Not currently, although once we have the basic Metrics framework we could look at adding them without much effort.
If I understand correctly, the most useful case is for periodically recording metrics about some external system possibly at a source or sink. I'd like to think about if there was a way to more cleanly support that use case -- eg., maybe a source or sink reports metrics about the external system it interacts with, and does so in a way that we can associate it with the source or sink directly.
What have you found them useful for?
Reply was deleted
Show more
Show less
Comment details cannot be verified
Vikas RK
Jan 26, 2017
Approver
In general, useful for anything not a counter or histogram. For Beam, agreed that the most utility seems to be when interacting with external systems. Although given that a DoFn is an arbitrary piece of code written by the user, I can see it being useful. A DoFn could be talking to an external system or using a data structure and want to record metrics periodically. Given that it is not much effort, it might be worth adding at some point.
Reply was deleted
Show more
Show less
Comment details cannot be verified
2 replies
New
Robert Bradshaw
Feb 27, 2017
Approver
Any objection to actually spelling out increment/decrement. Reviewing some code that actually used this, it wasn't as obvious without context.
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
Ben Chambers
Feb 27, 2017
Approver
Not particularly, although it would require a change to both Python and Java metrics APIs. It does decrease comparability to Dropwizard's Counter interface.
Reply was deleted
Show more
Show less
Comment details cannot be verified
Robert Bradshaw
Feb 28, 2017
Approver
Mirroring an existing interface is a fair justification (though I suppose so is internal consistency, it seems other than "P[ar]" we've shunned abbreviations in the beam API).
Reply was deleted
Show more
Show less
Comment details cannot be verified
2 replies
New
Ben Chambers
Oct 5, 2016
Approver
Should this be "Distribution" (which suggests CDF and/or descriptive statics about the distribution), "Histogram" (to mirror DropWizard metrics) or something else to capture the fact we don't plan to support quantiles (yet)?
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Zur
Mar 3, 2017
Approver
Replace link:“http://timer” with “http://metrics.dropwizard.io/3.1.0/apidocs/com/codahale/metrics/Timer.html”
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Zur
Mar 3, 2017
Approver
Add row: with text “Gauge - Last value reported. (Compare to Gauge) void set(long value);”
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Ryan Williams
Jan 5, 2019
Approver
Delete:“connect a call to”
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Robert Burke
Mar 27, 2018
Approver
Format list: remove from list
Format: indent first line, indent left
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Katarzyna Kucharczyk
Oct 30, 2018
Approver
Replace:“teh” with “the”
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Ryan Williams
Jan 5, 2019
Approver
Replace:“s`” with “`s”
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
0 replies
New
Jean-Baptiste Onofré
Oct 6, 2016
Approver
The API sounds good. I have a question, how to collect and "expose" those metrics. For instance, JMX is an interesting way to do that. I guess it's the responsibility of the runner (or another Beam module) to deal with metric. Right ?
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
Robert Bradshaw
Oct 6, 2016
Approver
The returned PipelineResult would have handles to access these metrics (API TBD).
Reply was deleted
Show more
Show less
Comment details cannot be verified
Ben Chambers
Oct 6, 2016
Approver
I'll add something about the in-language querying capability. In addition to being able to access metrics from the PipelineResult, runners may provide other ways of accessing the metrics. In the future, it would be good to add a plugin that allows reporting them.
As far as JMX, I don't think it works for our needs. We'll need the runner (Fn API) to collect the metrics and aggregate them across workers and it will need to do so in a way that is consistent with committed bundles.
Reply was deleted
Show more
Show less
Comment details cannot be verified
Ben Chambers
Oct 6, 2016
Approver
Added a sketch of querying metrics above.
Reply was deleted
Show more
Show less
Comment details cannot be verified
Apache Beam Project
Oct 6, 2016
Approver
I could see a runner using JMX internally to do this as a solution but we should keep that as a runner implementation detail.
Reply was deleted
Show more
Show less
Comment details cannot be verified
4 replies
New
Rafał Wojdyła
Oct 12, 2016
Approver
+1, is there a ticket open for this?
Suggestion was deleted
Show more
Show less
Comment details cannot be verified
Ben Chambers
Oct 12, 2016
Approver
No Jira yet. I suspect it will be a while before we implement this extension, but it is definitely something we'd like. Feel free to create a feature request for adding this ability!