问题一:Flink savepoint迁移问题
社区各位大佬,我遇到了一个Flink版本升级后,savepoint不会正确恢复问题。
版本从1.9.3 升级到1.11.0或1.11.3.
连接器使用的pulsar的连接器。在Source的checkpoint数据结构如下
oldUnionOffsetStates = stateStore.getUnionListState(
new ListStateDescriptor<>(
OFFSETS_STATE_NAME,
TypeInformation.of(new TypeHint>() {
})));
oldUnionSubscriptionNameStates =
stateStore.getUnionListState(
new ListStateDescriptor<>(
OFFSETS_STATE_NAME + "_subName",
TypeInformation.of(new TypeHint () {
})));
我在本地通过bin/flink savepoint 命令保存下1.9.3的state,然后停止flink 1.9.3的集群,启动1.11.3的flink集群,使用bin/flink run -s 参数恢复任务。
任务在启动后,会遇到下面的错误
2021-03-11 10:02:25
java.lang.Exception: Exception while creating StreamOperatorStateContext.
at org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.streamOperatorStateContext(StreamTaskStateInitializerImpl.java:222)
at org.apache.flink.streaming.api.operators.AbstractStreamOperator.initializeState(AbstractStreamOperator.java:248)
at org.apache.flink.streaming.runtime.tasks.OperatorChain.initializeStateAndOpenOperators(OperatorChain.java:290)
at org.apache.flink.streaming.runtime.tasks.StreamTask.lambda$beforeInvoke$1(StreamTask.java:506)
at org.apache.flink.streaming.runtime.tasks.StreamTaskActionExecutor$SynchronizedStreamTaskActionExecutor.runThrowing(StreamTaskActionExecutor.java:92)
at org.apache.flink.streaming.runtime.tasks.StreamTask.beforeInvoke(StreamTask.java:475)
at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:526)
at org.apache.flink.runtime.taskmanager.Task.doRun(Task.java:721)
at org.apache.flink.runtime.taskmanager.Task.run(Task.java:546)
at java.base/java.lang.Thread.run(Thread.java:832)
Caused by: org.apache.flink.util.FlinkException: Could not restore operator state backend for StreamSource_15541fe0b20ca4f01df784b178a8cc9d_(1/1) from any of the 1 provided restore options.
at org.apache.flink.streaming.api.operators.BackendRestorerProcedure.createAndRestore(BackendRestorerProcedure.java:135)
at org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.operatorStateBackend(StreamTaskStateInitializerImpl.java:283)
at org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.streamOperatorStateContext(StreamTaskStateInitializerImpl.java:156)
... 9 more
Caused by: org.apache.flink.runtime.state.BackendBuildingException: Failed when trying to restore operator state backend
at org.apache.flink.runtime.state.DefaultOperatorStateBackendBuilder.build(DefaultOperatorStateBackendBuilder.java:86)
at org.apache.flink.runtime.state.memory.MemoryStateBackend.createOperatorStateBackend(MemoryStateBackend.java:314)
at org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.lambda$operatorStateBackend$0(StreamTaskStateInitializerImpl.java:274)
at org.apache.flink.streaming.api.operators.BackendRestorerProcedure.attemptCreateAndRestore(BackendRestorerProcedure.java:142)
at org.apache.flink.streaming.api.operators.BackendRestorerProcedure.createAndRestore(BackendRestorerProcedure.java:121)
... 11 more
Caused by: java.iohttp://java.io.EOFException: No more bytes left.
at org.apache.flink.api.java.typeutils.runtime.NoFetchingInput.require(NoFetchingInput.java:79)
at com.esotericsoftware.kryo.iohttp://com.esotericsoftware.kryo.io.Input.readVarInt(Input.java:355)
at com.esotericsoftware.kryo.iohttp://com.esotericsoftware.kryo.io.Input.readInt(Input.java:350)
at com.esotericsoftware.kryo.serializers.UnsafeCacheFields$UnsafeIntField.read(UnsafeCacheFields.java:46)
at com.esotericsoftware.kryo.serializers.FieldSerializer.read(FieldSerializer.java:528)
at com.esotericsoftware.kryo.Kryo.readClassAndObject(Kryo.java:761)
at org.apache.flink.api.java.typeutils.runtime.kryo.KryoSerializer.deserialize(KryoSerializer.java:346)
at org.apache.flink.api.java.typeutils.runtime.TupleSerializer.deserialize(TupleSerializer.java:151)
at org.apache.flink.api.java.typeutils.runtime.TupleSerializer.deserialize(TupleSerializer.java:37)
at org.apache.flink.runtime.state.OperatorStateRestoreOperation.deserializeOperatorStateValues(OperatorStateRestoreOperation.java:191)
at org.apache.flink.runtime.state.OperatorStateRestoreOperation.restore(OperatorStateRestoreOperation.java:165)
at org.apache.flink.runtime.state.DefaultOperatorStateBackendBuilder.build(DefaultOperatorStateBackendBuilder.java:83)
... 15 more
请问题大佬们可以提供排查问题的办法或者解决方案吗?*来自志愿者整理的flink邮件归档
参考回答:
新的集群使用的是更新之后的 pulsar connector ?我看了下 pulsar-flink 的代码,这个更新对 state 是破坏性的。
- unionOffsetStates = stateStore.getUnionListState(
- new ListStateDescriptor<>(
- OFFSETS_STATE_NAME,
- TypeInformation.of(new TypeHint>() {
- })));
解决方法 :? 1. 尝试通过 state-processor-api 重写下 state ? 2. 1.9 的 pulsar-flink connector 可以跑在 1.11 的 flink 上吗?
感觉后面还有不兼容的更新
new ListStateDescriptor<>( OFFSETS_STATE_NAME, - TypeInformation.of(new TypeHint>() { + TypeInformation.of(new TypeHint>() { })));
不兼容的 state 更新,如果不好做 schema migration,可以试试 StatefulSinkWriterOperator 的方法:同时读取新旧 state + 只写新 state。
可以等 streamnative 的人确认下。
关于本问题的更多回答可点击原文查看:https://developer.aliyun.com/ask/359397?spm=a2c6h.13262185.0.0.54e839c0LrOat6
问题二:将每个tm的slot数从2降低到1,任务反而无法启动怎么办
您好: 任务的拓扑图如下,parallelism的设置是140,但是中间有个操作的并行度设置成了50。 集群剩余的slot总数是195。 如果将每个tm的slot数设置为2,任务能够正常启动,并且分配了70个tm和140个slot,符合预期。 可如果将每个tm的slot数设置为1,便只分配了115个slot。任务会卡在creating状态,并且几分钟后,会抛出异常,“Could not allocate all requires slots within timeout of 300000 ms. Slots required: 470, slots allocated: 388” 这里面有几个问题: 1. 将slot数设置为1后,异常中提示“Slots required: 470”,这个470似乎完全没有考虑slot share(我们并没有手动设置SlotSharingGroup)。这是为啥? 2. 将slot数设置为1后,异常中提示“slots allocated: 388”,而整个集群剩余的slot其实只有195个,这个388怎么来的? 3. 最大的并行度应是140,为何只分配了115个slot呢? 我们使用的flink版本是1.6.2。 期待你们的回复~*来自志愿者整理的flink邮件归档
参考回答:
建议看下集群剩余的内存情况,看是不是 140 个 TAskManager 内存不够了
关于本问题的更多回答可点击原文查看:https://developer.aliyun.com/ask/359418?spm=a2c6h.13262185.0.0.54e839c0YhZuwm
问题三:提交两个SQL任务,其中一个不生效怎么办?
大家好,我通过Yarn Per Job模式提交了两个任务,后一个提交的任务会变成前一个一样的任务,而且好像不产生实际作用(?),看起来像是在给前一个做HA,不知道是不是什么配置没搞好呢?
两个任务分别是 Kafka Topic1 -> Kafka Topic2 和 Kafka Topic2 -> Postgre,两个任务中的Topic2用的是同一个。
之前也尝试过,Topic1同时写入Topic2和Postgre,好像也同样不会生效,这是为什么呢?
谢谢大家。*来自志愿者整理的flink邮件归档
参考回答:
多个insert的话要用statementset去提交
关于本问题的更多回答可点击原文查看:https://developer.aliyun.com/ask/359419?spm=a2c6h.13262185.0.0.54e839c0YhZuwm
问题四:MapState 无法更新怎么办
主要代码如下:
class getRule extends KeyedBroadcastProcessFunction[String, KafkaStreamSource, List[Rule], KafkaStreamSource] { private var carEfenceState: MapState[String, Boolean] = _
override def open(parameters: Configuration): Unit = {
carEfenceState = getRuntimeContext.getMapState(new
MapStateDescriptor[String, Boolean]("carEfenceState", classOf[String],
classOf[Boolean]))
}
override def processBroadcastElement(in2: List[Rule], context:
KeyedBroadcastProcessFunction[String, KafkaStreamSource, List[Rule],
KafkaStreamSource]#Context, collector: Collector[KafkaStreamSource]): Unit =
{
context.getBroadcastState(ruleStateDescriptor).put("rules", in2)
}
override def processElement(kafkaSource: KafkaStreamSource,
readOnlyContext: KeyedBroadcastProcessFunction[String, KafkaStreamSource,
List[Rule], KafkaStreamSource]#ReadOnlyContext, collector:
Collector[KafkaStreamSource]): Unit = {
val ruleIterator =
readOnlyContext.getBroadcastState(ruleStateDescriptor).immutableEntries().iterator()
while (ruleIterator.hasNext) {
val ruleMap: Map.Entry[String, List[Rule]] = ruleIterator.next()
val ruleList: List[Rule] = ruleMap.getValue
for (rule <- ruleList) {
val mapKey = kafkaSource.vno + rule.id
val tempState = carEfenceState.get(mapKey)
val currentState = if (tempState != null) tempState else false
// 业务逻辑
if (!currentState) {
...
carEfenceState.put(mapKey, true)
...
} else if (currentState) {
...
carEfenceState.remove(mapKey)
...
}
}
}
}
}
c*来自志愿者整理的flink邮件归档
参考回答:
A read-only view of the {@link BroadcastState}.
*
- Although read-only, the user code should not modify the value returned by the {@link
- #get(Object)} or the entries of the immutable iterator returned by the {@link
- #immutableEntries()}, as this can lead to inconsistent states. The reason for this is that we do
- not create extra copies of the elements for performance reasons.
*
- @param The key type of the elements in the {@link ReadOnlyBroadcastState}.
- @param The value type of the elements in the {@link ReadOnlyBroadcastState}.
*/
这是源码中对ReadOnlyBroadcastState的描述,希望对你有帮助
关于本问题的更多回答可点击原文查看:https://developer.aliyun.com/ask/359504?spm=a2c6h.13262185.0.0.133c39c0Clkic3
问题五:BUG :DataStream 转 Table 后无法 触发窗口计算怎么办
1:当DataStream是由 一个table 经过 group by rowtime 转换过来的就无法触发窗口
例如: tableEnv.createTemporaryView("test3", tableEnv.sqlQuery("select msg,rowtime from test group by msg,rowtime"));
// 获得 DataStream,并定义wtm生成 SingleOutputStreamOperator r = tableEnv.toRetractStream(tableEnv.from("test3"), Row.class) .filter(x -> x.f0) // map ........ .returns(Types.TUPLE(Types.STRING, Types.LONG)) .assignTimestampsAndWatermarks( WatermarkStrategy.>forBoundedOutOfOrderness(Duration.ofSeconds(1)) .withTimestampAssigner(((element, recordTimestamp) -> element.f1)) );
// stream - 转 Table,指定Rowtime tableEnv.createTemporaryView("test5", r, $("msg"), $("rowtime").rowtime());
String sql5 = "select " + "msg," + "count(1) cnt" + " from test5 " + " group by TUMBLE(rowtime, INTERVAL '30' SECOND), msg " + ""; tableEnv.executeSql("insert into printlnRetractSink " + sql5);
结果: 无法触发窗口操作。 查调试源码: org.apache.flink.table.runtime.operators.window.WindowOperator // 返回的wtm永远都是 -9223372036854775808 public long getCurrentWatermark() { return internalTimerService.currentWatermark(); }
// 查看任务,watermark是正常在生成的。InternalTimerServiceImpl.advanceWatermark是正常为currentWatermark赋值。但是 internalTimerService.currentWatermark() 却拿的是-9223372036854775808
// 当 tableEnv.createTemporaryView("test3", tableEnv.sqlQuery("select msg,rowtime from test group by msg,rowtime")); 语句改为 tableEnv.createTemporaryView("test3", tableEnv.sqlQuery("select msg,rowtime from test"));
结果就是正确的。 所以这是一个bug吗??*来自志愿者整理的flink邮件归档
参考回答:
你好, 你的flink版本是多少? 之前有个bug是Table转datastream 会丢rowtime问题,看起来是这个问题。
我在[1]里修复了,你可以升级对应的版本试下。
祝好, Leonard [1]https://issues.apache.org/jira/browse/FLINK-21013 https://issues.apache.org/jira/browse/FLINK-21013
关于本问题的更多回答可点击原文查看:https://developer.aliyun.com/ask/359506?spm=a2c6h.13262185.0.0.133c39c0Clkic3