Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-10-11
13:41:55 openstackgerrit Merged openstack/nova-specs master: Fix followup comments of policy-defaults-refresh spec https://review.opendev.org/669196
13:54:13 mdbooth lyarwood: Re https://code.engineering.redhat.com/gerrit/182841 GAH! and thanks. Mind if we don't wait for tempest after I update the commit message?
13:54:59 lyarwood mdbooth: wrong window but yeah of course
13:55:28 mdbooth lyarwood: Hah, so it is
14:11:18 openstackgerrit Balazs Gibizer proposed openstack/nova master: Avoid using image with kernel in BDM large request func test https://review.opendev.org/688132
14:40:08 gibi stephenfin: quick question in https://review.opendev.org/#/c/686802/6//COMMIT_MSG@18
14:47:00 stephenfin gibi: replied
14:47:51 gibi stephenfin: +2
14:47:58 stephenfin ta
14:55:54 stephenfin gibi: If you're on a reviewing streak, I'd appreciate your eyes on the following patch too since it's NeutronFixture'y and you know that stuff, heh https://review.opendev.org/#/c/684344/
14:56:04 stephenfin Feel free to chuck something my way too
14:56:10 gibi stephenfin: reviewing it right no
14:56:13 gibi w
14:56:20 stephenfin nice :D
14:56:43 dansmith gibi: there are a few patches in this stack for which you reviewed the spec that you could +W if you want :)
14:56:44 dansmith gibi: https://review.opendev.org/#/c/687137/4
14:57:05 gibi dansmith: ack, I will look at it after stephenfin's patch
14:57:28 dansmith gibi: thanks
14:57:48 gibi stephenfin: -1 due to https://review.opendev.org/#/c/684344/15/nova/tests/functional/api_sample_tests/test_floating_ips.py@158
14:58:23 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_images() to conductor https://review.opendev.org/687139
14:58:23 openstackgerrit Dan Smith proposed openstack/nova master: Add image caching API for aggregates https://review.opendev.org/687140
14:58:24 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Add image precaching docs for aggregates https://review.opendev.org/687348
14:58:58 stephenfin gibi: Would a follow-up be okay? Can post that now
14:59:05 gibi stephenfin: sure
14:59:10 gibi let me change my vote
15:03:35 openstackgerrit Stephen Finucane proposed openstack/nova master: nova-net: Use deepcopy on value returned by NeutronFixture https://review.opendev.org/688139
15:29:29 stephenfin gibi: Back at you https://review.opendev.org/#/c/688132/
15:30:26 gibi stephenfin: thanks. Make sense. I need to get back to that
15:53:44 gibi dansmith: reviewd the image chache series left some comments here and there
15:55:19 dansmith gibi: so the return dict thing is leaving us some room for improvement in the future without having to bump the compute rpc version again
15:55:41 dansmith gibi: are you -1 on that or the logging asserts?
15:55:54 dansmith oh you said, because the driver bit is unused
15:56:06 gibi dansmith: you mean that now the rpc returns a dict so later we can add whatever we want to that dict without rpc bump?
15:56:36 dansmith gibi: no, it's returning that so that I could follow on with some stats logging in conductor.. not expecting to add stuff to the dict later, just expecting to use it
15:56:54 dansmith gibi: there was some discussion in those patches (IIRC) but definitely some here about what we might do in the future to log some stats,
15:57:15 dansmith gibi: like "image X seemed to fail on every host" or "34 new downloads, 127 existing", that kind of thing
15:57:18 gibi dansmith: if it will be used in the future then I'm OK with it
15:57:51 dansmith gibi: yep, plan is to use it, just trying to keep the base functionality to these patches and then we can hem and haw over how to calculate some stats in a future patch
16:01:16 gibi dansmith: cool. changed my votes as the rest of my comments can be done in a fup
16:01:43 dansmith gibi: roger, just saw as I was replying, I'll post a fup for the nits and get a WIP enqueued for the stats so I can point at that if the question comes up again :D
16:02:21 gibi dansmith: thanks
16:04:44 dansmith gibi: oh, heh on that libvirt test name... it's totally opposite, I dunno how I did that :P
16:05:53 gibi :)
16:06:24 dansmith I'll call it a "testing gibi's attention to detail" easter egg
16:06:31 gibi it worked :)
16:06:56 dansmith yeah you passed the test this time
16:07:41 gibi I'm wondering was there other tests I did not even notice?!
16:09:05 dansmith gibi: muahah :D
16:09:12 gibi stephenfin: do you want me to refactor the fake image service ? https://review.opendev.org/#/c/688132/1/nova/tests/functional/test_boot_from_volume.py@203
16:11:34 gibi stephenfin: https://github.com/openstack/nova/blob/ef6e49d5bc721840b331c87c6391a69309253ade/nova/tests/unit/image/fake.py#L45
16:12:18 stephenfin gibi: If you have time, but I won't block on that now
16:12:48 gibi stephenfin: I can do that later. making a todo...
16:13:02 stephenfin tbh, I'd like to stop using the 'stub_out_image_service' function entirely since it obscures things
16:13:15 stephenfin I've a big functional test cleanup in-progress. Can include that
16:15:28 gibi stephenfin: OK, I will ping you when I reach my todo to see if you have already started on it
16:15:40 stephenfin (y)
16:16:10 stephenfin +2
16:18:25 gibi Im leaving for today. Have a nice weeked you all!
16:22:17 stephenfin O/
16:53:47 openstackgerrit Dan Smith proposed openstack/nova master: Fix up some feedback on image precache support https://review.opendev.org/688172
16:53:47 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Log some stats for image pre-cache https://review.opendev.org/688173
17:37:35 openstackgerrit Dan Smith proposed openstack/nova master: Add image caching API for aggregates https://review.opendev.org/687140
17:37:36 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Add image precaching docs for aggregates https://review.opendev.org/687348
17:37:37 openstackgerrit Dan Smith proposed openstack/nova master: Fix up some feedback on image precache support https://review.opendev.org/688172
17:37:37 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Log some stats for image pre-cache https://review.opendev.org/688173
18:06:28 openstackgerrit Stephen Finucane proposed openstack/nova master: setup.cfg: Cleanup https://review.opendev.org/677969
18:06:29 openstackgerrit Stephen Finucane proposed openstack/nova master: Stop testing Python 2 https://review.opendev.org/687954
18:13:30 melwitt o/
18:19:49 openstackgerrit Stephen Finucane proposed openstack/nova master: Stop testing Python 2 https://review.opendev.org/687954
18:28:02 openstackgerrit Merged openstack/nova master: Add cache_image() driver method and libvirt implementation https://review.opendev.org/687137
18:41:42 openstackgerrit Merged openstack/nova master: Add cache_image() support to the compute/{rpcapi,api,manager} https://review.opendev.org/687138
18:52:19 melwitt zzzeek: hey, are you around?
18:52:32 zzzeek melwitt: heya
18:52:42 melwitt o.
18:52:47 melwitt o/
18:52:51 melwitt question for you
18:53:18 zzzeek yep
18:55:02 melwitt zzzeek: do you happen to know why if this write adds a record in a single request with project_id=NULL, a non-independent read of records matching project_id=NULL will return no rows? https://github.com/openstack/nova/blob/master/nova/db/sqlalchemy/api.py#L4101
18:55:45 zzzeek melwitt: well in SQL there is no "= NULL" that works, it has to be "is NULL"
18:55:54 zzzeek melwitt: SQLAlchemy makes that conversion in most cases
18:56:12 zzzeek melwitt: however, sometimes it cant
18:56:32 zzzeek melwitt: depends on context
18:57:21 melwitt zzzeek: I did learn that recently and found that our query does make the conversion correctly. what I found is that if the insert of the record happens with "independent" and a read *without* independent happens in the same request looking for "is NULL" it will not find the record that was written. it behaves as though the inserted record is not reflected in the current session
18:57:27 zzzeek melwitt: im not seeing what the query is here but if you were to do query(Foo).with_parent(Bar(id=None)) you might see that this does not in fact return Foo with bar_id=NULL
18:58:21 zzzeek melwitt: ah well that is a transaction isolation issue
18:58:26 melwitt and if I use "independent" in the read, it _will_ find the record. that could make some sense if the session caches stuff it knows about in the current transaction
18:58:55 melwitt the weird thing is that it does not behave this way if it is not project=NULL. when project_id is not NULL, it will find the record fine without using "independent" on the read
18:59:08 zzzeek melwitt: OK so there are two levels to that. the first is, if you want to assume your transaction is non-isolated, you can say query(MyObject).populate_existing().filter(...)...
18:59:50 zzzeek melwitt: that asusmes you already have MyObject loaded and some related part of it is not being updated
19:00:15 zzzeek melwitt: if it is straight up, query(MyObject) returns no row, and the row is there, then this would be like a repeatable read problem
19:01:09 zzzeek melwitt: basically if transaction A starts, then you do sometihgn in transaction B, you can't rely that transaction A can see what you just committed in B
19:01:21 melwitt ohhhh
19:01:26 zzzeek melwitt: with a list of caveats a mile long
19:03:00 melwitt so transaction A inserts the record, transaction B reads the record and doesn't see it, yet transaction C (if added) will see what A committed. is the behavior I'm observing
19:04:55 zzzeek melwitt: yes if transaction C started after A was finished doing its work. transaction A would only have had to have committed if isolatoin level is serializable which it is not
19:05:21 zzzeek melwitt: it's mostly about, im a transaction, I read some data, now that data is part of a "version" that i will forever see until my transaction ends
19:05:49 zzzeek if i didnt read that data yet, then i dont know anything about it and based on isolation i might see the work of other transations
19:06:18 zzzeek also my previous line about A not having to commit is incorrect. it has to have committed unless isoaltion is read uncommitted, or if theres some quirky mysql behavior going on
19:08:24 zzzeek melwitt: yeah mysql is doing repeatable read by default over here. transaction A runs INSERT, but hasnt committed, B can see nothing no matter when it was started
19:08:40 zzzeek A then commits. B can only see something if it hasn't tried to read that table already

Earlier   Later