[MNG-7285] [3.8.x] Replace ThreadLocal in MavenProject with locking in MojoExecutor to fix MNG-6843

[famod]

As mentioned here #570 (comment), this removes the ThreadLocal from MavenProject that was introduced in 3.8.2 and instead adds locking to MojoExecutor to prevent MNG-6843.

The major advantage is that there are no threading visibility issues anymore when passing around MavenProjects.

The major downside is that MavenProject artifacts are not "protected" anymore from concurrent modifications, just as in 3.8.1 and before.
Another downside is higher overall build runtime in case an aggregating goal takes some time to finish as it will block all other mojo executions until finished (in parallel build mode).

It does fix MNG-6843 which can be verified via https://github.com/famod/parallel-testbed/tree/compiler-fix

PS: The root cause can only be solved with a proper refeactoring like #475 but I don't think even that is enough.

Following this checklist to help us incorporate your
contribution quickly and easily:

• Make sure there is a JIRA issue filed
  for the change (usually before you start working on it). Trivial changes like typos do not
  require a JIRA issue. Your pull request should address just this issue, without
  pulling in other changes.
• Each commit in the pull request should have a meaningful subject line and body.
• Format the pull request title like [MNG-XXX] - Fixes bug in ApproximateQuantiles,
  where you replace MNG-XXX with the appropriate JIRA issue. Best practice
  is to use the JIRA issue title in the pull request title and in the first line of the
  commit message.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Run mvn clean verify to make sure basic checks pass. A more thorough check will
  be performed on your pull request automatically.
• You have run the Core IT successfully.

If your pull request is about ~20 lines of code you don't need to sign an
Individual Contributor License Agreement if you are unsure
please ask on the developers list.

To make clear that you license your contribution under
the Apache License Version 2.0, January 2004
you have to acknowledge this by using the following check-box.

• I hereby declare this contribution to be licenced under the Apache License Version 2.0, January 2004

• In any other case, please file an Apache Individual Contributor License Agreement.

[michael-o]

Let's do the revert in one commit, the new change in a separate one.

[famod]

Let's do the revert in one commit, the new change in a separate one.

Oh well, I first had it this that way but that will be two reverts. But ok, I can do that...

[michael-o]

@rmannibucau Do you consider this as a better approach than thread local? What about Olivier's PR?

[rmannibucau]

@michael-o yes, the fact it handles concurrency=1 to fall back on the well known and used behavior is a clear important step and the lock solves the original issue I think - I don't see a broken case like that at least. The thread local was creating bugs in mojo, inherited or not (even if both didnt have the exact same) so better to not try it further IMHO.

[famod]

Oliver's PR doesn't help when thread pools are involved, btw.

@gnodet I'd love to hear your take on this.

[gnodet]

@famod I will do asap. At first glance, I wonder if this is somewhat related to https://issues.apache.org/jira/browse/MNG-7156 and #476. The problem and locking strategy are different, but if the problem is the access to the MavenProject, I wonder if #476 would solve both problems...

[mickaelistria]

Maybe you should consider adding #570 or similar test into this PR to validate this fix and also prevent from future regressions.

[michael-o]

I will take care of the merging when done.
@famod Can you provide a PR for master when we have agreed on a solution?

[famod]

@michael-o

Can you provide a PR for master when we have agreed on a solution?

Sure!

@gnodet
Thanks for the pointer, I'll have a look #476 later

[gnodet]

@famod I will do asap. At first glance, I wonder if this is somewhat related to https://issues.apache.org/jira/browse/MNG-7156 and #476. The problem and locking strategy are different, but if the problem is the access to the MavenProject, I wonder if #476 would solve both problems...

I've tried to reproduce the problem with the latest mvnd which includes a fix similar to #476 with no success. I think this is to be expected since my PR provides an exclusive lock on a given project so that the same project can not be built twice in parallel (from the reactor and from a forked lifecycle), while your projects makes sure any plugin running as an aggregator can only run alone.
@famod I think my proposal would not cause any delay in the build and that it should also fix the problem since the same project would not be build concurrently, thus the resolved artifacts should not be an issue.

[famod]

I confirm that #476 seems to fix MNG-6843 as well. I've built maven-3.8.x with that LockingEventSpy and there were no failures in https://github.com/famod/parallel-testbed/tree/compiler-fix 👍

Overall, I would prefer #476 over this because it also prevents inconsistencies in other places. It's also more fine grained in that it seems to reduce the locking scope to the actual project being executed whereas with my approach, all mojo executions have to wait for an aggregating execution, no matter which project they are about to be executed for.

@gnodet

I think my proposal would not cause any delay in the build

If a "regular" mojo execution wants to run for project A, but an aggregating mojo execution comes first, chances are that A is indeed locked and the other execution has to wait for that. So there will be a "delay" but there is no way around that, a lock is a lock.

What I don't fully understand yet: When is ProjectStarted fired, especially in the case for an aggregating execution (which involves multiple projects)?

[gnodet]

I confirm that #476 seems to fix MNG-6843 as well. I've built maven-3.8.x with that LockingEventSpy and there were no failures in https://github.com/famod/parallel-testbed/tree/compiler-fix 👍

Overall, I would prefer #476 over this because it also prevents inconsistencies in other places. It's also more fine grained in that it seems to reduce the locking scope to the actual project being executed whereas with my approach, all mojo executions have to wait for an aggregating execution, no matter which project they are about to be executed for.

@gnodet

I think my proposal would not cause any delay in the build

If a "regular" mojo execution wants to run for project A, but an aggregating mojo execution comes first, chances are that A is indeed locked and the other execution has to wait for that. So there will be a "delay" but there is no way around that, a lock is a lock.

Yes, obviously. What I meant was less delay, because of the finer grained lock.

What I don't fully understand yet: When is ProjectStarted fired, especially in the case for an aggregating execution (which involves multiple projects)?

The ProjectStarted is fired when a project's build is began, i.e. once for a given project in a reactor, whereas during an aggregating execution, there's the notion of forked lifecycle, and the ForkedProjectStarted event is fired from inside the MojoExecutor.

[famod]

@gnodet

whereas during an aggregating execution, there's the notion of forked lifecycle, and the ForkedProjectStarted event is fired from inside the MojoExecutor.

I was under the impression that forked lifecycles/goals and aggregating executions are two different things?
I'm mean your locking approach seems to work, so I won't complain, but I haven't fully understood why.

[famod]

@michael-o I see you did the reversions in #595. Shall I update this PR so that it can be included in 3.8.4 or do you want to go with #476 instead?

[michael-o]

Please update

[famod]

Done, all tests are still passing, including the "parallel-testbed".

[michael-o]

I have tried #476 on the testbed repo, I still see failures, unfortunately.

[gnodet]

Superseded by #628 for 3.x and #627 for master.