chore: avoid IT job hangs caused by orphan Spring Boot JVM

[ZheSun88]

Summary

Validation runs occasionally lose ~30 minutes on it-tests (11, ...) because
the step runs to the job timeout instead of finishing — costing a full re-run
attempt to produce green. Example:
https://github.com/vaadin/flow/actions/runs/25435931074/job/74614298073
(attempt 1 cancelled at 30:00, attempt 2 succeeded.)

Current behavior (root cause)

The flow-test-redeployment module starts its app via
spring-boot-maven-plugin:start (pre-IT) and stops it via :stop (post-IT).
Spring DevTools is on the classpath and the test churns class files (touching
Application.class to trigger reloads), which in this run produced 8
DevTools-driven restarts in 20 seconds, the last firing 442 ms before
spring-boot:stop ran:

12:47:56.241 [File Watcher] Restarting due to 1 class path change ...
12:47:56.683 [INFO] --- spring-boot-maven-plugin:4.1.0-RC1:stop ...
[ERROR] Spring application lifecycle JMX bean not found.
Could not stop application gracefully:
org.springframework.boot:type=Admin,name=SpringApplication

stop queries the admin MBean exactly once and there is no retry; during a
restart the bean is briefly unregistered, so the call fails. The forked
Spring Boot JVM (PID 3969 in this run) is therefore never killed. Maven
moves on under -fae, finishes 16 more modules, and exits with BUILD
FAILURE — but the orphan JVM still holds the write end of the pipe inherited
from mvn ... | tee -a mvn-it-tests-11.out (.github/workflows/validation.yml).
tee cannot reach EOF, the bash pipeline cannot complete, and the job
hangs until GitHub force-cancels at the 30-min job timeout. The runner
cleanup confirms the orphan: Terminate orphan process: pid (3969) (java).

Fix

Two small, independent changes:

1. flow-tests/test-redeployment/pom.xml — insert a maven-antrun-plugin
   execution bound to post-integration-test (declared before
   spring-boot-maven-plugin so it runs first in that phase) that sleeps 5 seconds
   before spring-boot:stop runs. This module's tests deliberately trigger DevTools
   restarts; the original failure was a 442 ms race between the last restart and
   stop-server querying the admin JMX bean. The 5-second quiet period lets
   DevTools finish re-registering the bean so stop-server can shut the app down
   cleanly instead of leaving an orphan JVM behind.

2. .github/workflows/validation.yml — add timeout-minutes: 22 on the
   Run ITs step (the job-level 30-min cap stays as the outer fence). If a
   future plugin produces a similar shape of hang, the step now fails after
   22 min instead of 30, leaving room within the original job budget for
   report packaging, artifact upload, and the existing run_attempt > 1
   retry path.

3. .github/workflows/validation.yml — replace mvn ... | tee ...out with file
   redirection (>"$LOG" 2>&1 + tail -f --pid) so an orphan child JVM holding
   an inherited stdout fd can no longer pin the step open. Add a follow-up Kill
   leftover Java processes step (if: always()) that SIGTERM/SIGKILLs any remaining
   java processes before report packaging — defensive cleanup against any
   future plugin that leaks a JVM.

[github-actions]

Test Results

 1 404 files  ±0   1 404 suites  ±0   1h 16m 42s ⏱️ -46s
10 127 tests ±0  10 057 ✅ ±0  70 💤 ±0  0 ❌ ±0 
10 602 runs  ±0  10 523 ✅ ±0  79 💤 ±0  0 ❌ ±0 

Results for commit cd2a39e. ± Comparison against base commit 8102527.

♻️ This comment has been updated with latest results.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[mcollovati]

Looks a bit tricky, but hopefully it fixes the issue 😄