Thursday, 2026-08-27

opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher master: Add Workload Optimization goal and Workload Preemption strategy  https://review.opendev.org/c/openstack/watcher/+/100017003:01
dviroel#startmeeting watcher13:01
opendevmeetMeeting started Thu Aug 27 13:01:16 2026 UTC and is due to finish in 60 minutes.  The chair is dviroel. Information about MeetBot at http://wiki.debian.org/MeetBot.13:01
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.13:01
opendevmeetThe meeting name has been set to 'watcher'13:01
* dviroel was reviewing a code and almost missed the meeting start13:01
dviroelhi all o/13:01
morenodo/13:01
chandankumaro/13:02
dviroelcourtesy ping:  sean-k-mooney rlandy13:02
dviroeladditional ping: winiciusallan[m] 13:02
dviroel:) 13:02
dviroelok, so let's start with today's meeting agenda13:03
dviroel#link https://etherpad.opendev.org/p/openstack-watcher-irc-meeting#L34 (Meeting agenda)13:03
dviroelfeel free to add your own topics to the agenda13:03
dviroellets start with the first one13:04
dviroel#topic Announcements13:04
dviroel1. Elections13:04
dviroeljust in case you missed:13:04
dviroelPolls for TC elections are now open and will remain open for you to your vote until Sep 16, 2026 23:45 UTC13:04
dviroelif you want to know if you are elegiblt to vote13:05
dviroelcheck Ian email for more info13:05
dviroel#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/thread/CU6SOGKPFEBTMGJ63T54G72E2MSJTZGG/13:05
winiciusallan[m]o/13:05
dviroelnext announcement13:06
dviroel2. Hibiscus Schedule13:06
dviroel#link https://releases.openstack.org/hibiscus/schedule.html13:06
dviroelwe reach hibiscus milestone-3 13:06
dviroelfeature freeze is today13:07
dviroelwhich mean that featureful patch should be landed after this point13:07
dviroelbut some expections may be granted by the ptl/core team 13:08
dviroelwe should cover the open patch in the next topic13:08
dviroelnote that today is also the deadline for releasing client libraries13:09
dviroelthe python-watcherclient release patch is open13:09
dviroel#link https://review.opendev.org/c/openstack/releases/+/100207513:09
dviroelwe may go with current version, which adds audit-template default_parameters13:10
dviroelbut we can discuss that in a few13:10
dviroelany other announcement?13:10
dviroel#topic Reviews13:11
dviroelstartin with 13:11
dviroel#link https://review.opendev.org/c/openstack/watcher-dashboard/+/99813113:11
dviroelmorenod patch that adds new playwright tests13:11
morenodyep. all requirements are now merged, we can review it13:12
dviroelI think that I will have more time to review it next week, after we cover the features that we still need to land13:12
dviroelnice, thanks for the work morenod 13:12
dviroelchandankumar: ptal also when yo have some time :)13:12
dviroelok, so next we can go through the reviews in the status etherpad13:13
dviroelgoing to start with13:13
dviroel1. Preemtible Instances13:13
dviroelsince winiciusallan[m] is with us13:13
dviroelso you splitted the changes13:13
dviroelwe have a proposal for 2 new actions13:13
dviroel1000168: Add new delete and shelve actions | https://review.opendev.org/c/openstack/watcher/+/100016813:13
dviroelwhich sean-k-mooney and I have been reviewing13:14
dviroelit is in a good shape I think, and winiciusallan[m] has been updting based on our comments13:14
dviroelso I guess that is close to be merged13:14
dviroellikely today13:14
dviroelwiniciusallan[m]: so please fix the failing ci jobs13:15
winiciusallan[m]i addressed most of your coments and from sean as well13:15
dviroelwiniciusallan[m]: what about the tempest change?13:15
winiciusallan[m]yes, I just noticed that zuul failed13:15
dviroelpep8, easy fix, no worries13:15
winiciusallan[m]I have the devstack deployment, I'll debug why the tempest test is failing and will do my best to push this today13:15
dviroel#link https://review.opendev.org/c/openstack/watcher-tempest-plugin/+/100218813:15
dviroelack, don't forget to add the config option for these new tests13:16
dviroelsince they can't be skipped based on api microversions13:16
winiciusallan[m]yes, the flag option, right?13:16
dviroelwe need to skip them for all other stable branches13:16
dviroelyes13:17
winiciusallan[m]noted13:17
dviroellet me know if you need help to setup the tempest on your local env13:17
dviroelthe next one is 13:17
dviroel#link https://review.opendev.org/c/openstack/watcher/+/100017013:17
winiciusallan[m]I'll probably need, but I ping you thanks13:17
dviroelwhich adds the workload preemption strategy13:17
dviroelI have been reviewing it, but there are still some threads that we raised some issues13:18
winiciusallan[m]I left some questions mainly about filtering the instances13:18
dviroelit is very likely that we will need a few more days to cover everything, not sure if sean-k-mooney will be able to review it today too13:19
winiciusallan[m]If we would like to exclusively filter them by audit_scope of if we should do this in some way in the code13:19
sean-k-mooneyam i will be spending more tiem reviewing watcher today yes13:19
sean-k-mooneyif the issue in the first patch are adress ill look at the second13:19
dviroelwiniciusallan[m]: so i added a comment about "watcher_exclude=True", as teim-ci pointed out, we should not stop/shelve instances that have "watcher_exclude=True" since ther are exluded via scope13:20
dviroelbut they should be consider as resources allocated in the node/cluster13:21
sean-k-mooneyhum yes..13:21
sean-k-mooneybut i tough we expclied such instnace form the cdm13:21
sean-k-mooneyso they shoudl not be in the set of isntance the stragey sees?13:21
dviroelthey still exists in the cdm to strategies can account there resource allocation too13:22
sean-k-mooneyhum ok so that diffent the the compute nodes13:22
sean-k-mooneywhich we just drop entirly13:22
dviroelcompute nodes are removed yes13:22
sean-k-mooneyack13:22
sean-k-mooneywe may want a helper that returns only the "manageable instances"13:23
dviroelcorrect13:23
sean-k-mooneyinstead of implemnting that filtering in every stragey13:23
dviroelin the cdm itself or in a additional strategy helper13:23
dviroelyes13:23
sean-k-mooneyyep13:23
dviroelwe already have 3 bugs because of that13:23
dviroelbut for now, we need to avoid deleting thenm13:24
sean-k-mooneyack on the resource_id vs server_id13:24
sean-k-mooneyso we need to revert back yes?13:24
dviroelnot sure if winiciusallan[m] already made that change13:24
sean-k-mooneyoh it wasnt actuly updated13:24
dviroelyeah13:24
winiciusallan[m]no, not yet13:25
winiciusallan[m]becuase in the spec we are using resource_id, so I'd like to confirm if that's ok to rename it13:25
dviroeli think that we can keep it13:25
sean-k-mooneyya for now we can keep it as is13:26
sean-k-mooneybased on what dviroel noted13:26
dviroelnot sure if other places in the code really check RESOURCE_ID or hardcoded "resource_id"13:26
dviroelso we may face issue13:26
dviroelwith more time we could try 13:26
dviroelfor now, resource_id should be safer13:27
sean-k-mooneyit was a minor issue13:27
sean-k-mooneyit just looked like over generalisation13:27
sean-k-mooneyi.e. impliging it might be used to delete a volume or something later13:27
dviroelok, if winiciusallan[m] have time to keep working on these changes13:27
sean-k-mooneyi think we can ignore that comment13:27
dviroelit is very likely that https://review.opendev.org/c/openstack/watcher/+/1000170 will need an exception13:27
dviroelshelve/delete actions is feasible to get in I think13:28
dviroelor not, lets see13:28
winiciusallan[m]hope not :)13:28
dviroelwiniciusallan[m]: do you have some time early next week to update the patches if needed?13:28
winiciusallan[m]yes! I'll be likely online on sunday and all monday13:29
dviroelit more about the watcher changes, the tempest ones we can merge later, but would be good to have them validating the features13:29
dviroelwiniciusallan[m]: ack13:29
dviroelsean-k-mooney and others, works for you?13:30
winiciusallan[m]sean-k-mooney: PTAL when you have some time, your reviews are always valuable13:30
winiciusallan[m]I'll try to do a quick turnaround13:30
dviroelnote that monday is holiday in some places, so few os us may not be around13:30
dviroelwe may have joan back on monday too13:31
dviroelok, so  lets keep reviewing it13:31
sean-k-mooneydviroel: i think so13:31
dviroelack13:31
sean-k-mooneyi woudl be reluctant to extend beyond the next meeting but we can check in then13:31
dviroelah yes, me too13:32
dviroelsince winiciusallan[m] will be around, i think that we can solve most of the things earlier13:32
winiciusallan[m]sounds good13:32
dviroelnext feature in the list13:32
dviroel2. Audit Pipeline13:32
dviroelah, this one has a lot of changes 13:32
dviroel#link https://review.opendev.org/q/topic:%22bp/audit-pipeline%22+is:+open13:33
dviroeltopic shows everything, including tempest and watcherclient13:33
dviroelfirst thing13:33
dviroelI don't think we will get the client merged this week13:34
dviroelso we will likely need to wait to have 1.8 in the next release of the client13:34
dviroelsean-k-mooney: ^13:34
dviroelwhich is not a big issue, we also don't have the dashboard support for it yet (but I would like to propose in the next release too)13:35
sean-k-mooneyits ok for the client supprot to come later13:35
sean-k-mooneyit would not eb the first time that the client supprot landed in the cycle after an api feature13:35
sean-k-mooneywe prefer to avoid it but its not a blocker for the main feature to merge13:36
dviroelyeah, I already faced the same thing in manila once13:36
sean-k-mooneyi have left some comment on the inial patch however 13:36
sean-k-mooneyim not convice by the new interface fucntion yoru addign to the base stragies13:36
dviroeli missed this i think13:37
dviroeli was looking at the DB ones13:37
sean-k-mooneyDATASOURCE_METRICS is already aviabel as a class methond and you can access it directly from anywehre you can call get_datasouce_metrics13:37
dviroelit is yes13:38
sean-k-mooneyalso the get_perod(self,resouce) does not match the usage13:38
sean-k-mooneyalmot noting passes resouce 13:38
dviroelsome strategies allow period to be configures per resource13:38
sean-k-mooneyand often its just delegatin to a hard coded value or the property13:38
dviroelas input parameters13:38
dviroelthat was the reason13:38
dviroelso an strategy may have different period values based on the resource13:39
sean-k-mooneyright but this really does not look liek a clean interfce13:39
sean-k-mooneyif we are adopting thse fucntion we shoudl remove the properties13:39
sean-k-mooneyso we dont have 2 ways to get the same info13:39
dviroelyeah, makes sense13:39
dviroelwe can replace them13:39
dviroeli didn't remove to be backward compatible13:40
dviroelis there a risk of breaking things outside the in tree strategies13:40
dviroelor maybe deprecate the properties and remove later13:40
dviroelinstead of just remove them13:41
sean-k-mooneywell nothing outside the tree can inherit form the straties13:41
sean-k-mooneythe only one they can inertit form woudl be base13:41
sean-k-mooneyeven then we do not really provide any stablity garentees to out of tree straties13:41
dviroelcorrect, it seems that period and others are not part of the Base13:42
sean-k-mooneycorrect13:42
dviroelso it should be fine13:42
sean-k-mooneyalso one fo the get_ fucntions i think never uses resouce13:42
sean-k-mooneythe other 2 i belive do in at least one case13:42
dviroelack, but if some strategy start to use it, we would add that later13:44
sean-k-mooneyoh no all 3 have 1+ usage13:44
dviroelah ok13:44
sean-k-mooneying genreal only one13:44
dviroeli don't really remember the details13:44
sean-k-mooneyso if we clean up the duplciaiton i think its fine13:44
dviroelbut I remember that 1 or 2 use them13:44
dviroelack13:44
dviroelagree, since Base does not have them13:44
sean-k-mooneyi woudl almost prefer to not use this and use the input parmater directly however13:45
sean-k-mooneyand just have constants for the kesy but this is ok13:45
dviroelyeah, but input_paramters may not have a pattern to follow13:45
dviroelyeah13:45
sean-k-mooneyi assume we will want to normalise this in the future13:45
dviroelwe should13:45
sean-k-mooneyi.e. make these supproted as input parmater on alls stragies13:45
sean-k-mooneyok in that case the propoal makes more sense13:46
dviroelstill, the get_ could fetch dirctly from input_paramters later13:46
dviroelyeah13:46
sean-k-mooneyyep and if we have a defautl we want to express we can do that13:46
sean-k-mooneyget(resouce,default)13:46
dviroelwe may discuss this more in the next ptg, wrt to filtering and base methods for strategies13:46
sean-k-mooneyack13:46
sean-k-mooneyso the intial patch and db change were the two i looked at yesterday13:47
dviroelthanks a lot13:47
sean-k-mooneyill try and look deeper in the serise13:47
dviroeli think that I agree with yout comments in  the DB patch13:47
dviroeli think that i replied them13:47
sean-k-mooneyso what we could do13:47
sean-k-mooneyis proceed with the first patch as is, and you could do the property cleanup in a followup13:47
dviroelyes13:48
sean-k-mooneyif you can push a draft for that on the end of the sereis ill swap to +2w13:48
dviroelack13:48
dviroeleasier  I think13:48
dviroelsince we already have Alfredo's vote13:48
dviroelthe DB patch, I implemented a bit different, not setting "name" as mandatory in the DB (so i also skipped the generated name for it, which is what Audit does)13:49
dviroelbut we can move to follow the Audit pattern13:50
dviroeland turn a nullable=False and generate a name if user don't provide one13:50
dviroelsame for audit_stage13:50
dviroelI think that covers most of your concerns13:50
dviroeluniqueConstraints make sense too13:51
sean-k-mooneywe coudl jsut set the name to the pipeline name+ postion13:51
sean-k-mooneyif you dont provide one13:51
dviroelaudit aldo set datetime on them13:51
dviroele.g.: workload_balance-2026-07-17T17:38:03.75534413:51
sean-k-mooneythat might be because fo the contious audit and template interaction13:51
sean-k-mooneyif we did that on the pipelien13:52
sean-k-mooneyit woudl hapen automaticlly13:52
dviroelname it is also an unique constraints I think13:52
dviroelfor audit13:52
sean-k-mooneywhat did we say for the api13:52
sean-k-mooneyso i was expectign the api to require the name of the pipelien at a minium13:52
sean-k-mooneybut i dont recall what we said for stages13:52
dviroelso for Audit name  is optional13:53
sean-k-mooneyok well we do not need to requrie a name13:53
dviroelso i was going in the same way for the AuditPipeline api13:53
dviroelbut we always populate one for Audits13:53
sean-k-mooneythe pipeline and postion is enough to enofroce the uiqueness of a stage13:54
dviroelaudit: https://docs.openstack.org/api-ref/resource-optimization/#id1313:54
dviroelagree on the position+pipeline13:54
sean-k-mooneyack im fine with takign the same approch13:54
dviroelack13:54
dviroeli will update the code then in DB and in the API then13:54
dviroellets follow the same thing so it do not turns to be confusing13:55
dviroelack13:55
dviroelthe next thing would be the notifications patch13:55
dviroelwhich can require some changes, but we could also do as follow up13:55
dviroelthe core change would be:13:56
dviroel1. https://review.opendev.org/c/openstack/watcher/+/1000031 Add AuditPipeline execution handler and cascade execution mode13:56
dviroelthe handle logic13:56
dviroel2. https://review.opendev.org/c/openstack/watcher/+/1001009 - Add efficacy indicators for Audit Pipeline action plans13:57
dviroeland 13:57
dviroel3. https://review.opendev.org/c/openstack/watcher/+/997427 - Add Audit Pipeline REST API endpoints 13:57
dviroelwhich I also didn't receive feedbacks from other reviewers13:57
sean-k-mooneythe only feedback i had on the notification patch is we proably over engenerired it and we shoudl get rid of the terse class13:57
sean-k-mooneyand jsut null out the stages when we want to be breif13:58
dviroelit is possible13:58
sean-k-mooneythe thing is the class information is encode in the object13:58
dviroelit is the way that other notifications were built13:58
sean-k-mooneyand im not sure client will wnat to have to treate the terse and verbose classes seperatly13:58
dviroelyeah, the major difference is the stages info13:59
dviroelthat a client could ignore13:59
sean-k-mooneywe can declare it as a nullable/optional list of stage objects13:59
sean-k-mooneyand then the client can just check for null/None before trying to use it14:00
sean-k-mooneythat what we do in nova14:00
sean-k-mooneybut either is fien14:00
dviroelack14:00
sean-k-mooneyim not going to block on that14:00
sean-k-mooneyill see if i can revew the remaining changes today14:00
dviroelyeah, my feeling is that we saw some decisions in the past that we can improve, while adding new stuff14:01
sean-k-mooneybut i may not have tiem14:01
dviroelack thanks sean-k-mooney 14:01
dviroelso this is another feature that will require a FFE14:01
sean-k-mooneywe could follow the existing pattern for now and revist next cycle14:01
dviroelthere is a risk of not getting even next week, but that would be fine14:01
dviroelwe can see if Joan can also review it when he comes back14:02
sean-k-mooneyi think up until the next meeting woudl also be ok14:02
dviroelalright14:02
sean-k-mooneyat which point i stongly suggest we punt14:02
sean-k-mooneybut yes joan will be back next week14:02
sean-k-mooneyso that allows another pair of eye to review14:02
dviroelack14:02
dviroelbefore we close the meeting14:03
dviroelsince we also lack more reviewers for the client14:03
dviroelI think  i will unblock the client release today14:03
dviroelwdyt sean-k-mooney?14:03
sean-k-mooneyi think its fien for that to proceed14:04
dviroelyeah, no need to rush things I guess14:04
sean-k-mooneywe can alwasy do an early release next cycle14:04
dviroelrigh14:04
dviroelok we are out of time14:04
sean-k-mooneythe other part of this14:04
dviroelyeah?14:05
sean-k-mooneyis we will eventully need to add pipeline supprot to watcher-dashboard14:05
sean-k-mooneyso that will defintly not be this cycle14:05
dviroelyeah, I mentioned that too14:05
sean-k-mooneybut that is ok14:05
dviroelack14:05
sean-k-mooneywell i asume you dont have it implemnted with playright tests?14:05
dviroelwe may have both next cycle, i can work in the dashboard 14:05
dviroelnope14:05
sean-k-mooneyya i think its best to take our time with that14:05
sean-k-mooneyand not rush14:06
dviroelack14:06
dviroelthat was my idea too14:06
dviroelso, et's wrap up for today14:06
dviroelwe can continue to chat here in the irc about the reviews14:06
dviroelwe will meet again next week14:06
dviroelthank you all for participating14:06
dviroel#endmeeting14:06
opendevmeetMeeting ended Thu Aug 27 14:06:53 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:06
opendevmeetMinutes:        https://meetings.opendev.org/meetings/watcher/2026/watcher.2026-08-27-13.01.html14:06
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/watcher/2026/watcher.2026-08-27-13.01.txt14:06
opendevmeetLog:            https://meetings.opendev.org/meetings/watcher/2026/watcher.2026-08-27-13.01.log.html14:06
opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher master: Add new delete and shelve actions  https://review.opendev.org/c/openstack/watcher/+/100016814:47
opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher master: Add Workload Optimization goal and Workload Preemption strategy  https://review.opendev.org/c/openstack/watcher/+/100017014:47
opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher-tempest-plugin master: Add tests for delete and shelve actions  https://review.opendev.org/c/openstack/watcher-tempest-plugin/+/100218815:03
opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher master: Add new delete and shelve actions  https://review.opendev.org/c/openstack/watcher/+/100016815:07
opendevreviewWinicius Allan Bezerra da Silva proposed openstack/watcher master: Add Workload Optimization goal and Workload Preemption strategy  https://review.opendev.org/c/openstack/watcher/+/100017015:07

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!