Aias00 opened a new issue, #6858:
URL: https://github.com/apache/shenyu/issues/6858
## Description
`close()` (lines 141-150) cancels the `watchFutures` and deregisters the
service, but never calls `executor.shutdown()` or `ttlScheduler` shutdown. The
`executor` field (line 69) is a `ScheduledThreadPoolExecutor` with daemon
threads (minor leak on registry re-creation). More critically, `TtlScheduler`
(line 41) creates its scheduler via
`Executors.newSingleThreadScheduledExecutor()`, which uses the default thread
factory producing a **non-daemon** thread. This thread is never shut down —
`close()` only calls `ttlScheduler.remove(newService.getId())` which cancels
the heartbeat task but leaves the scheduler alive. Since `@Join(isSingleton =
false)` creates a new repository instance per discovery, each
`AbstractDiscoveryProcessor.removeDiscovery` → `close()` cycle leaks a
non-daemon scheduler thread. On graceful Spring context close, the non-daemon
TtlScheduler thread prevents JVM exit: the JVM will not start its shutdown
sequence while a non-daemon thread is alive, and the shu
tdown hook that calls `close()` only fires after shutdown begins — a deadlock.
## Location
-
`shenyu-registry/shenyu-registry-consul/src/main/java/org/apache/shenyu/registry/consul/ConsulInstanceRegisterRepository.java:69,141-150`
-
`shenyu-registry/shenyu-registry-consul/src/main/java/org/apache/shenyu/registry/consul/TtlScheduler.java:41`
## Impact
JVM hang on graceful shutdown (the non-daemon TtlScheduler thread blocks
exit). Thread leak on each discovery add/remove cycle in the admin, eventually
exhausting thread limits.
## Suggested fix
Add a `shutdown()` method to `TtlScheduler` that calls
`scheduler.shutdownNow()`, and call it from
`ConsulInstanceRegisterRepository.close()`. Also call `executor.shutdownNow()`
in `close()`. Alternatively, make the TtlScheduler's thread factory daemon:
`Executors.newSingleThreadScheduledExecutor(ShenyuThreadFactory.create("consul-ttl",
true))`.
## Related existing
None — distinct from N34 (#6703, `UpstreamCheckService.close` leaks
invokeExecutor) which covers the admin-side
`shenyu-admin/.../UpstreamCheckService`, not the `shenyu-registry-consul`
module. The shutdown-hook deadlock via non-daemon thread is a distinct root
cause.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
