~
-- --
??: "shengjk1";
: 2019??3??27??(??) 12:10
??: "user-zh@flink.apache.org";
: "user-zh";
: ?? flink savepoints ?? checkpoints??
??flink-connector-kafkakafka
??flink-connector-kafkakafka
0.8zkkafka??(??kafka??API)
Flinkoffsetcheckpoint??
??checkpointkafka??APIoffset??
??checkpoint??savepoint
1??flink
kafka??kafka??offsetflink??zookeeper
??
-- --
??: "baiyg25...@hundsun.com";
: 2019??3??27??(??) 11:38
??: "user-zh";
: : ?? flink savepoints ?? checkpoints??
savepoint
??savepoints
baiyg25...@hundsun.com
?? 2019-03-27 11:03
user-zh
?? ?? flink
??savepointsavepoints
-- --
??: "baiyg25...@hundsun.com";
: 2019??3??27??(??) 11:01
??: "user-zh";
: : flink savepoints ?? checkpoints??
sql:
select EVENTTIME,ID,EVENT_ID,MSISDN,TS
from (select a.*,ROW_NUMBER() over(partition by EVENT_ID,MSISDN order by TS
desc) AS rw
from table1 a
) where rw = 1
tableEnv.toRetractStream(结果表, Row.class).print();
输出结果,分析结果发现,第二条的 1553652720961584 比第一条的时间 1553652720927835 更大,同时输出一条
checkpoints:
checkpoint??
savepoints??
??flink savepoints ??
checkpointscheckpoints??savepointssavepoints??savepoints??
PS??checkpoints,
it might be related to this issue
https://issues.apache.org/jira/browse/FLINK-10774
On Tue, Mar 26, 2019 at 4:35 PM Fritz Budiyanto wrote:
> Hi All,
>
> We're using Flink-1.4.2 and noticed many dangling connections to Kafka
> after job deletion/recreation. The trigger here is Job
Hi,
> Would then the assumption that this possibility ( part reported length >
> part file size ( reported by FileStatus on NN) ) is only attributable to
> this edge case be correct ?
Yes, I think so.
> Or do you see a case where in though the above is true, the part file would
> need
Hi All,
We're using Flink-1.4.2 and noticed many dangling connections to Kafka after
job deletion/recreation. The trigger here is Job cancelation/failure due to
network down event followed by Job recreation.
Our flink job has checkpointing disabled, and upon job failure (due to network
Sorry to flood this thread, but keeping my experiments:
so far i've been using retract to a Row and then mapping to a dynamic pojo
that is created (using ByteBuddy) according to the select fields in the SQL.
Considering the error I'm trying now to remove thr usage in Row and use the
dynamic type
Debugging locally it seems like the state descriptor of "GroupAggregateState"
is creating an additional field (TypleSerializer of SumAccumulator)
serializer within the RowSerializer. Im guessing this is what causing
incompatibility? Is there any work around i can do?
--
Sent from:
Hi Fabian,
It seems like it didn't work.
Let me specify what i have done:
i have a SQL that looks something like:
Select a, sum(b), map[ 'sum_c', sum(c), 'sum_d', sum(d)] as my_map FROM...
GROUP BY a
As you said im preventing keys in the state forever by doing idle state
retention time (+ im
Hi,
I did write recently about our problems with 1.7.2 for which we still
haven't found a solution and the cluster is very unstable. I am trying to
point now to a different problem that maybe it is related somehow and we
don't understand.
When we restart a Flink Session in Yarn, we see it takes
Hello,
Can someone please explain me the functionality of blob server in flink ?
Thanks,
Manju
There are 2 out files and a log file. I will describe the steps to take
below, but I have not found anything to indicate there's a problem when it
cannot allocate any resources (but otherwise runs).
--start default free tier aws instanceedit security group to allow 8081
incomingsudo
Thanks Yun Tang, we will keep that in mind!
> 26 марта 2019 г., в 11:27, Yun Tang написал(а):
>
> Hi Llya
>
> I believe Java is the main implementation language of Flink internals,
> flink-core is the kernel module and implemented in Java.
>
> What's more:
> FILP6: Replace Scala implemented
Thank you for your email.
Would then the assumption that this possibility ( part reported length >
part file size ( reported by FileStatus on NN) ) is only attributable to
this edge case be correct ?
Or do you see a case where in though the above is true, the part file would
need truncation as
Hi Vishal,
I’ve come across the same problem. The problem is that by default the file
length is not updated when the output stream is not closed properly.
I modified the writer to update file lengths on each flush, but it comes with
some overhead, so this approach should be used when strong
Hi Konstantin
I think there is no direct relationship between registering
chill-protobuf/chill-thrift for Protobuf/Thrift type with Kryo and enforcing
POJO to use Kryo.
For both Protobuf and Thrift types, they will be extracted as GenericTypeInfo
within TypeExtractor which would use
不好意思,我理解错了,更正一下:
APPEND 流是没有这个字段的,只有更新流才有,true 表示 APPEND ,false 表示
update,这个值应该是流发出数据时自己带的,不是人为赋值的。。。
发件人: 邓成刚【qq】
发送时间: 2019-03-26 18:27
收件人: user-zh
主题: 回复: 实现 UpsertStreamTableSink, BatchTableSink 接口代码
这里面决定 update 或 delete 的 Boolean型值 怎么赋?
这里的 Boolean 值 是流类型决定,如果流是APPEND,则为true,
Hi Jacky,
you're right that we are currently lacking documentation for the
`mesos-appmaster-job.sh` script. I've added a JIRA issue to cover this [1].
In order to use this script you first need to store a serialized version of
the `JobGraph` you want to run somewhere where the Mesos appmaster
这里面决定 update 或 delete 的 Boolean型值 怎么赋?
这里的 Boolean 值 是流类型决定,如果流是APPEND,则为true, update,为false,你直接打印流会有这个字段
不知道我的理解是否正确,期待大佬解答。。。
邓成刚【qq】
发件人: baiyg25...@hundsun.com
发送时间: 2019-03-26 18:02
收件人: user-zh
主题: 实现 UpsertStreamTableSink, BatchTableSink 接口代码
大家好!
伙伴们,附件有实现 blink 中
HI,各位大佬:
发现一个很诡异的问题:使用SQL API时,在窗口上group by,JOB 5分钟后会timeout,但如果改成select *
就能正常消费kafka。。。
说明:本地模式和提交JOB均存在此异常
相关信息:
blink 1.5.1
kafka 1.1.1
flink-connector-kafka-0.11_2.11-1.5.1-sql-jar.jar
消费正常的code:
String sql = "select * from table1"
Table sip_distinct_event_id = tableEnv.sqlQuery(
Good point, Konstantin, that makes sense.
On Tue, Mar 26, 2019 at 10:37 AM Konstantin Knauf
wrote:
> Hi Stephan,
>
> I am in favor of renaming forceKryo() instead of removing it, because users
> might plugin their Protobuf/Thrift serializers via Kryo as advertised in
> our documentation [1].
大家好!
伙伴们,附件有实现 blink 中 flink-table 模块 UpsertStreamTableSink, BatchTableSink
接口代码 ,自实现类放在 flink-jdbc 模块 org.apache.flink.api.java.io.jdbc 包下。大神们帮忙看看呗!
江湖救急啊!
目前实现后,在代码中调用,报异常:
Exception in thread "main" org.apache.flink.table.api.TableException: Arity [4]
of result
Hi,
I have a job (with Flink 1.6.4) which uses rocksdb incremental checkpointing,
but the checkpointing always fails with `IllegalStateException`,
because hen performing `RocksDBIncrementalSnapshotOperation`, rocksdb finds
that `localBackupDirectory`, which should be created earlier
by rocksdb
Hi Stephan,
I am in favor of renaming forceKryo() instead of removing it, because users
might plugin their Protobuf/Thrift serializers via Kryo as advertised in
our documentation [1]. For this, Kryo needs to be used for POJO types as
well, if I am not mistaken.
Cheers,
Konstantin
[1]
Thanks for bringing up this DISCUSS Timo!
Java Expression DSL is pretty useful for java user. When we have the Java
Expression DSL, Java API will become very rich and easy to use!
+1 from my side.
Best,
Jincheng
Dawid Wysakowicz 于2019年3月26日周二 下午5:08写道:
> Hi,
>
> I really like the idea of
Hi
Flink高可用相关配置的存储目录,当存储路径配置成HDFS时,相关namenode高可用性由HDFS支持,对上层完全透明。
祝好
唐云
From: 戴嘉诚
Sent: Tuesday, March 26, 2019 16:57
To: user-zh@flink.apache.org
Subject: RocksDB中指定nameNode 的高可用
嘿,我想询问一下,flink中的RocksDB位置
Hi Community,
Can anyone help me understand the execution sequence in batch mode?
1. Can I set slot isolation in batch mode? I can only find the
slotSharingGroup API in streaming mode.
2. When multiple data source parallel instances are allocated to the same
slot, how does flink run those data
Hi,
I really like the idea of introducing Java Expression DSL. I think this
will solve many problems e.g. right now it's quite tricky how string
literals work in scala (sometimes it might go through the
ExpressionParser and it will end up as an UnresolvedFieldReference),
another important problem
Hi luyj,
Currently, TableAPI does not have the trigger, due to the behavior of the
windows(unbounded, tumble, slide, session) is very clear.The behavior of
each window is as follows:
- Unbounded Window - Each set of keys is a grouping, and each event
triggers a calculation.
- Tumble
Compatibility is really important for checkpointed state.
For that, you can always directly specify GenericTypeInfo or AvroTypeInfo
if you want to continue to treat a type via Kryo or Avro.
Alternatively, once https://issues.apache.org/jira/browse/FLINK-11917 is
implemented, this should happen
嘿,我想询问一下,flink中的RocksDB位置
我指定了hdfs路径,但是,这里是强指定nameNode的地址,但是我的hdfs是有个两个nameNode地址的,这里能否有个功能,当active
nameNode挂掉了,类似hdfs的HA那样,能无缝切换nameNode地址吗?不然,当nameNode挂掉了, 我指定的flink也会一并挂掉
大家好!
有没有伙伴对 blink 中 flink-table 模块下的 UpsertStreamTableSink , BatchTableSink
这两个接口比较熟悉?或者对TableSink这块处理原理比较熟悉?我想实现这两个接口,以实现JDBC更新功能,自己看源码只能看懂表面,希望熟悉的伙伴能给些指导。。。
baiyg25...@hundsun.com
Hi Llya
I believe Java is the main implementation language of Flink internals,
flink-core is the kernel module and implemented in Java.
What's more:
FILP6: Replace Scala implemented JobManager.scala and TaskManager.scala to new
JobMaster.java and TaskExecutor.java
FILP32: Make flink-table
Hello,
I am using Table API to do some aggregation based on time window.
In DataStream API, there is trigger to control when the aggregation
function should be invoked. Is there similar thing in Table API?
Because I am using large time window, like a day. I want the intermediate
result every
收到,感谢姬平老师的专业解答。
谢谢各位!
发件人: 胥平勇(姬平)
发送时间: 2019-03-26 15:01:15
收件人: bigdatayunzhongyan
抄送: user-zh; Bowen Li
主题: 回复:fw:Blink SQL报错
Hi bigdatayunzhongyan:
1. SQL语法不支持:
这个可以参照代码里面TpcDsBatchExecPlanTest的单测,我们使用的sql query也都放在了工程里。看看是不是有些query的语法有些区别。
2. 执行方式:
我们自己benchmark的时候采用的是依赖tableEnv
Hi Stephan
I prefer to remove 'enableForceKryo' since Kryo serializer does not work
out-of-the-box well for schema evolution stories due to its mutable properties,
and our built-in POJO serializer has already supported schema evolution.
On the other hand, what's the backward compatibility plan
Hello,
our dev-team is choosing a language for developing Flink jobs. Most likely that
we will use flink-streaming api (at least in the very beginning). Because of
Spark jobs developing experience we had before the choice for now is scala-api.
However recently I’ve found a
Hi Son
I think it might be because of not assigning operator ids to your Filter and
Map functions, you could refer to [1] to assign ids to your application.
Moreover, if you have ever removed some operators, please consider to add
--allowNonRestoredState [2] option.
[1]
非常谢谢您的解答,这个问题是zk中有失败任务的jobGraph,导致每次启动群集就会去检索,删除zk中残余后重启即可解决。
Thank you for your reply!
发件人: baiyg25...@hundsun.com
发送时间: 2019-03-26 09:40
收件人: user-zh
主题: Re: Re: flink ha模式进程hang!!!
是不是跟这个访问控制有关?
high-availability.zookeeper.client.acl: open
baiyg25...@hundsun.com
发件人: Han Xiao
发送时间:
45 matches
Mail list logo