I've also observed the same thing happened to nova-compute processes. In that case, upstart 'looses track' of the process and calls to restart it end up launching a second nova-compute process, though the first remains responsive, servicing requests from the scheduler with bad config (the defaults).
I have a feeling https://review.openstack.org/#/c/8733/ has something to do with this, and also that the disappearance of the original traceback is the result of moving to openstack-common logging functionality.
I've also observed the same thing happened to nova-compute processes. In that case, upstart 'looses track' of the process and calls to restart it end up launching a second nova-compute process, though the first remains responsive, servicing requests from the scheduler with bad config (the defaults).
I have a feeling https:/ /review. openstack. org/#/c/ 8733/ has something to do with this, and also that the disappearance of the original traceback is the result of moving to openstack-common logging functionality.
Thread @ https:/ /lists. secondlife. com/pipermail/ eventletdev/ 2011-February/ 000952. html seems consistent with what I'm seeing. I've applied a small testing patch to our repository that disables the logging calls from signal handler to see if that helps.