Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2024-08-26
16:58:13 melwitt sean-k-mooney: I just looked at the review comments you mention and yeah, that's the same conclusion that was reached with my patch is that there is no way to cancel a native thread (from the main thread). so it's either let it keep running (and assume it exits on its own eventually) or switch to processes. or like you said somehow run a timer in the child thread to exception exit itself, if there is a way we can do that
16:59:02 sean-k-mooney we might be able to do that with funtools.partil and have it create a time before callign the orginal funciton
16:59:06 melwitt the code as-is in the patch will just move on after the timer expires and leave the worker thread
16:59:10 sean-k-mooney but that feels even more hacky
16:59:48 sean-k-mooney i could use a process pool but that feels a bit heavy weight. if people are ok with that its easy to do i think
17:00:08 sean-k-mooney process pools are ok provide the funciton does not try to modify any shared sate
17:00:35 sean-k-mooney i dont think this code does but thats somethign we would need to be careful of
17:00:44 melwitt yeah. with my patch the review just kept going around in circles with nobody happy with anything so I hopefully you will have better luck
17:01:13 sean-k-mooney im feelign like i might be better off moving that patch to later in the series
17:01:52 sean-k-mooney however i see value in makeing the api eventlet free first
17:02:08 sean-k-mooney since otherwise its the simplest service to do that with
19:17:50 opendevreview melanie witt proposed openstack/nova master: DNM testing https://review.opendev.org/c/openstack/nova/+/925635
19:19:19 opendevreview melanie witt proposed openstack/nova master: DNM testing https://review.opendev.org/c/openstack/nova/+/925635
19:20:54 sean-k-mooney melwitt: by the way this is somehting i was playing with https://github.com/SeanMooney/os-vol
19:30:08 melwitt sean-k-mooney: that is very different :)
19:30:21 sean-k-mooney yep kind of intentionally
19:30:57 sean-k-mooney i was playing with the idea of storage backends abstracting how the storage is provisioned and allocating volumed which are then consumed
19:31:31 sean-k-mooney with the connection between a volume and a backend indirect though a storage pool
19:32:07 sean-k-mooney so you would not create a file or rbd volume
19:32:21 sean-k-mooney you would ask the storage pool to allocate a volume of a given size from its backend
19:33:09 sean-k-mooney that volume object is the logical handel to the backing storage
19:33:25 melwitt hm ok
19:34:10 sean-k-mooney my idea was you could have multiple backend and have adaptor that can take data form diffen sources and import it into a volume
19:35:04 sean-k-mooney i have not really fleshed out the interface much as i said i was just playing with that desgin to see what it looked like
19:35:38 melwitt that's cool
19:36:45 sean-k-mooney im not saying you should not consinue with your experiments by the way, the design im going for might be overly generic for what we want
19:37:09 sean-k-mooney on the other hand i kind of like the idea of having this in a self contianed lib or module with a stict inteface
19:37:58 sean-k-mooney the reason i have host_attach in the interafce for example is so we can allocate a file (lvm volume, rbd volume)
19:38:09 sean-k-mooney attach it to the host
19:38:17 sean-k-mooney call makefs on it and unmount it
19:38:34 sean-k-mooney without having to care where the data is stored
19:39:00 melwitt sure. where I was coming from is I wondered if we can pare down/simplify imagebackend as a baby step and whether or not it could make image format tracking easier in some way
19:39:13 sean-k-mooney yep
19:39:23 sean-k-mooney i think that would be valueable to do in anycase
19:40:11 melwitt I was thinking priority one is getting the format tracking done and I thought maybe removing some of the confusing layers of indirection could help facilitate that
19:40:34 sean-k-mooney yep i agree
19:42:15 sean-k-mooney i dont know if you looked at some of the tests i created https://github.com/SeanMooney/os-vol/blob/master/os_vol_tests/unit/backends/test_flat_file.py#L38-L63 but as i said this is just something im toying with form time to time
19:42:38 sean-k-mooney i have not considered how you could use this in nova really
19:43:04 sean-k-mooney having a simplifed code path would help if we were ever going to consider going this way
19:44:54 melwitt dansmith: I tried a minimalistic refactor of imagebackend things with the goal of helping somehow with the image format tracking we want to do https://review.opendev.org/c/openstack/nova/+/925635 I dunno if something like that could help with the ideas you had for the format tracking
19:45:35 dansmith melwitt: ah cool, I'll queue up a look
19:46:14 melwitt sean-k-mooney: ah yeah, I didn't see the tests. that does make it easier to understand
19:47:25 sean-k-mooney i defintly have some bugs but im going to keep playing wiht it on and off and see if i can build it out to be something more useful but its just a toy for now.
21:01:58 opendevreview Merged openstack/nova stable/2024.1: Functional test test_boot_reschedule_with_proper_pci_device_count https://review.opendev.org/c/openstack/nova/+/926556
21:07:38 opendevreview Merged openstack/nova stable/2024.1: Fix PCI passthrough cleanup on reschedule https://review.opendev.org/c/openstack/nova/+/926557
21:31:43 opendevreview Merged openstack/nova master: Deprecate AMI image formats https://review.opendev.org/c/openstack/nova/+/926144
#openstack-nova - 2024-08-27
07:03:17 gibi sean-k-mooney: I have no preference between eventlet an healthcheck. I will take a look at the healthcheck series today
07:09:30 gibi sean-k-mooney: on the timer, I will reply in the review
10:09:19 opendevreview Stephen Finucane proposed openstack/python-novaclient master: functional: Handle multiple networks https://review.opendev.org/c/openstack/python-novaclient/+/927249
10:09:20 opendevreview Stephen Finucane proposed openstack/python-novaclient master: Fix Python 3.12 compatibility https://review.opendev.org/c/openstack/python-novaclient/+/927251
10:09:20 opendevreview Stephen Finucane proposed openstack/python-novaclient master: tox: Add Python-specific functional envs https://review.opendev.org/c/openstack/python-novaclient/+/927250
10:44:16 opendevreview Michael Still proposed openstack/nova master: libvirt: Add extra specs for USB redirection. https://review.opendev.org/c/openstack/nova/+/927254
15:11:23 bauzas that's been a while but we'll have a nova meeting here in 50 mins
15:15:06 stephenfin sean-k-mooney: Here's the fix for the novaclient issue https://review.opendev.org/c/openstack/python-novaclient/+/927249/ I know you were saying that should test the default configuration of auto-configuring a network, but that will never work with multiple possible networks. If we want to test that properly we'll need a separate job that removes e.g. the share network, but that seems like overkill
15:15:26 stephenfin So hopefully the bit of logic I've copied in there if sufficient
15:15:33 sean-k-mooney the legacy test is testing before the auto network feature was added
15:16:04 sean-k-mooney so ya the issue here is that the job is not conigured to allw the test to run
15:16:19 opendevreview Stephen Finucane proposed openstack/python-novaclient master: tox: Add Python-specific functional envs https://review.opendev.org/c/openstack/python-novaclient/+/927250
15:16:19 opendevreview Stephen Finucane proposed openstack/python-novaclient master: functional: Handle multiple networks https://review.opendev.org/c/openstack/python-novaclient/+/927249
15:16:20 opendevreview Stephen Finucane proposed openstack/python-novaclient master: Fix Python 3.12 compatibility https://review.opendev.org/c/openstack/python-novaclient/+/927251
15:16:32 stephenfin (just spotted so leftover print statements)
15:16:42 sean-k-mooney i saw them
15:16:49 stephenfin s/so/some/
15:16:51 sean-k-mooney but my point it this is ment to be testing when we do not specify a nic
15:16:57 sean-k-mooney so adding a nic invaldiates the test
15:17:15 stephenfin so do we add a skipTest if self.multiple_networks?
15:17:18 sean-k-mooney it would be better to use the skip test funciontion if we detect multiple networks
15:17:32 sean-k-mooney yep that was one of the opetions
15:17:46 sean-k-mooney i was also lookig at modifyitg the job to remove the share network
15:18:07 stephenfin though I'll note there's nothing in the test name or docstring that says this needs to rely on being able to auto-configure a network
15:18:18 stephenfin so what makes you say that we're testing that?
15:18:18 sean-k-mooney its not
15:18:30 sean-k-mooney auto configuring the network is an diffent feature then we are testing
15:18:41 stephenfin right, but we're testing the '--image-with' falg
15:18:43 stephenfin *flag
15:18:46 sean-k-mooney this is testing the bahvior or there is only one network
15:19:02 stephenfin having to/not to pass the '--nic' option is secondary to the main test, from the looks of things?
15:19:06 sean-k-mooney which is diffent then auto config which will create it if there is none
15:19:27 stephenfin right, but do we care? that's not the aim of this test
15:19:49 sean-k-mooney it is the aim to ensure that the old command line argument work as openstack evolves
15:20:02 sean-k-mooney at least fo the legacy version that is testin 2.1
15:20:46 sean-k-mooney stephenfin: if we have a sperate test case for this then im fine with changing this one
15:20:50 stephenfin Again, right, but the test is testing the '--image-with' option, not the '--nic' option (or lack thereof). My change proves the '--image-with' option works on both 2.1 and 2.latest
15:20:59 stephenfin okay, lemme see
15:21:33 sean-k-mooney i think this is the only test without a nic specifed
15:21:36 stephenfin that was easy. It's a short file. None there. Do we add one (with the skip)?
15:22:23 stephenfin relying on there only being one test seems fairly fragile and would prevent us running said tests against a local devstack (at least without doing what CI does and e.g. deleting the 'shared' network)
15:22:30 stephenfin *one network
15:24:21 sean-k-mooney so i think this regression was intoduced by https://github.com/openstack/devstack/commit/24b65adc9cedff9c7a8ab412fb39613ef5d4a627
15:24:50 stephenfin Hardly. That's there since yoga
15:24:57 stephenfin per the GitHub UI anyway
15:25:27 sean-k-mooney maybe not but was srbac enabeld in neutron by default for that long
15:26:06 sean-k-mooney ok im torn personally i want to remvoe the cli testing and the shell module
15:26:20 sean-k-mooney on the other hand while its there we shoudl nto reduce test coverage
15:26:36 sean-k-mooney so if you want to add a test taht will skip if multiple networks i guess that fine
15:26:43 sean-k-mooney but it also kind of useless without fixing the ci
15:26:49 stephenfin agreed
15:26:59 stephenfin maybe we just merge that and move on with our lives? 🤷
15:27:16 stephenfin undoubtedly better than making the whole thing non-voting
15:27:21 sean-k-mooney if we do then i think we need to delete the cli next cycle

Earlier   Later