Merge lp:~cjwatson/rabbitfixture/signal-esrch into lp:rabbitfixture

[Colin Watson]

Ignore ESRCH errors in RabbitServerRunner._signal.

This can happen if the server process exits by itself just before we try to signal it.

[Jürgen Gmach (jugmac00)]

How did you come up with this solution?

review: Approve

[Colin Watson (cjwatson)]

I saw http://lpbuildbot.canonical.com/builders/lp-db-devel-xenial/builds/2297/steps/shell/logs/summary, which wasn't quite the exact failure mode I remembered noticing before, and asked myself "how could that happen?". Since the code immediately preceding that was a check for whether the process is still running, the obvious answer was that it must be a race: the process exited by itself between the check for whether it was still running and the attempt to send a signal to it.

The private `_signal` method is only used to send various signals to the process that are intended to terminate it, either cooperatively (`SIGTERM`) or abruptly (`SIGKILL`). Therefore, it is not a problem if the process has already gone away, and the right answer is to simply ignore errors that mean that.

Finally, since `rabbitfixture` hasn't yet dropped Python 2 support and I didn't want to block a fix for this bug on that, I looked up https://docs.python.org/3/library/exceptions.html to confirm that `ProcessLookupError` corresponds to `ESRCH` and used the old approach of catching `OSError` and filtering by `errno` rather than catching `ProcessLookupError` directly, since that's only two more lines of code.

=== modified file 'NEWS.rst'
--- NEWS.rst	2021-02-02 10:56:39 +0000
+++ NEWS.rst	2022-07-22 19:43:08 +0000
@@ -2,6 +2,13 @@
 NEWS for rabbitfixture
 ======================
 
+0.5.1
+=====
+
+- Ignore ``ESRCH`` errors in ``RabbitServerRunner._signal``, since this can
+  happen if the server process exits by itself just before we try to signal
+  it.
+
 0.5.0 (2021-02-02)
 ==================
 

=== modified file 'rabbitfixture/server.py'
--- rabbitfixture/server.py	2021-02-01 13:15:24 +0000
+++ rabbitfixture/server.py	2022-07-22 19:43:08 +0000
@@ -465,10 +465,15 @@
     def _signal(self, code):
         """Send a signal to the server process and all its children."""
         # We need to send the signal to the process group, since on Ubuntu
-        # 14.04 an later /usr/sbin/rabbitmq-server is a shell script wrapper
-        # that spawns the actual Erlang runtime sub-process which what does
-        # actually do the work and listen for connections.
-        os.killpg(os.getpgid(self.process.pid), code)
+        # 14.04 and later /usr/sbin/rabbitmq-server is a shell script
+        # wrapper that spawns the actual Erlang runtime sub-process which is
+        # what actually does the work and listens for connections.
+        try:
+            os.killpg(os.getpgid(self.process.pid), code)
+        except OSError as e:
+            if e.errno == errno.ESRCH:
+                pass
+            raise
 
 
 class RabbitServer(Fixture):