Ulf Lilleengen created DISPATCH-478:
---------------------------------------
Summary: Dispatch hung in shutdown due to multiple concurrent
signal handlers
Key: DISPATCH-478
URL: https://issues.apache.org/jira/browse/DISPATCH-478
Project: Qpid Dispatch
Issue Type: Bug
Reporter: Ulf Lilleengen
Priority: Minor
When I ran qdrouterd with perf and pressed Ctrl+C, perf would send SIGTERM as
well as SIGINT right after each other to the router.
Scenario:
The first thread would pick up and handle the first signal in
handle_signals_LH(). However, within the function, it unlocks the
qd_server->lock before calling the registered handler.
When it unlocks this lock, some other thread will pick up the second signal and
jump into the qd_server_pause() code, where it will wait indefinitely for
threads to pause (I saw this in GDB, where 1 thread was 'missing', and all
others marked as canceled). The original handler will have canceled the thread
trying to pause all others, but it is not able to jump out.
Patch: https://github.com/apache/qpid-dispatch/pull/95
This patch ensures that only 1 signal handler can run at a time, which fixes
the issue for me. Note that it will still potentially mask signals while
handling a signal, but the current set of signals (SIGHUP, SIGQUIT, SIGINT,
SIGTERM) either shutdown the router or do nothing.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]