Optimised integration between Context Propagation and Mutiny

[Sanne]

This unlocks significant performance improvements as it allows to skip CP processing in a Mutiny chain, following the existing documentation we have in the CP guide.

(Apparently it wasn't being applied to Mutiny - not sure if to classify it as a bugfix, let's say it's an efficiency improvement)

[quarkus-bot]

Failing Jobs - Building ad2cd96

Status	Name	Step	Failures	Logs	Raw logs
✔️	JVM Tests - JDK 11
✖	JVM Tests - JDK 11 Windows	Upload gc.log	⚠️ Check →	Logs	Raw logs
✔️	JVM Tests - JDK 17

[Sanne]

@geoand , @famod : there's an odd failure on a windows specific github action... but it might be related :/ Do you know about that task?

I see it has the context propagation 1.2 in the path, which my PR updated to 1.2.2

No idea how this triggered it, but I see the github action is:

     - name: Upload gc.log
        uses: actions/upload-artifact@v2
        with:
          name: "GC log - JDK ${{matrix.java.name}}"
          path: |
            **/windows-java-11.txt
            !**/build/tmp/**

While the errror message states:

Error: ENOENT: no such file or directory, realpath 'D:\a\quarkus\quarkus\devtools\gradle\gradle-extension-plugin\build\tmp\test\work\.gradle-test-kit\caches\modules-2\metadata-2.97\descriptors\org.eclipse.microprofile.context-propagation\microprofile-context-propagation-parent\1.2\a8be1fe3b3911d3d3425fe720cf42835'

Is it possible that last line in the path exclusions needs to be expressed in windows format? I'm just guessing, really not familiar with this.

[Sanne]

Looks like on windows slashes can be expressed in both ways (according to docs at least..)

• https://github.com/actions/toolkit/tree/main/packages/glob#glob-behavior

But also, a missing file should generate a warning by default:

• https://github.com/actions/upload-artifact#customization-if-no-files-are-found

So I'm not sure. I wonder if it's racy: the gradle cache daemon could have deleted the file after the expression had included it, but before it could have finished the transfer? Since it failed on Windows, I'm not surprised an opened file wasn't locked against deletion.

But then again the exlusion pattern should have prevented this.. I'm confused.

[Sanne]

oh, just found this is an existing problem:

• Try to prevent path failure during 'Upload gc.log' in Windows job #20679

The relation with context-propagation was a coincidence apparently .. I guess I'll merge then!

[famod]

Re-adding 2.3 backport label after the following discussion: https://quarkusio.zulipchat.com/#narrow/stream/187038-dev/topic/Mailer.3A.20.22Retries.20exhausted.22.20without.20any.20further.20info