Maybe it is a known issue[1] and has already been resolved in 1.12.2(will
release soon).
BTW, I think it is unrelated with the aliyun oss info logs.
[1]. https://issues.apache.org/jira/browse/FLINK-20992
Best,
Yang
Lei Wang 于2021年2月8日周一 下午2:22写道:
> Flink standalone HA. Flink version 1.12.1
Hi 祖安,
state抽象的数据结构,无论是value state,list state还是map state,其都是对应流计算处理中的当前key对应的数据结构。以map
state具体来说对于每个正在处理的current key (由key selector选择出来 [1]),都有一个对应
的map存储相关的数据,如果你每次都发现对应的map为空,很有可能是因为你的key
selector选择出来的key每次都不相同,很大概率是当前处理的record不同导致。
另外,map.isEmpty() 的调用是需要额外开销的(尤其对于RocksDB state
图就是哪个报错
建表语句如下,表示公共表,我也没有改的权限.
CREATE EXTERNAL TABLE `exchange_rate`(`grass_region` string COMMENT
'country', `currency` string COMMENT 'currency', `exchange_rate`
decimal(25,10) COMMENT 'exchange rate')
PARTITIONED BY (`grass_date` date COMMENT 'partition key, -MM-dd')
ROW FORMAT SERDE
Flink standalone HA. Flink version 1.12.1
2021-02-08 13:57:50,550 ERROR
org.apache.flink.runtime.util.FatalExitExceptionHandler [] - FATAL:
Thread 'cluster-io-thread-30' produced an uncaught exception. Stopping the
process...
java.util.concurrent.RejectedExecutionException: Task
你好,图挂了,可以贴一下hive建表的DDL和join的语句是怎么写的么?
On Mon, Feb 8, 2021 at 10:33 AM macia kk wrote:
> Currently the join key in Temporal Table Join can not be empty.
>
> 我的 Hive 表 join DDL 没有设置 is not null ,但是都是有值的,还是会报这个错
>
> [image: image.png]
>
--
Best regards!
Rui Li
yes, but I use stop not cancel, which also stop and cancel the job together.
Yun Gao 于2021年2月8日周一 上午11:59写道:
> Hi yidan,
>
> One more thing to confirm: are you create the savepoint and stop the job
> all together with
>
> bin/flink cancel -s [:targetDirectory] :jobId
>
> command ?
>
> Best,
>
Hi Jan,
From my view, I think in Flink Window should be as a "high-level" operation for
some kind
of aggregation operation and if it could not satisfy the requirements, we could
at least turn to
using the "low-level" api by using KeyedProcessFunction[1].
In this case, we could use a ValueState
那你可能需要把你的JobManager和TaskManager的日志发一下,才能进一步分析
主要需要确认的是连的端口是正确的,如果网络层面没有问题,那就有可能是哪个配置项使用了某个特定端口导致的
Best,
Yang
Junpb 于2021年2月8日周一 上午9:30写道:
> 你好,
> 我的测试环境yarn有三个节点,当TM启动只有一个时,JM和Tm随机启动在任何节点上都很正常,只有TM变为两个时,会出现报错。
> 每次启动JM和TM端口都是随机的,以上配置是确保2个TM启动,我现在怀疑是我其他配置导致的错误,谢谢
>
> Best,
> Bi
>
>
>
> --
>
Hi yidan,
One more thing to confirm: are you create the savepoint and stop the job all
together with
bin/flink cancel -s [:targetDirectory] :jobId
command ?
Best,
Yun
--Original Mail --
Sender:赵一旦
Send Date:Sun Feb 7 16:13:57 2021
Recipients:Till
Hi Dan
The SQL add the uuid by default is for the case that users want execute
multiple bounded sql and append to the same directory (hive table), thus
a uuid is attached to avoid overriding the previous output.
The datastream could be viewed as providing the low-level api and
thus it does not
Hi Marco,
Sorry that current statebackend is a global configuration and could
not be configured differently for different operators.
One possible alternative option to this requirements might be set rocksdb
as the default statebackend, and for those operators that want to put state
in
Hello Team,
As we have two kafka connectors "upsert-kafka" and "kafka".
I am facing issue with "upsert-kafka" while reading avro message serialized
using "io.confluent.kafka.serializers.KafkaAvroDeserializer".
Please note "kafka" connector is working while reading avro message
serialized
Hi experts,
I want to cache a temporary table for reuse it
Flink version 1.10.1
the table is consumer from kafka, struct like:
create table a (
field1 string,
field2 string,
field3 string,
field4 string
)
the sample code looks like:
val settings =
Currently the join key in Temporal Table Join can not be empty.
我的 Hive 表 join DDL 没有设置 is not null ,但是都是有值的,还是会报这个错
[image: image.png]
请教大佬们: 一个最简单pyflink UDF跑起来,报 Failed to create stage bundle factory!
INFO:root:Initializing python harness: 在IdeaIJ上可以运行、大家有遇到过吗?谢谢~
/opt/module/flink-1.11.1/bin/flink run -m localhost:8081 -pyexec
/opt/python36/bin/python3 -py udf.py
[hadoop@hadoop01 pyflink]$ /opt/python36/bin/python3 -V
你这代码贴的乱七八糟。。。
你需要再richjoinfunction里面设置valuestate的生命周期,他不随着窗口而销毁,窗口只会销毁自己设定的state,有空你可以看看window的源码,里面有清理state的逻辑
--
Sent from: http://apache-flink.147419.n8.nabble.com/
Hi,
MemoryStateBackend and FsStateBackend both hold keyed state in
HeapKeyedStateBackend [1], and the main structure to store data is StateTable
[2] which holds POJO format objects. That is to say, the object would not be
serialized when calling update().
On the other hand, RocksDB
keyedStream? key不同可能是。
谌祖安 于2021年2月7日周日 下午6:00写道:
> 您好!
>
>
> 重载procesElement方法,每条stream数据处理时,put数据进入map,后面每条数据处理时先判断是否在map中有相同的key值,无则新增后put进map,有则加工后put进map。
> 在实际代码中发现,每次写入put后,map有数据,但处理下一条数据时,先从map读取数据,发现map已经为空。
> 请问是哪里写错了吗? 和 flink官网中 state.update(current);有什么不同吗?
>
> 以下为代码:
>
你好,
我的测试环境yarn有三个节点,当TM启动只有一个时,JM和Tm随机启动在任何节点上都很正常,只有TM变为两个时,会出现报错。
每次启动JM和TM端口都是随机的,以上配置是确保2个TM启动,我现在怀疑是我其他配置导致的错误,谢谢
Best,
Bi
--
Sent from: http://apache-flink.147419.n8.nabble.com/
Hi.
*Context*
I'm migrating my Flink SQL job to DataStream. When switching to
StreamingFileSink, I noticed that the part files now do not have a uuid in
them. "part-0-0" vs "part-{uuid string}-0-0". This is easy to add with
OutputFileConfig.
*Question*
Is there a reason why the base
On Thu, Feb 04, 2021 at 04:26:42PM +0800, ChangZhuo Chen (陳昌倬) wrote:
> Hi,
>
> We have problem connecting to queryable state client proxy as described
> in [0]. Any help is appreciated.
>
> * The port 6125 is opened in taskmanager pod.
>
> ```
> root@-654b94754d-2vknh:/tmp# ss -tlp
>
Using FsStateBackend.
I was under the impression that ValueState.value will serialize an object which
is stored in the local state backend, copy the serialized object and
deserializes it. Likewise update() would do the same steps copying the object
back to local state backend.And as a
感谢回复、jar添加到lib下没重启服务、 自己S 13了;
不过又在PyFlink 应用UDF(在SQL应用udf函数)过程中遇到如下问题;把udf函数去掉,pyflink 又可以执行..
请问有遇到过嘛?谢谢~
---
Caused by: java.lang.RuntimeException: Failed to create stage bundle
factory! INFO:root:Initializing python harness:
您好!
重载procesElement方法,每条stream数据处理时,put数据进入map,后面每条数据处理时先判断是否在map中有相同的key值,无则新增后put进map,有则加工后put进map。
在实际代码中发现,每次写入put后,map有数据,但处理下一条数据时,先从map读取数据,发现map已经为空。
请问是哪里写错了吗? 和 flink官网中 state.update(current);有什么不同吗?
以下为代码:
private MapState map; //定义map
@Override
public void
执行pyflink提交任务的报错、又遇到过的大佬嘛?谢谢!
/opt/module/flink-1.11.1/bin/flink run -m localhost:8081 -pyexec
/opt/python36/bin/python3 -py NtPyFlink.py
Caused by: java.lang.RuntimeException: Failed to create stage bundle
factory! INFO:root:Initializing python harness:
Hi,
flink-connector-jdbc_2.11-1.11.1.jar
有添加在flink/lib下,只能保证在作业执行的时候,可以找到对应的class,在客户端提交的时候,会编译作业,从报错看,是客户端编译作业的时候找不到对应的class。
可以试试这里的方法:https://ci.apache.org/projects/flink/flink-docs-release-1.11/dev/python/table-api-users-guide/dependency_management.html#java-dependency
> 在
Hi,
Flink自1.11 版本之后就已经支持了hadoop3 [1][2],具体来讲就是将 HADOOP_CLASSPATH 配置成运行机器上的hadoop3
相关jar包即可。
你也可以参照 [3] 的测试步骤
[1] https://issues.apache.org/jira/browse/FLINK-11086
[2]
https://ci.apache.org/projects/flink/flink-docs-stable/deployment/resource-providers/yarn.html#supported-hadoop-versions
[3]
在k8s上部署sesson模式的flink集群遇到jobmanager报如下错误,请问这是什么原因造成的?要如何fix?
2021-02-07 08:21:41,873 INFO org.apache.flink.runtime.rpc.akka.AkkaRpcService
[] - Starting RPC endpoint for
org.apache.flink.runtime.dispatcher.StandaloneDispatcher at
akka://flink/user/rpc/dispatcher_1 .
2021-02-07
It also maybe have something to do with my job's first tasks. The second
task have two input, one is the kafka source stream(A), another is
self-defined mysql source as broadcast stream.(B)
In A: I have a 'WatermarkReAssigner', a self-defined operator which add an
offset to its input watermark and
It also maybe have something to do with my job's first tasks. The second
task have two input, one is the kafka source stream(A), another is
self-defined mysql source as broadcast stream.(B)
In A: I have a 'WatermarkReAssigner', a self-defined operator which add an
offset to its input watermark and
The first problem is critical, since the savepoint do not work.
The second problem, in which I changed the solution, removed the 'Map'
based implementation before the data are transformed to the second task,
and this case savepoint works. The only problem is that, I should stop the
job and
截图也没办法反应动态变化的过程。
目前是10机器的Standalone集群,状态在5G左右。通过flink-client端提交任务,然后web-ui刷新就一直转圈,过一会(几十秒大概)就OK啦,然后刚刚OK一瞬间会有很多个处于Initialize状态的任务,然后慢慢(10s内吧)没掉。
flink-client端的话,有时候正常提交完成,有时候出现报错(类似说是重复任务的)。
zilong xiao 于2021年2月7日周日 下午3:25写道:
> 有截图吗?
>
> 赵一旦 于2021年2月7日周日 下午3:13写道:
>
> > 这个问题现在还有个现象,我提交任务,web
32 matches
Mail list logo