Add Helper Function to Build classifierF (to pass to client and server's Middleware)

[kevinmeredith]

Adds a function, classifierFMethodWithOptionallyExcludedPath, to Prometheus.

Given an exclude function, return a 'classifier' function, i.e. for application in
[[org.http4s.client.middleware.Metrics#apply]] and [[org.http4s.server.middleware.Metrics#apply]].

Let's say you want a classifier that excludes integers since your paths consist of:

    * GET    /users/{integer}    = get_users_*
    * POST   /users                = post_users
    * PUT    /users/{integer}    = put_users_*
    * DELETE /users/{integer} = delete_users_*

In such a case, we could use:

   classifierFMethodWithOptionallyExcludedPath[F] { 
         str: String => scala.util.Try(str.toInt).isSuccess 
   }

This idea, namely adding a classifier with an exclude: String => Boolean, was inspired by an internal library that @ChristopherDavenport wrote.

[rossabaker]

Summoning @ChristopherDavenport, since I think he does the most with metrics.

[rossabaker]

I'm surprised http4s-dsl is available in this project. That seems like a bad coupling. Can we do this without it?

[kevinmeredith]

I'm surprised http4s-dsl is available in this project. That seems like a bad coupling.

I only used it since I saw it available via https://github.com/http4s/http4s/blob/series/0.20/build.sbt#L153.

Can we do this without it?

Sure. Can you please advise if you know, off-hand, how to get the path as a List[String], @rossabaker?

[rossabaker]

You could copy this code and build a List rather than a Path.

In the long run, turning a path into a list of segments ought to be core functionality on a URI.

[rossabaker]

So this dependency already exists, to support the PrometheusExportService. I think we could re-express it without that, but that's a separate ticket.

[kevinmeredith]

Thanks, Ross, for telling me how to get a path list from a String. I did that, as well as removed the Http4sDsl usage via commits at #3167 (commits).

[rossabaker]

Are you needing this in 0.20.x?

[kevinmeredith]

Are you needing this in 0.20.x?

I would not say need, but it'd be nice. If it's a pain to support it in 0.20.x, please let me know, Ross.

[rossabaker]

I'd like to phase out 0.20.x releases in general and focus on keeping 0.21 healthy and developing the next version, but if somebody sends PRs, there's no particular reason we can't release another 0.20.

[rossabaker]

Comments below. This should still pass MiMa checks and be eligible for 0.20.18 and/or 0.21.1.

[rossabaker]

You could copy this code and build a List rather than a Path.

In the long run, turning a path into a list of segments ought to be core functionality on a URI.

[kevinmeredith]

This should still pass MiMa checks and be eligible for 0.20.18 and/or 0.21.1.

Understood, @rossabaker. I ran mimaReportBinaryIssues from project root successfully. I'll check the build's result tomorrow.

I believe that I've addressed your review findings, Ross. Can you please take a look?

I'd like to phase out 0.20.x releases in general and focus on keeping 0.21 healthy and developing the next version, but if somebody sends PRs, there's no particular reason we can't release another 0.20.

I interpret this to mean that it's OK for me to target series/0.20, but let me know, please, if I understood wrongly.

[kevinmeredith]

Since the other tests succeeded per https://travis-ci.org/http4s/http4s/builds/648221349?utm_source=github_status&utm_medium=notification, I'm guessing the test failure in https://travis-ci.org/http4s/http4s/jobs/648221350?utm_medium=notification&utm_source=github_status was the result of flakiness.

[info] BlazeClientSpec
[info] Blaze Http1Client should
[info]   + raise error NoConnectionAllowedException if no connections are permitted for key (383 ms)
[info]   + make simple https requests (940 ms)
[info]   + behave and not deadlock (994 ms)
[error]   x behave and not deadlock on failures with parTraverse (30 seconds, 186 ms)
[error]    None is not Some (BlazeClientSpec.scala:160)

As a result, I'll close, and then re-open, the PR to prompt another build.

[kevinmeredith]

Looks like another, I'm guessing, flaky test is to blame per https://travis-ci.org/http4s/http4s/jobs/648223657?utm_medium=notification&utm_source=github_status. I will close, and then re-open.

To make sure I follow, is it necessary to get the build green, @rossabaker, even if you have high confidence that a PR was not to blame for a failing test?

[rossabaker]

I'm rerunning the test, but the failure is definitely not related to this ticket.

Thanks for the cleanups.

[rossabaker]

And yes, if you need this on 0.20.18, it's not a problem to keep it on this branch. I backported the GitHub Actions build and threw in a Tomcat bugfix in separate PRs to make it more worth our while.

[kevinmeredith]

Can you please review this one, @ChristopherDavenport, again?

[kevinmeredith]

The most recent build appears to have run into a transient error.

[error] coursier.ResolutionException: 1 download error
[error]     Caught java.nio.channels.OverlappingFileLockException while downloading https://repo1.maven.org/maven2/com/fasterxml/jackson/core/jackson-databind/2.9.8/jackson-databind-2.9.8.jar
[error] (play-json / update) coursier.ResolutionException: 1 download error
[error]     Caught java.nio.channels.OverlappingFileLockException while downloading https://repo1.maven.org/maven2/com/fasterxml/jackson/core/jackson-databind/2.9.8/jackson-databind-2.9.8.jar
[error] Total time: 299 s (04:59), completed Feb 16, 2020, 4:53:50 AM
##[error]Process completed with exit code 1.

I'll close, and then re-open.

[kevinmeredith]

I see a few tests failed:

[info] Blaze Http1Client should
[info]   + raise error NoConnectionAllowedException if no connections are permitted for key (1 second, 131 ms)
[info]   + make simple https requests (1 second, 158 ms)
[info]   + behave and not deadlock (1 second, 182 ms)
[error]   x behave and not deadlock on failures with parTraverse (30 seconds, 576 ms)
[error]    None is not Some (BlazeClientSpec.scala:160)
[error] org.http4s.client.blaze.BlazeClientSpec.$anonfun$new$21(BlazeClientSpec.scala:160)
[info]   + behave and not deadlock on failures with parSequence (201 ms)
[info]   + obey response header timeout (212 ms)
[info]   + unblock waiting connections (1 second, 55 ms)
[info]   + reset request timeout (2 seconds, 57 ms)
[info]   + drain waiting connections after shutdown (1 second, 139 ms)
[info]   + cancel infinite request on completion (1 second, 151 ms)
[info]   + doesn't leak connection on timeout (1 second, 161 ms)
[error]   x call a second host after reusing connections on a first (5 seconds, 68 ms)
[error]    Some(Left(())) != Some(Right(simple path)) (BlazeClientSpec.scala:324)
[error] org.http4s.client.blaze.BlazeClientSpec.$anonfun$new$82(BlazeClientSpec.scala:324)
[error] Actual:   Some(Left(()))
[error] Expected: Some(Right(simple path))
[info]   + raise a ConnectionFailure when a host can't be resolved (26 ms)
[info] Total for specification BlazeClientSpec
[info] Finished in 36 seconds, 859 ms
[info] 13 examples, 18 expectations, 2 failures, 0 error
[error] Failed: Total 66, Failed 2, Errors 0, Passed 61, Skipped 3, Pending 1

• https://github.com/http4s/http4s/pull/3167/checks?check_run_id=448513735
• https://github.com/http4s/http4s/pull/3167/checks?check_run_id=448513742

but I highly doubt that my changes introduced such tests' failures.

[rossabaker]

Yes, three are still a few tests that are flaky. I was hoping the move to GitHub Actions would help, and, well, maybe, but not entirely.

[kevinmeredith]

Hi @rossabaker - please let me know what I can do to influence the merging of this PR. Thanks.

[rossabaker]

@ChristopherDavenport, are all your comments addressed?

[kevinmeredith]

Hi @rossabaker and @ChristopherDavenport - could you please cut a release per v0.20.19...series/0.20? I don't have a rush, but I'd like to start using this function. Thanks!

[kevinmeredith]

Could you please take a look at #3167 (comment), Ross or Chris?

I believe I have a use case for this in prod today.

[rossabaker]

I was trying to get another fix in, but life has been upside down. Let's get this out now. Versions are cheap.

[kevinmeredith]

I was trying to get another fix in, but life has been upside down. Let's get this out now. Versions are cheap.

Thank you, @rossabaker !

[kevinmeredith]

I was trying to get another fix in, but life has been upside down

Yes, unfortunately so. I'm glad to have spoken with you last week at NEScala. I hope you're doing well.