Hello,
Is it possible to register user metric outside task/operator (not in RichMapFunction#open)
Thanks,
Alexey |
Hi Alexey, could you describe what you want to achieve? Most metrics are bound to a specific task (available in RuntimeContext). You can also access them in custom operators and state backends. Then you have some metrics bound to taskmanager and even java processes, but I don't see an easy way to access them. I'm also pulling in Chesnay; he can probably help you more (but I guess we really need to understand your need). On Wed, Mar 10, 2021 at 5:44 PM Alexey Trenikhun <[hidden email]> wrote:
|
Couple of use cases, I have metric representing job version, currently it bound for a task, but I want bound it to job manager, another example I have dump to on OOM exception configured, and on start, I want to check content of directory with dumps and if
not empty increase restarted-due-to-OOM counter
From: Arvid Heise <[hidden email]>
Sent: Thursday, March 11, 2021 5:22:30 AM To: Alexey Trenikhun <[hidden email]> Cc: Flink User Mail List <[hidden email]>; Chesnay Schepler <[hidden email]> Subject: Re: User metrics outside tasks Hi Alexey,
could you describe what you want to achieve? Most metrics are bound to a specific task (available in RuntimeContext). You can also access them in custom operators and state backends.
Then you have some metrics bound to taskmanager and even java processes, but I don't see an easy way to access them.
I'm also pulling in Chesnay; he can probably help you more (but I guess we really need to understand your need).
On Wed, Mar 10, 2021 at 5:44 PM Alexey Trenikhun <[hidden email]> wrote:
|
I too think this would be a useful capability for the job manager to be able
to send metrics easily. Sometimes additional compute responsibilities are placed in the job manager and having a convenient way to add telemetry data into a metrics stream would be very useful. -- Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/ |
Hi Bob and Alexey, I double-checked and there is currently no way to achieve what you want. The good news is that the OOM part should be addressed through FLINK-20833 [1], maybe it's even suitable for other issues. A "workaround" (I don't think it's a workaround) for your issues would actually be to check the logs and automatically parse certain patterns. That should be a common setup for all big data applications. Here you are free to emit whatever you want. The parsed entries should go into your metric systems if applicable. On Thu, Mar 11, 2021 at 6:05 PM Bob Tiernay <[hidden email]> wrote: I too think this would be a useful capability for the job manager to be able |
Free forum by Nabble | Edit this page |