MINOR: Add method to Herder to control logging of connector configs during validation by chia7712 · Pull Request #8263 · apache/kafka

chia7712 · 2020-03-10T13:51:30Z

Connector validation API is useful in validating connector configs before starting connector. And our application counts on it to generate friendly feedback to users. However, Connector validation API always log all configs when handling validation request since it creates AbstractConfig to check the overridable configs (introduced by #6624) and AbstractConfig, by default, logs all configs. In our case, the worker logs are filled with the following messages when there are a log of validation requests.

INFO AbstractConfig values:
xxx
xxx
xxx

It seems to me the response of Connector validation API is good enough. By contrast, the logs are a bit noisy and useless.

Committer Checklist (excluded from commit message)

Verify design and implementation
Verify test coverage and CI build status
Verify documentation (including upgrade notes)

ConcurrencyPractitioner · 2020-03-10T21:13:51Z

https://builds.apache.org/job/kafka-pr-jdk11-scala2.13/5097/testReport/

chia7712 · 2020-03-11T03:19:26Z

the flaky "KafkaBasedLogTest.testSendAndReadToEnd" is already fixed by #8255

chia7712 · 2020-03-14T04:07:44Z

@mageshn Could you take a look? thanks!

chia7712 · 2020-03-21T09:28:53Z

the related code were introduced by #6624. @ijuma Could you take a look?

kkonstantine

I don't think I agree with the justification of this change.

If by Connect validation API you are referring to the /config/validate endpoint of the Connect REST API, I'd argue that just a response from this endpoint, as a way to list the connector configs, is hardly sufficient.

To the contrary, Worker logs that contain a connector's configuration among other valuable information are the single most important source of information when troubleshooting connectors.

I'd suggest closing without merging, unless I'm missing something here.

chia7712 · 2020-03-22T05:22:59Z

@kkonstantine thanks for suggestion!

If by Connect validation API you are referring to the /config/validate endpoint of the Connect REST API, I'd argue that just a response from this endpoint, as a way to list the connector configs, is hardly sufficient.
To the contrary, Worker logs that contain a connector's configuration among other valuable information are the single most important source of information when troubleshooting connectors.

I totally agree that the connector's configuration are the useful information when troubleshooting connectors. However, the "configuration" I mentioned is not used by any connectors. Instead, it is just a temporary result. For example, users may use /config/validate multi-times before sending creation request and worker log are full of INFO AbstractConfig values.

At any rate, could we change the log level from INFO to DEBUG at least for this case?

kkonstantine · 2020-03-22T05:37:34Z

I totally agree that the connector's configuration are the useful information when troubleshooting connectors. However, the "configuration" I mentioned is not used by any connectors. Instead, it is just a temporary result. For example, users may use /config/validate multi-times before sending creation request and worker log are full of INFO AbstractConfig values.

At any rate, could we change the log level from INFO to DEBUG at least for this case?

Unfortunately AbstractHerder#validateConnectorConfig is not only used when you hit the endpoint {connector}/config/validate. It's also used by other endpoints when you create or reconfigure a connector.

Maybe you want to describe more your use case, but taking a step back, by default Connect now has the option to export the logs in files and rotate those files - a setting that can be configured even further to accommodate situations where the log becomes more verbose.

I'm not sure that by silencing this log we are gaining a lot.

chia7712 · 2020-03-22T06:01:19Z

Unfortunately AbstractHerder#validateConnectorConfig is not only used when you hit the endpoint {connector}/config/validate. It's also used by other endpoints when you create or reconfigure a connector.

My idea is to add an new construct to AbstractConfig for this case so log level is "able" to be DEBUG. for example, new AbstractConfig (xxx, xxx, debugLevel = true)

Maybe you want to describe more your use case

In short, our UI offers the visual "feedback" of configs validation. Users can get the validation response immediately after entering any config change (for example, users see the suggestion "this field must be a number" immediately after they enter a string). In order to address the function, we send a bunch of validation request to worker.. The worst case - a lot of users are configuring their connectors - there is a log file contains only INFO AbstractConfig values...

I'm not sure that by silencing this log we are gaining a lot.

I'm going to close this PR as a result of objection.

but taking a step back, by default Connect now has the option to export the logs in files and rotate those files - a setting that can be configured even further to accommodate situations where the log becomes more verbose.

@kkonstantine thanks for suggestions !

ijuma · 2020-03-22T08:14:57Z

I actually agree with @chia7712 that it's pretty surprising to see these.logs when calling a validate endpoint.

kkonstantine · 2020-03-22T17:04:34Z

Thanks for the follow up @ijuma.
Indeed, maybe I wasn't clear in the beginning, but I was referring to the proposed implementation in its current form, that would silence these logs under all circumstances (roughly calls to validate, create, and configure a connector).

As we were discussing the issue with @chia7712, I also didn't like the fact that validating a connector could generate this type of noise. In my last comment I highlighted the need to decouple the calls to Herder#validateConnectorConfig depending on whether this call happens from the {connector}/config/validate endpoint, or while creating or reconfiguring a connector.

Such decoupling would probably involve adding a method
ConfigInfos validateConnectorConfig(Map<String, String> connectorConfig, boolean doLog);
next to the existing one, to the Herder interface or the AbstractHerder class only.

What do you all think?

ijuma · 2020-03-22T17:11:35Z

Sounds good to me.

kkonstantine · 2020-03-22T17:18:05Z

@chia7712 I'm reopening this PR. Feel free to amend the commit to follow the more detailed approach described above.

chia7712 · 2020-03-22T17:39:01Z

I'm reopening this PR. Feel free to amend the commit to follow the more detailed approach described above.

@kkonstantine Got it. the commit is updated!

kkonstantine

Thanks @chia7712 !
Left a few comments re: method overrides.

kkonstantine · 2020-03-22T19:14:12Z

Actually, it's the new method that needs a default implementation if we choose to add it to this interface.

Thanks for suggestions!

Just curious, is Herder.java a public interface?

I trace the history of Herder.java. It seems to me Herder.java is not public interface so it should be fine to add new methods directly. For example, the commit added three new methods without default implementation.

Am I missing something here?

First, to your first question, no this interface is not public api, but as a general principle we are trying to maintain compatibility regardless, wherever it makes sense. Here, keeping both methods makes the most sense.

Now, w/r/t the commit that you mention, indeed there's a difference.
That commit was the implementation of KIP-558. And that's how we document changes to public interfaces, by writing and discussing publicly KIPs. You may read more here: https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Improvement+Proposals

In this PR a KIP would be an overkill, and given that this is not public interface, we can easily maintain compatibility the way I described above. The other difference is that here a default implementation makes sense, while in KIP-558 it didn't make much sense.

Hope this helps.

@kkonstantine thanks for the great explanation!!! I have addressed your comments in the latest commit. Please take a look. thanks.

kkonstantine · 2020-03-22T19:15:00Z

A per argument inline comment is not something I've seen being used widely in Connect's codebase.

kkonstantine · 2020-03-22T19:15:32Z

Suggested change

return herder.validateConnectorConfig(connectorConfig,

// the validated configs don't need to be logger.

false);

// the validated configs don't need to be logger.

return herder.validateConnectorConfig(connectorConfig, false);

kkonstantine · 2020-03-22T19:18:03Z

Related to the other comment, we'll need to keep the @Override of the previous method, and just have it call the new method that will also be overridden.

chia7712 · 2020-03-23T16:34:02Z

flaky StandbyTaskEOSIntegrationTest is traced by #8330

chia7712 · 2020-03-23T16:47:55Z

retest this please

kkonstantine · 2020-03-24T00:33:42Z

retest this please

kkonstantine

Latest build was successful.
Added a last commit with minor doc fixes.

LGTM
Merging to trunk.
Thanks @chia7712 !

chia7712 force-pushed the fix_unused_logging_in_validation branch from db4bd30 to 314a11e Compare March 10, 2020 13:52

ConcurrencyPractitioner mentioned this pull request Mar 10, 2020

[KAFKA-8522] Streamline tombstone and transaction marker removal #7884

Closed

chia7712 force-pushed the fix_unused_logging_in_validation branch from 314a11e to 256f1d5 Compare March 11, 2020 03:20

kkonstantine reviewed Mar 22, 2020

View reviewed changes

kkonstantine added the connect label Mar 22, 2020

chia7712 closed this Mar 22, 2020

kkonstantine reopened this Mar 22, 2020

chia7712 force-pushed the fix_unused_logging_in_validation branch from 256f1d5 to 9658d6a Compare March 22, 2020 17:38

kkonstantine reviewed Mar 22, 2020

View reviewed changes

chia7712 force-pushed the fix_unused_logging_in_validation branch 2 times, most recently from e3e277a to d054740 Compare March 23, 2020 12:04

Minor: remove useless log messages from connector configs validation

dbd10d4

chia7712 force-pushed the fix_unused_logging_in_validation branch from d054740 to dbd10d4 Compare March 24, 2020 03:43

MINOR: Slight rewording and fixing typos

c60d83e

kkonstantine approved these changes Mar 24, 2020

View reviewed changes

kkonstantine merged commit a5c2578 into apache:trunk Mar 24, 2020

kkonstantine changed the title ~~Minor: remove useless log messages from connector configs validation~~ MINOR: Add method to Herder to control logging of connector configs during validation Mar 24, 2020

chia7712 deleted the fix_unused_logging_in_validation branch March 25, 2024 15:23

-        return herder.validateConnectorConfig(connectorConfig,
-            // the validated configs don't need to be logger.
-            false);
+        // the validated configs don't need to be logger.
+        return herder.validateConnectorConfig(connectorConfig, false);

Conversation

chia7712 commented Mar 10, 2020

Committer Checklist (excluded from commit message)

Uh oh!

ConcurrencyPractitioner commented Mar 10, 2020

Uh oh!

chia7712 commented Mar 11, 2020

Uh oh!

chia7712 commented Mar 14, 2020

Uh oh!

chia7712 commented Mar 21, 2020

Uh oh!

kkonstantine left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chia7712 commented Mar 22, 2020

Uh oh!

kkonstantine commented Mar 22, 2020

Uh oh!

chia7712 commented Mar 22, 2020

Uh oh!

ijuma commented Mar 22, 2020

Uh oh!

kkonstantine commented Mar 22, 2020

Uh oh!

ijuma commented Mar 22, 2020

Uh oh!

kkonstantine commented Mar 22, 2020

Uh oh!

chia7712 commented Mar 22, 2020

Uh oh!

kkonstantine left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chia7712 commented Mar 23, 2020

Uh oh!

chia7712 commented Mar 23, 2020

Uh oh!

kkonstantine commented Mar 24, 2020

Uh oh!

kkonstantine left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

kkonstantine left a comment •

edited

Loading