15:30:03 #startmeeting Performance Team 15:30:04 Meeting started Tue Jan 9 15:30:03 2018 UTC and is due to finish in 60 minutes. The chair is DinaBelova. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:30:06 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:30:08 The meeting name has been set to 'performance_team' 15:30:27 okay, let's see if someone has survived the holiday season 15:31:52 tovin07 o/ 15:33:00 ilyashakhat o/ 15:34:34 o/ 15:34:56 wow, I see a survivor :) 15:35:01 tovin07 I have a feeling we'll be alone today 15:35:02 :D 15:35:28 I saw that ilyashakhat publish a patch a few minutes ago in cinder 15:35:36 hope that he can join us 15:36:17 okay, let's wait for a few more moments 15:37:17 hi! 15:37:20 I'm here 15:37:24 yay! 15:37:31 good to see both of you :) 15:37:35 okay, let's proceed 15:37:43 no action items for a while 15:37:50 so switching to the current status 15:37:55 #topic Current status 15:38:06 tovin07 how are you doing sir? anything to share? 15:38:26 yes 15:38:28 a little bit 15:39:10 when testing ilyashakhat's patch https://review.openstack.org/#/c/519664/ 15:39:29 I found an issue with OSProfiler in Nova 15:39:46 http://paste.openstack.org/show/641262/ 15:40:28 and it is related to this change (from very long time ago) https://review.openstack.org/#/c/254703/20/nova/utils.py 15:41:02 line 1174 - 1175 from `spawn` function 15:41:08 I also so this trace couple of times.. 15:41:37 it's interesting :) I can swear that 11 months ago it used to work :D 15:41:45 yes 15:41:53 for me the issue happens rarely 15:42:13 tovin07 ilyashakhat it's quite interesting 15:43:02 as obviously what the lines you've tovin07 mentioned is doing is just initiating profiler if not context with it is in place 15:43:07 as I see here, because we re-initialize osprofiler when it is already initialized will clear current profiler object 15:43:31 --> the self._name queue will be empty 15:43:40 oh, yeah, it might be the reason 15:43:57 I think we should have `a check` here 15:44:04 tovin07 can you provide a patch? 15:44:12 it looks like minor one 15:44:12 yes, I will 15:44:24 cool 15:44:26 thank you 15:44:51 why this re-initialization is needed? 15:45:33 ilyashakhat frankly speaking I'm not sure - patch ir really old and I do not remember the details 15:45:51 Boris Pavlovic 15:45:51 Jan 19, 2016 15:45:51 Patch Set 19: Code-Review-1 15:45:51 We need to handle green threads as well, like here: 15:45:51 https://github.com/boris-42/nova/commit/6f81fbc45db8ed02c723e534f4a0780845e4d1a8 15:45:59 that's the reason 15:46:25 additionally iyour link is 404 not found to me :) 15:46:29 after this comment, DinaBelova adds these code to the utils.py file 15:46:38 yes, it 404 for now 15:46:59 I cannot find any cache of that link on the web too 15:47:04 yeah, I se this comment 15:47:36 I guess it was not accurate change introduced occasionally without a check 15:48:19 thanks tovin07 for finding it 15:48:37 btw, profiler state is defined as threading.local() - does it mean to stay per green thread too? 15:49:17 (that may be a reason why re-initialization is needed) 15:49:49 it was supposed to be so 15:51:42 tovin07 anything else to add? 15:51:46 so the issue in Nova may occur when there is already some profiling data and then a new thread is forked 15:51:59 ilyashakhat yes, it looks like this 15:52:04 so we need to properly 'fork' profiler state too 15:52:13 and this is happening not too often 15:52:28 that's why we do not see it too much 15:52:39 yeah... 15:52:55 let's move on 15:52:58 ack 15:53:04 ah 15:53:12 tovin07 anything else to share? 15:53:41 I see this happen with `openstack server create` and some other commands 15:54:30 tovin07 ack 15:54:34 ilyashakhat let's proceed with you :) anything to share with us? how did your holidays go? 15:55:08 holidays went well :) so I don't have much to share ;) 15:55:18 ilyashakhat :D 15:55:24 that's good too :D 15:55:26 I've posted a patch to Cinder to fix initialization of WSGI 15:55:46 similar patch in Nova still hangs 15:55:48 yes 15:56:01 i tested your cinder patch too 15:56:02 yeah, I see it 15:56:17 nova patch is quite old... 15:56:29 well, Nova review queue sometimes may go crazy 15:56:48 however, I see that traces from cinder are quite short 15:56:56 although for 2 lines patch it looks a bit overkill 15:57:07 DinaBelova, yes 15:57:33 tovin07, yes, I see only wsgi events -- looks like there's some more issue 15:58:57 it looks like new year brought more gifts than expected :) 15:58:58 more bugs :D 15:59:05 from my side nothing specific to share 15:59:21 I will reach nova cores in nova channel to get another +2 and A+1 for ilyashakhat patch 15:59:30 tovin07 thanks 15:59:32 thanks! 15:59:37 #topic Open Discussion 16:00:02 okay, so tovin07 ilyashakhat - our brave holiday survivors - anything to share in addition? 16:00:06 :) 16:00:11 I have :) 16:00:15 ilyashakhat shoot 16:00:27 I'm going to finish https://review.openstack.org/#/c/521385/ 16:00:50 thanks ilyashakhat 16:01:00 yaaay 16:01:01 so we will have `trace list` command 16:01:07 it's good to hear 16:01:22 DinaBelova, do we have any deadline to release and be included into Q? 16:01:42 (if it's not too late already) 16:02:06 well, we did not have too many changes 16:02:08 usually no, but osprofiler is under oslo umbrella 16:02:18 yeah 16:02:30 therefore, it will depend on oslo release cycle 16:03:05 Final non-client library release deadline: January 18 16:03:18 ilyashakhat so we have some time 16:03:20 a bit but still 16:03:25 ok 16:04:21 thanks ilyashakhat 16:04:26 anything else? 16:05:03 nope 16:05:03 :D 16:05:06 it looks like nope :) if so, thanks tovin07 and ilyashakhat for showing up :) 16:05:06 nope 16:05:14 have a nice evening! 16:05:17 #endmeeting