You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@flink.apache.org by makeyang <ri...@hotmail.com> on 2018/05/25 07:56:58 UTC
latency critical job
some job is latency critical job which means it can't accept certain
threadhold of latency
so will flink provide timeout operator in near future which means when one
operator timeout, the jobmanager will schedule a new operator which starts
from previous state of the OP and keep dealing with new events and discard
the events are processing.
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
Re: latency critical job
Posted by makeyang <ri...@hotmail.com>.
Rong Rong:
my flink version is 1.4.2
since we are using the docker env which is sharing disk-io, based on our
observation, disk-io spike cased by other process in the same physical
machine can lead to long time operator processing.
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
Re: latency critical job
Posted by makeyang <ri...@hotmail.com>.
Timo:
thanks for u suggestion
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
Re: latency critical job
Posted by Rong Rong <wa...@gmail.com>.
Hi Makeyang,
+1 on Timo's point. We have been dealing with these kind of problems before
and in general Flink can handle latency if implemented correctly and
assigned correct amount of computation resource (depending on what kind of
resource isolation/containerization you are doing) to handle additional
traffic spikes. In addition we have our external monitoring process on top
of [2] to provide extra auto-scaling capability.
Can you share more information regarding which operator as actually causing
the latency to pile up (you should be able to fetch that using [1]) and
which Flink version you are using?
Thanks,
Rong
On Fri, May 25, 2018 at 6:23 AM, Timo Walther <tw...@apache.org> wrote:
> Hi,
>
> usually Flink should have constant latency if the job is implemented
> correctly. But if you want to implement something like an external
> monitoring process., you can use the REST API [1] and metrics [2] to model
> such an behavior by restarting your application. In theory, you could also
> implement a mechanism in your job that throws a runtime exception to cancel
> you job if a certain threshold is violated and thus would trigger a restart
> if you set a restart strategy.
>
> Regards,
> Timo
>
> [1] https://ci.apache.org/projects/flink/flink-docs-release-1.4/
> monitoring/rest_api.html
> [2] https://ci.apache.org/projects/flink/flink-docs-release-1.4/
> monitoring/metrics.html
>
> Am 25.05.18 um 09:56 schrieb makeyang:
>
> some job is latency critical job which means it can't accept certain
>> threadhold of latency
>> so will flink provide timeout operator in near future which means when one
>> operator timeout, the jobmanager will schedule a new operator which starts
>> from previous state of the OP and keep dealing with new events and discard
>> the events are processing.
>>
>>
>>
>>
>> --
>> Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.
>> nabble.com/
>>
>
>
>
Re: latency critical job
Posted by Timo Walther <tw...@apache.org>.
Hi,
usually Flink should have constant latency if the job is implemented
correctly. But if you want to implement something like an external
monitoring process., you can use the REST API [1] and metrics [2] to
model such an behavior by restarting your application. In theory, you
could also implement a mechanism in your job that throws a runtime
exception to cancel you job if a certain threshold is violated and thus
would trigger a restart if you set a restart strategy.
Regards,
Timo
[1]
https://ci.apache.org/projects/flink/flink-docs-release-1.4/monitoring/rest_api.html
[2]
https://ci.apache.org/projects/flink/flink-docs-release-1.4/monitoring/metrics.html
Am 25.05.18 um 09:56 schrieb makeyang:
> some job is latency critical job which means it can't accept certain
> threadhold of latency
> so will flink provide timeout operator in near future which means when one
> operator timeout, the jobmanager will schedule a new operator which starts
> from previous state of the OP and keep dealing with new events and discard
> the events are processing.
>
>
>
>
> --
> Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/