); + connector.stop(); + taskContext.close(); + + log.info("Kafka sink stopped."); + } + + @Override + public void open(Map<String, Object> config, SinkContext ctx) throws Exception { + kafkaSinkConfig = PulsarKafkaConnectSinkConfig.load(config); + Objects.requireNonNull(kafkaSinkConfig.getTopic(), "Kafka topic is not set"); + topicName = kafkaSinkConfig.getTopic(); + unwrapKeyValueIfAvailable = kafkaSinkConfig.isUnwrapKeyValueIfAvailable(); + + String kafkaConnectorFQClassName = kafkaSinkConfig.getKafkaConnectorSinkClass(); + kafkaSinkConfig.getKafkaConnectorConfigProperties().entrySet() + .forEach(kv -> props.put(kv.getKey(), kv.getValue())); + + defaultKeySchema = (Schema)Schema.class + .getField(kafkaSinkConfig.getDefaultKeySchema()).get(null); + defaultValueSchema = (Schema)Schema.class + .getField(kafkaSinkConfig.getDefaultValueSchema()).get(null); + + Class<?> clazz = Class.forName(kafkaConnectorFQClassName); + connector = (SinkConnector) clazz.getConstructor().newInstance(); + + Class<? extends Task> taskClass = connector.taskClass(); + sinkContext = new PulsarKafkaSinkContext(); + connector.initialize(sinkContext); + connector.start(Maps.fromProperties(props)); + + List<Map<String, String>> configs = connector.taskConfigs(1); + configs.forEach(x -> { + x.put(OFFSET_STORAGE_TOPIC_CONFIG, kafkaSinkConfig.getOffsetStorageTopic()); + x.put(PULSAR_SERVICE_URL_CONFIG, kafkaSinkConfig.getPulsarServiceUrl()); + }); + task = (SinkTask) taskClass.getConstructor().newInstance(); + taskContext = + new PulsarKafkaSinkTaskContext(configs.get(0), task::open); + task.initialize(taskContext); + task.start(configs.get(0)); + + batchSize = kafkaSinkConfig.getBatchSize(); + lingerMs = kafkaSinkConfig.getLingerTimeMs(); + scheduledExecutor.scheduleAtFixedRate(() -> + this.flushIfNeeded(true), lingerMs, lingerMs, TimeUnit.MILLISECONDS); + + isRunning = true; + log.info("Kafka sink started : {}.", props); + } + + private void flushIfNeeded(boolean force) { + if (force || numPendingRecords.get() >= batchSize) { + scheduledExecutor.submit(this::flush); + } + } + + public void flush() { + if (log.isDebugEnabled()) { + log.debug("flush requested, pending: {}, batchSize: {}", + numPendingRecords.get(), batchSize); + } + + if (numPendingRecords.getAndSet(0) == 0) { + return; + } + + Map<TopicPartition, OffsetAndMetadata> currentOffsets = taskContext.currentOffsets(); + CompletableFuture<Void> flushCf; + synchronized (this) { + flushCf = pendingFlush; + pendingFlush = new CompletableFuture<>(); + } + + try { + task.flush(currentOffsets); + taskContext.flushOffsets(currentOffsets); + flushCf.complete(null); + } catch (Throwable t) { + log.error("error flushing pending records", t); + flushCf.completeExceptionally(t); + } + } + + /** + * org.apache.kafka.connect.data.Schema for the object + * @param obj + * @return org.apache.kafka.connect.data.Schema + */ + public static Schema getKafkaConnectSchemaForObject(Object obj, Schema defaultSchema) { + if (obj == null) { + return defaultSchema; + } + + if (primitiveTypeToSchema.containsKey(obj.getClass())) { + return primitiveTypeToSchema.get(obj.getClass()); + } + + // Other types are not supported yet. + // Will fallback to defaults provided. + return defaultSchema; + } + + private SinkRecord toSinkRecord(Record<Object> sourceRecord) { + final int partition = 0; + final Object key; + final Object value; + if (unwrapKeyValueIfAvailable && sourceRecord.getValue() instanceof KeyValue) { + KeyValue<Object, Object> kv = (KeyValue<Object, Object>) sourceRecord.getValue(); + key = kv.getKey(); + value = kv.getValue(); + } else { + key = sourceRecord.getKey().orElse(null); + value = sourceRecord.getValue(); + } + final Schema keySchema = getKafkaConnectSchemaForObject(key, defaultKeySchema); + final Schema valueSchema = getKafkaConnectSchemaForObject(value, defaultValueSchema); + + long offset = taskContext.currentOffset(topicName, partition).incrementAndGet(); Review comment: @sijie The Record<> may or may not contain sequenceID (Optional) sequenceID, at the best, is `Optional.of(FunctionCommon.getSequenceId(message.getMessageId()))` - as i understand this applies to persistent topics only. Record may or may not contain Message (Optional): sourceRecord.getMessage() which may return sequenceID or -1 (kafka has a thing against negative offsets, SinkRecord does not allow them). MessageId can return byte[] (sourceRecord.getMessage().get().getMessageId().toByteArray()) of something - not guaranteed to be offset, rather a byte[] of anything, i.e. ledgerId, entryId, batchSize, topicName etc. Or it can throw UnsupportedOperationException AFAICT usable sequenceID can be retrieved from data produced by kafka source (and there is absolutely no point in connecting kafka source to kafka sink in pulsar) or from the data read from bookkeeper. I can do something along the lines of ``` long offset = sourceRecord.getRecordSequence() .orElse(taskContext.currentOffset(topicName, partition).incrementAndGet()); ``` This assumes that topic either has sequenceID for all records, or does not have them for all records. Next problem is that kafka's offsets increase monotonically (per partition). In case of BK, offsets is `(ledgerId << 28) | entryId` and is not guaranteed to be sequential, as I understand. I don't know how kafka sinks react to that. Some sinks ignore offsets, others may track specific invariants. Simply assigning the offsets the way I do now is the simplest solution. It won't allow Exactly-once processing if the kafka sink deduping/overwriting data by offset id (and not the key, and if it supports this) but it will work as AtLeast-once. I.e. if everything restarts it won't re-assign the same offsets (so deduping sink won't overwrite older data) but in case of replay/retry it will assign new offsets to the data potentially processed by the sink. What do you hink? -- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
