CI: Use Java 26#1912
Conversation
|
@kwin this seems quite consistent: JDK transport UTs fail on Java 26? |
Seems some internal retry mechanism has changed in Java 26... |
|
With Java 26 the num of internal retries is 5 times the number of retries controlled via Methanol (=3 by default). Prior Java 26 the internal number of retries was 2... The retry handling got changed in openjdk/jdk@e8db14f#diff-41e3c3f66a6d78612e230f2546ac15e3f82619efe333a731b64f59a9e4991816R613. |
|
Please rebase on top of master now @cstamas. |
|
Ok, still not working reliably, I would tend to disable the ITs related to retries in JDKTransporter as those are too brittle. WDYT @cstamas ? |
|
Yup, go for it. |
gnodet
left a comment
There was a problem hiding this comment.
AI Review — PR #1912: CI: Use Java 26
Hi @cstamas, thanks for keeping CI up to date with the latest JDK!
⚠️ JDK 26 CI jobs are red (medium)
All 6 JDK 26 jobs (ubuntu/windows/macos × Maven 3.9.16/4.0.0-rc-5) are currently failing. The failure is in JdkTransporterTest.testRetryHandler_defaultCount_negative, caused by changed HTTP client retry semantics in Java 26.
A partial fix was merged to master on July 2 ("Limit internal retries to 1 with Java 26+") but @kwin reported on July 8 that it's "still not working reliably" and proposed disabling the brittle retry ITs. You agreed with that approach.
The PR branch appears to be 66 commits behind master and would need:
- A rebase onto current master
- The agreed-upon retry IT disabling (which hasn't been pushed yet)
Once both are done, CI should go green and this can merge.
🤖 This review was generated by ForgeBot using a maker/checker pattern (reviewer + independent verifier). All findings were independently confirmed.
Up CI matrix from Java 25 to Java 26.