| oschwart | hey folks o/ do you know how can I modify some lines in oslo.messaging on devstack and then see the changes on designate? | 09:36 |
|---|---|---|
| oschwart | I tried adding some log prints and creating/using a virtual environment (/opt/stack/data/venv) | 09:37 |
| oschwart | I ran `pip install -e .` and restarted the designate services | 09:37 |
| oschwart | but I didn't see my log prints when I ran journalctl -u "devstack@designate*" | rg 'OSLO_MESSAGING_CHECK' | 09:37 |
| tkajinam | oschwart, I wonder if people in #openstack-qa have a better idea | 10:26 |
| tkajinam | It has been a while since I looked into devstack env, and a lot has been changed (mainly due to global venv change) since then so I'm not quite familiar with recent devstack tbvh, | 10:27 |
| oschwart | thanks tkajinam, I will try debugging a little more and if I won't find it I will write there | 10:40 |
| ralonsoh | damani[m], hberaud[m] hello folks. I'm having some issues with https://github.com/openstack/neutron/blob/master/neutron/service.py#L310-L313. In a nutshell: we create several workers, different all of them, and then we call launch_service() | 14:26 |
| ralonsoh | but it seems now that this call can be done once per process | 14:26 |
| ralonsoh | https://paste.opendev.org/show/bkxcZXwNjp12r6YDecCI/ | 14:26 |
| ralonsoh | so I would prefer not to change how we spawn the services (I mean, not to create one process per worker type). Is there an alternative implementation? | 14:27 |
| ralonsoh | more context: this is the Neutron API periodic process. Any service plugin, extension or whatever can create a PeriodicWorker (https://github.com/openstack/neutron/blob/master/neutron/worker.py#L33). This process will collect all of them and spawn them | 14:29 |
| damani[m] | ralonsoh, you're right, i will do a patch to fix that | 14:34 |
| ralonsoh | damani[m], thanks! add me to the review list | 14:35 |
| ralonsoh | damani[m], hello. I need to leave just now. I've seen this: https://paste.opendev.org/show/bMAHVkOuXb1Dp6Szk635/ | 15:18 |
| ralonsoh | do you want me to report a bug? just to document it | 15:18 |
| ralonsoh | https://bugs.launchpad.net/oslo.service/+bug/2113775 | 15:20 |
| itamarst | hello, I am trying to figure out what it takes to make oslo.service not use fork() because fork() is fundamentally broken | 15:48 |
| itamarst | (and getting eventlet to support fork() in Python 3.13 is going to be a pain) | 15:49 |
| itamarst | ("support", I should say, cause fork() is fundamentally broken so you can't really support it) | 15:49 |
| itamarst | it looks like there's a newly added thread pool based implementation - to what extent is this a sufficient replacement for using subprocesses? | 15:50 |
| itamarst | if the thread pool is not sufficient for all use cases, I suspect it's possible to make a process pool that doesn't rely on fork(), e.g. there's cloudpickle that is much better at pickling arbitrary objects than Python's built-in pickle | 16:13 |
| itamarst | (fork() is fundamentally broken any time you use a thread before the fork() happens, which is very easy to do by accident, and then you get deadlocks, memory corruption, and other fun stuff, it's really not a thing one should rely on, ever) | 16:14 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!