[DISCUSS] FLIP-112: Support User-Defined Metrics for Python UDF

[Hequn Cheng-2]

Hi everyone,

FLIP-58 adds the support for Python UDFs, but user-defined metrics
have not been supported yet. With metrics, users can report and monitor
the UDF status to get a deeper understanding of the execution,
so in this FLIP, we want to support metrics for Python UDFs.

Previously, Jincheng and I discussed offline about the support of
metrics for Python UDFs. We'd like to achieve three goals for
supporting metrics for Python UDFs:
- Support user-defined metrics including Counters, Gauges, Meters,
  Distributions in Python UDFs.
- Support defining user scopes.
- Support defining user variables.

More details can be found in the FLIP wiki page[1] and we are looking
forward
to your feedback.

Best,
Hequn

[1]
https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF

[Xingbo Huang]

Hi Hequn,
thanks for drafting the FLIP and kicking off the discussion.

+1 for this feature.
I think this feature will be extremely convenient for PyFlink users.

Best,
Xingbo

Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：

> Hi everyone,
>
> FLIP-58 adds the support for Python UDFs, but user-defined metrics
> have not been supported yet. With metrics, users can report and monitor
> the UDF status to get a deeper understanding of the execution,
> so in this FLIP, we want to support metrics for Python UDFs.
>
> Previously, Jincheng and I discussed offline about the support of
> metrics for Python UDFs. We'd like to achieve three goals for
> supporting metrics for Python UDFs:
> - Support user-defined metrics including Counters, Gauges, Meters,
>   Distributions in Python UDFs.
> - Support defining user scopes.
> - Support defining user variables.
>
> More details can be found in the FLIP wiki page[1] and we are looking
> forward
> to your feedback.
>
> Best,
> Hequn
>
> [1]
>
> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
>

[Wei Zhong-2]

Hi Hequn,

Thanks for driving this. +1 for the metrics support for Python UDF, which makes it much easier for users to monitor the execution of Python UDFs.

Best,
Wei

> 在 2020年3月10日，15:32，Xingbo Huang <[hidden email]> 写道：
>
> Hi Hequn,
> thanks for drafting the FLIP and kicking off the discussion.
>
> +1 for this feature.
> I think this feature will be extremely convenient for PyFlink users.
>
> Best,
> Xingbo
>
> Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：
>
>> Hi everyone,
>>
>> FLIP-58 adds the support for Python UDFs, but user-defined metrics
>> have not been supported yet. With metrics, users can report and monitor
>> the UDF status to get a deeper understanding of the execution,
>> so in this FLIP, we want to support metrics for Python UDFs.
>>
>> Previously, Jincheng and I discussed offline about the support of
>> metrics for Python UDFs. We'd like to achieve three goals for
>> supporting metrics for Python UDFs:
>> - Support user-defined metrics including Counters, Gauges, Meters,
>>  Distributions in Python UDFs.
>> - Support defining user scopes.
>> - Support defining user variables.
>>
>> More details can be found in the FLIP wiki page[1] and we are looking
>> forward
>> to your feedback.
>>
>> Best,
>> Hequn
>>
>> [1]
>>
>> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
>>

[Dian Fu-2]

Hi Hequn,

Thanks for driving this. +1 to this feature.

Just one minor comment: It seems that we will add an API get_metric_group for the Python class FunctionContext, could you update the FLIP reflecting this?

Thanks,
Dian

> 在 2020年3月10日，下午3:38，Wei Zhong <[hidden email]> 写道：
>
> Hi Hequn,
>
> Thanks for driving this. +1 for the metrics support for Python UDF, which makes it much easier for users to monitor the execution of Python UDFs.
>
> Best,
> Wei
>
>
>> 在 2020年3月10日，15:32，Xingbo Huang <[hidden email]> 写道：
>>
>> Hi Hequn,
>> thanks for drafting the FLIP and kicking off the discussion.
>>
>> +1 for this feature.
>> I think this feature will be extremely convenient for PyFlink users.
>>
>> Best,
>> Xingbo
>>
>> Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：
>>
>>> Hi everyone,
>>>
>>> FLIP-58 adds the support for Python UDFs, but user-defined metrics
>>> have not been supported yet. With metrics, users can report and monitor
>>> the UDF status to get a deeper understanding of the execution,
>>> so in this FLIP, we want to support metrics for Python UDFs.
>>>
>>> Previously, Jincheng and I discussed offline about the support of
>>> metrics for Python UDFs. We'd like to achieve three goals for
>>> supporting metrics for Python UDFs:
>>> - Support user-defined metrics including Counters, Gauges, Meters,
>>> Distributions in Python UDFs.
>>> - Support defining user scopes.
>>> - Support defining user variables.
>>>
>>> More details can be found in the FLIP wiki page[1] and we are looking
>>> forward
>>> to your feedback.
>>>
>>> Best,
>>> Hequn
>>>
>>> [1]
>>>
>>> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
>>>
>

[Hequn Cheng-2]

@Dian Fu <[hidden email]>  Thanks a lot for the advice. The FLIP
page has been updated.

Best,
Hequn

On Thu, Mar 12, 2020 at 4:19 PM Dian Fu <[hidden email]> wrote:

> Hi Hequn,
>
> Thanks for driving this. +1 to this feature.
>
> Just one minor comment: It seems that we will add an API get_metric_group
> for the Python class FunctionContext, could you update the FLIP reflecting
> this?
>
> Thanks,
> Dian
>
> > 在 2020年3月10日，下午3:38，Wei Zhong <[hidden email]> 写道：
> >
> > Hi Hequn,
> >
> > Thanks for driving this. +1 for the metrics support for Python UDF,
> which makes it much easier for users to monitor the execution of Python
> UDFs.
> >
> > Best,
> > Wei
> >
> >
> >> 在 2020年3月10日，15:32，Xingbo Huang <[hidden email]> 写道：
> >>
> >> Hi Hequn,
> >> thanks for drafting the FLIP and kicking off the discussion.
> >>
> >> +1 for this feature.
> >> I think this feature will be extremely convenient for PyFlink users.
> >>
> >> Best,
> >> Xingbo
> >>
> >> Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：
> >>
> >>> Hi everyone,
> >>>
> >>> FLIP-58 adds the support for Python UDFs, but user-defined metrics
> >>> have not been supported yet. With metrics, users can report and monitor
> >>> the UDF status to get a deeper understanding of the execution,
> >>> so in this FLIP, we want to support metrics for Python UDFs.
> >>>
> >>> Previously, Jincheng and I discussed offline about the support of
> >>> metrics for Python UDFs. We'd like to achieve three goals for
> >>> supporting metrics for Python UDFs:
> >>> - Support user-defined metrics including Counters, Gauges, Meters,
> >>> Distributions in Python UDFs.
> >>> - Support defining user scopes.
> >>> - Support defining user variables.
> >>>
> >>> More details can be found in the FLIP wiki page[1] and we are looking
> >>> forward
> >>> to your feedback.
> >>>
> >>> Best,
> >>> Hequn
> >>>
> >>> [1]
> >>>
> >>>
> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
> >>>
> >
>
>

[jincheng sun]

Hi Hequn,

+1, thank you for this discussion, and metrics are very important for
monitoring the running state of Python UDF.

Best,
Jincheng


Hequn Cheng <[hidden email]> 于2020年3月12日周四 下午11:39写道：

> @Dian Fu <[hidden email]>  Thanks a lot for the advice. The FLIP
> page has been updated.
>
> Best,
> Hequn
>
> On Thu, Mar 12, 2020 at 4:19 PM Dian Fu <[hidden email]> wrote:
>
> > Hi Hequn,
> >
> > Thanks for driving this. +1 to this feature.
> >
> > Just one minor comment: It seems that we will add an API get_metric_group
> > for the Python class FunctionContext, could you update the FLIP
> reflecting
> > this?
> >
> > Thanks,
> > Dian
> >
> > > 在 2020年3月10日，下午3:38，Wei Zhong <[hidden email]> 写道：
> > >
> > > Hi Hequn,
> > >
> > > Thanks for driving this. +1 for the metrics support for Python UDF,
> > which makes it much easier for users to monitor the execution of Python
> > UDFs.
> > >
> > > Best,
> > > Wei
> > >
> > >
> > >> 在 2020年3月10日，15:32，Xingbo Huang <[hidden email]> 写道：
> > >>
> > >> Hi Hequn,
> > >> thanks for drafting the FLIP and kicking off the discussion.
> > >>
> > >> +1 for this feature.
> > >> I think this feature will be extremely convenient for PyFlink users.
> > >>
> > >> Best,
> > >> Xingbo
> > >>
> > >> Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：
> > >>
> > >>> Hi everyone,
> > >>>
> > >>> FLIP-58 adds the support for Python UDFs, but user-defined metrics
> > >>> have not been supported yet. With metrics, users can report and
> monitor
> > >>> the UDF status to get a deeper understanding of the execution,
> > >>> so in this FLIP, we want to support metrics for Python UDFs.
> > >>>
> > >>> Previously, Jincheng and I discussed offline about the support of
> > >>> metrics for Python UDFs. We'd like to achieve three goals for
> > >>> supporting metrics for Python UDFs:
> > >>> - Support user-defined metrics including Counters, Gauges, Meters,
> > >>> Distributions in Python UDFs.
> > >>> - Support defining user scopes.
> > >>> - Support defining user variables.
> > >>>
> > >>> More details can be found in the FLIP wiki page[1] and we are looking
> > >>> forward
> > >>> to your feedback.
> > >>>
> > >>> Best,
> > >>> Hequn
> > >>>
> > >>> [1]
> > >>>
> > >>>
> >
> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
> > >>>
> > >
> >
> >
>

[Hequn Cheng-2]

Hi everyone,

If there are no more concerns, I will raise the vote next week on Monday.
Thank you all for the feedback.

Best,
Hequn


On Fri, Mar 13, 2020 at 9:00 AM jincheng sun <[hidden email]>
wrote:

> Hi Hequn,
>
> +1, thank you for this discussion, and metrics are very important for
> monitoring the running state of Python UDF.
>
> Best,
> Jincheng
>
>
> Hequn Cheng <[hidden email]> 于2020年3月12日周四 下午11:39写道：
>
> > @Dian Fu <[hidden email]>  Thanks a lot for the advice. The FLIP
> > page has been updated.
> >
> > Best,
> > Hequn
> >
> > On Thu, Mar 12, 2020 at 4:19 PM Dian Fu <[hidden email]> wrote:
> >
> > > Hi Hequn,
> > >
> > > Thanks for driving this. +1 to this feature.
> > >
> > > Just one minor comment: It seems that we will add an API
> get_metric_group
> > > for the Python class FunctionContext, could you update the FLIP
> > reflecting
> > > this?
> > >
> > > Thanks,
> > > Dian
> > >
> > > > 在 2020年3月10日，下午3:38，Wei Zhong <[hidden email]> 写道：
> > > >
> > > > Hi Hequn,
> > > >
> > > > Thanks for driving this. +1 for the metrics support for Python UDF,
> > > which makes it much easier for users to monitor the execution of Python
> > > UDFs.
> > > >
> > > > Best,
> > > > Wei
> > > >
> > > >
> > > >> 在 2020年3月10日，15:32，Xingbo Huang <[hidden email]> 写道：
> > > >>
> > > >> Hi Hequn,
> > > >> thanks for drafting the FLIP and kicking off the discussion.
> > > >>
> > > >> +1 for this feature.
> > > >> I think this feature will be extremely convenient for PyFlink users.
> > > >>
> > > >> Best,
> > > >> Xingbo
> > > >>
> > > >> Hequn Cheng <[hidden email]> 于2020年3月9日周一 上午11:32写道：
> > > >>
> > > >>> Hi everyone,
> > > >>>
> > > >>> FLIP-58 adds the support for Python UDFs, but user-defined metrics
> > > >>> have not been supported yet. With metrics, users can report and
> > monitor
> > > >>> the UDF status to get a deeper understanding of the execution,
> > > >>> so in this FLIP, we want to support metrics for Python UDFs.
> > > >>>
> > > >>> Previously, Jincheng and I discussed offline about the support of
> > > >>> metrics for Python UDFs. We'd like to achieve three goals for
> > > >>> supporting metrics for Python UDFs:
> > > >>> - Support user-defined metrics including Counters, Gauges, Meters,
> > > >>> Distributions in Python UDFs.
> > > >>> - Support defining user scopes.
> > > >>> - Support defining user variables.
> > > >>>
> > > >>> More details can be found in the FLIP wiki page[1] and we are
> looking
> > > >>> forward
> > > >>> to your feedback.
> > > >>>
> > > >>> Best,
> > > >>> Hequn
> > > >>>
> > > >>> [1]
> > > >>>
> > > >>>
> > >
> >
> https://cwiki.apache.org/confluence/display/FLINK/FLIP-112%3A+Support+User-Defined+Metrics+in++Python+UDF
> > > >>>
> > > >
> > >
> > >
> >
>