fix(docker): supervise Java process in entrypoints instead of tail -f /dev/null

[bitflicker64]

Purpose of the PR

• relates to [Bug] Docker containers never restart when Java process crashes — broken cron monitor + tail -f /dev/null supervision #3043

All three docker-entrypoint.sh files used tail -f /dev/null to keep the container alive. When Java crashed, tail kept running and the container stayed up with no Java inside — Docker restart policy never fired.

Main Changes

PD and Store: replace daemon start + tail -f /dev/null with -d false, which triggers the foreground branch added in #3047. exec java replaces the shell with Java directly, so the entrypoint blocks and exits with Java's real exit code when Java exits.

Server: keep daemon start (wait-partition.sh must run after Java is already up in the background). Replace tail -f /dev/null with tail --pid=$(cat ./bin/pid) -f /dev/null. When Java exits, tail --pid returns, entrypoint exits 1, restart policy fires.

Note: server entrypoint always exits 1 — start-hugegraph.sh calls disown before returning so Java is not a child of the entrypoint shell and its real exit code is not available. This is sufficient for --restart=on-failure.

Component	Before	After
PD	daemon start + tail -f /dev/null	-d false → exec java blocks
Store	daemon start + tail -f /dev/null	-d false → exec java blocks
Server	daemon start + tail -f /dev/null	daemon start + tail --pid supervision

Verifying these changes

• Need tests and can be verified as follows:
  • docker run --restart=on-failure → kill -9 Java inside container → container exits and Docker restarts it automatically
  • Verified locally: 3 kill-9 tests on server container, PD restart loop confirmed

Does this PR potentially affect the following parts?

• Dependencies
• Modify configurations
• The public API
• Other affects
• Nope

Documentation Status

• Doc - TODO — README/docs update for Docker deployment lifecycle will follow in a separate PR

[imbajin]

Blocking: no. Summary: server Docker entrypoint still needs graceful shutdown forwarding. Evidence: static review of hugegraph-server/hugegraph-dist/docker/docker-entrypoint.sh; bash -n passed for all three changed entrypoints.

[imbajin]

⚠️ Forward shutdown to HugeGraphServer

Evidence: PD and Store now execute their foreground scripts with -d false, but the server entrypoint still starts start-hugegraph.sh in daemon mode and then blocks on tail --pid="$PID" -f /dev/null at this line. Docker sends SIGTERM to PID 1, so this shell/tail path can exit without signaling the Java PID; the server then waits for Docker's forced kill instead of a graceful shutdown.

Please add a TERM/INT trap that kills and waits for $PID, or otherwise run the server foreground path after the post-start checks, so docker stop shuts HugeGraphServer down cleanly.

[bitflicker64]

Good catch added a TERM/INT trap before the tail --pid block in f20b8a4:

PID=$(cat ./bin/pid 2>/dev/null || true)
if [[ -n "$PID" ]]; then
    trap 'kill -TERM "$PID" 2>/dev/null; while kill -0 "$PID" 2>/dev/null; do sleep 1; done; exit 0' TERM INT
    tail --pid="$PID" -f /dev/null
    exit 1
fi

When dumb-init forwards SIGTERM from docker stop, the trap fires: sends SIGTERM to Java, polls with kill -0 until Java exits cleanly, then exits 0. The tail --pid crash-detection path still exits 1 so restart policy fires on a crash.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.25%. Comparing base (1f61c48) to head (f20b8a4).

Additional details and impacted files

@@              Coverage Diff              @@
##             master    #3051       +/-   ##
=============================================
+ Coverage     36.11%   93.25%   +57.14%     
+ Complexity      338       65      -273     
=============================================
  Files           803        9      -794     
  Lines         68234      267    -67967     
  Branches       8964       22     -8942     
=============================================
- Hits          24640      249    -24391     
+ Misses        40936        8    -40928     
+ Partials       2658       10     -2648     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[imbajin]

+1 from my side. The previous server SIGTERM concern is addressed at the current head: PD/Store now use the foreground exec path, and Server keeps the hstore post-start check while supervising the Java PID with crash detection plus TERM/INT forwarding. CI is green and I don’t see remaining blockers.