Work around problems with ports in t/cycle #52
Conversation
|
Hm, t/cycle failed again. I think we may need to add some debugging help to travis output. |
|
Retrying binding(Address already in use) messages appear only by testing with |
The comments in this code suggest that a busy server can run into problems and that's why the code retries up to 20 times. I've raised that to 50 to deal with frequent failures in Travis CI. Unfortunately additions to this value seem to have diminishing returns. Presumably there are things that slip into window between checking if the port is available and binding gearmand to it that don't let go at all. We also change the Retry to info, instead of debug, as that is useful information for a system administrator while the daemon starts up, and should not be relegated to debug-only logs. This closes #50
|
I'm going to rethink the approach. I think this patch does reduce the incidence of the problem, but the goal is to eliminate it and I think I have a scheme for doing that. The artifacts change is just a config change to get artifacts stored so that when these tests fail we can download them and attempt to debug. |
That sounds good. |
The comments in this code suggest that a busy server can run into
problems and that's why the code retries up to 20 times. I've raised
that to 30 to deal with frequent failures in Travis CI.
We also change the Retry to info, instead of debug, as that is useful
information for a system administrator while the daemon starts up, and
should not be relegated to debug-only logs.
This closes #50