adapter增加同步异常重试

This commit is contained in:
mcy
2018-12-23 18:16:47 +08:00
parent 34b5f9c927
commit 354d64069f
8 changed files with 233 additions and 192 deletions

View File

@@ -2,11 +2,7 @@ package com.alibaba.otter.canal.adapter.launcher.loader;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.*;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
@@ -80,16 +76,17 @@ public abstract class AbstractCanalAdapterWorker {
futures.forEach(future -> {
try {
if (!future.get()) {
logger.error("Outer adapter write failed");
throw new RuntimeException("Outer adapter sync failed! ");
}
} catch (InterruptedException | ExecutionException e) {
// ignore
} catch (Exception e) {
future.cancel(true);
throw new RuntimeException(e);
}
});
});
}
private void writeOut(final List<FlatMessage> flatMessages) {
protected void writeOut(final List<FlatMessage> flatMessages) {
List<Future<Boolean>> futures = new ArrayList<>();
// 组间适配器并行运行
canalOuterAdapters.forEach(outerAdapters -> {
@@ -119,63 +116,64 @@ public abstract class AbstractCanalAdapterWorker {
futures.forEach(future -> {
try {
if (!future.get()) {
logger.error("Outer adapter write failed");
throw new RuntimeException("Outer adapter sync failed! ");
}
} catch (InterruptedException | ExecutionException e) {
// ignore
} catch (Exception e) {
future.cancel(true);
throw new RuntimeException(e);
}
});
});
}
@SuppressWarnings("unchecked")
protected void mqWriteOutData(int retry, long timeout, final boolean flatMessage, CanalMQConnector connector,
protected boolean mqWriteOutData(int retry, long timeout, int i, final boolean flatMessage, CanalMQConnector connector,
ExecutorService workerExecutor) {
for (int i = 0; i < retry; i++) {
try {
List<?> messages;
if (!flatMessage) {
messages = connector.getListWithoutAck(100L, TimeUnit.MILLISECONDS);
} else {
messages = connector.getFlatListWithoutAck(100L, TimeUnit.MILLISECONDS);
}
if (messages != null) {
Future<Boolean> future = workerExecutor.submit(() -> {
if (flatMessage) {
// batch write
writeOut((List<FlatMessage>) messages);
} else {
for (final Object message : messages) {
writeOut((Message) message);
}
try {
List<?> messages;
if (!flatMessage) {
messages = connector.getListWithoutAck(100L, TimeUnit.MILLISECONDS);
} else {
messages = connector.getFlatListWithoutAck(100L, TimeUnit.MILLISECONDS);
}
if (messages != null && !messages.isEmpty()) {
Future<Boolean> future = workerExecutor.submit(() -> {
if (flatMessage) {
// batch write
writeOut((List<FlatMessage>) messages);
} else {
for (final Object message : messages) {
writeOut((Message) message);
}
return true;
});
try {
future.get(timeout, TimeUnit.MILLISECONDS);
} catch (Exception e) {
future.cancel(true);
throw e;
}
return true;
});
try {
future.get(timeout, TimeUnit.MILLISECONDS);
} catch (Exception e) {
future.cancel(true);
throw e;
}
connector.ack();
break;
} catch (Throwable e) {
if (i == retry - 1) {
connector.ack();
} else {
connector.rollback();
}
logger.error(e.getMessage(), e);
try {
TimeUnit.SECONDS.sleep(1L);
} catch (InterruptedException e1) {
// ignore
}
}
return true;
} catch (Throwable e) {
if (i == retry - 1) {
connector.ack();
logger.error(e.getMessage() + " Error sync but ACK!");
return true;
} else {
connector.rollback();
logger.error(e.getMessage() + " Error sync and rollback, execute times: " + (i + 1));
}
try {
Thread.sleep(500);
} catch (InterruptedException e1) {
// ignore
}
}
return false;
}
/**

View File

@@ -3,8 +3,11 @@ package com.alibaba.otter.canal.adapter.launcher.loader;
import java.util.List;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import com.alibaba.otter.canal.protocol.FlatMessage;
import com.alibaba.otter.canal.protocol.Message;
import org.apache.kafka.common.errors.WakeupException;
import com.alibaba.otter.canal.client.adapter.OuterAdapter;
@@ -61,7 +64,17 @@ public class CanalAdapterKafkaWorker extends AbstractCanalAdapterWorker {
connector.disconnect();
break;
}
mqWriteOutData(retry, timeout, flatMessage, connector, workerExecutor);
if (retry == -1) {
retry = Integer.MAX_VALUE;
}
for (int i = 0; i < retry; i++) {
if (!running) {
break;
}
if (mqWriteOutData(retry, timeout, i, flatMessage, connector, workerExecutor)) {
break;
}
}
}
} catch (Exception e) {
logger.error(e.getMessage(), e);

View File

@@ -3,7 +3,11 @@ package com.alibaba.otter.canal.adapter.launcher.loader;
import java.util.List;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import com.alibaba.otter.canal.protocol.FlatMessage;
import com.alibaba.otter.canal.protocol.Message;
import org.apache.kafka.common.errors.WakeupException;
import com.alibaba.otter.canal.client.adapter.OuterAdapter;
@@ -39,7 +43,8 @@ public class CanalAdapterRocketMQWorker extends AbstractCanalAdapterWorker {
;
ExecutorService workerExecutor = Executors.newSingleThreadExecutor();
int retry = canalClientConfig.getRetries() == null || canalClientConfig.getRetries() == 0 ? 1 : canalClientConfig.getRetries();
int retry = canalClientConfig.getRetries() == null
|| canalClientConfig.getRetries() == 0 ? 1 : canalClientConfig.getRetries();
long timeout = canalClientConfig.getTimeout() == null ? 30000 : canalClientConfig.getTimeout(); // 默认超时30秒
while (running) {
@@ -56,7 +61,17 @@ public class CanalAdapterRocketMQWorker extends AbstractCanalAdapterWorker {
connector.disconnect();
break;
}
mqWriteOutData(retry, timeout, flatMessage, connector, workerExecutor);
if (retry == -1) {
retry = Integer.MAX_VALUE;
}
for (int i = 0; i < retry; i++) {
if (!running) {
break;
}
if (mqWriteOutData(retry, timeout, i, flatMessage, connector, workerExecutor)) {
break;
}
}
}
} catch (Exception e) {
logger.error(e.getMessage(), e);

View File

@@ -20,10 +20,10 @@ import com.alibaba.otter.canal.protocol.Message;
*/
public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
private static final int BATCH_SIZE = 50;
private static final int SO_TIMEOUT = 0;
private static final int BATCH_SIZE = 50;
private static final int SO_TIMEOUT = 0;
private CanalConnector connector;
private CanalConnector connector;
/**
* 单台client适配器worker的构造方法
@@ -61,10 +61,14 @@ public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
while (!running)
; // waiting until running == true
ExecutorService workerExecutor = Executors.newSingleThreadExecutor();
int retry = canalClientConfig.getRetries() == null
|| canalClientConfig.getRetries() == 0 ? 1 : canalClientConfig.getRetries();
long timeout = canalClientConfig.getTimeout() == null ? 300000 : canalClientConfig.getTimeout(); // 默认超时5分钟
if (retry == -1) {
// 重试次数-1代表异常时一直阻塞重试
retry = Integer.MAX_VALUE;
}
// long timeout = canalClientConfig.getTimeout() == null ? 300000 :
// canalClientConfig.getTimeout(); // 默认超时5分钟
Integer batchSize = canalClientConfig.getBatchSize();
if (batchSize == null) {
batchSize = BATCH_SIZE;
@@ -90,6 +94,9 @@ public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
}
for (int i = 0; i < retry; i++) {
if (!running) {
break;
}
Message message = connector.getWithoutAck(batchSize); // 获取指定数量的数据
long batchId = message.getId();
try {
@@ -97,29 +104,19 @@ public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
if (batchId == -1 || size == 0) {
Thread.sleep(500);
} else {
Future<Boolean> future = workerExecutor.submit(() -> {
if (logger.isDebugEnabled()) {
logger.debug("destination: {} batchId: {} batchSize: {} ",
canalDestination,
batchId,
size);
}
long begin = System.currentTimeMillis();
writeOut(message);
if (logger.isDebugEnabled()) {
logger.debug("destination: {} batchId: {} elapsed time: {} ms",
canalDestination,
batchId,
System.currentTimeMillis() - begin);
}
return true;
});
try {
future.get(timeout, TimeUnit.MILLISECONDS);
} catch (Exception e) {
future.cancel(true);
throw e;
if (logger.isDebugEnabled()) {
logger.debug("destination: {} batchId: {} batchSize: {} ",
canalDestination,
batchId,
size);
}
long begin = System.currentTimeMillis();
writeOut(message);
if (logger.isDebugEnabled()) {
logger.debug("destination: {} batchId: {} elapsed time: {} ms",
canalDestination,
batchId,
System.currentTimeMillis() - begin);
}
}
connector.ack(batchId); // 提交确认
@@ -127,10 +124,11 @@ public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
} catch (Exception e) {
if (i != retry - 1) {
connector.rollback(batchId); // 处理失败, 回滚数据
logger.error(e.getMessage() + " Error sync and rollback, execute times: " + (i + 1));
} else {
connector.ack(batchId);
logger.error(e.getMessage() + " Error sync but ACK!");
}
logger.error("sync error!", e);
Thread.sleep(500);
}
}
@@ -151,8 +149,6 @@ public class CanalAdapterWorker extends AbstractCanalAdapterWorker {
}
}
}
workerExecutor.shutdown();
}
@Override