| *** openstack has joined #openstack-oslo | 00:53 | |
| *** dims_ has joined #openstack-oslo | 00:53 | |
| *** bogdando has joined #openstack-oslo | 00:54 | |
| *** openstack has joined #openstack-oslo | 00:56 | |
| *** openstack has quit IRC | 00:57 | |
| *** openstack has joined #openstack-oslo | 00:59 | |
| *** rpodolyaka has joined #openstack-oslo | 00:59 | |
| *** hdd_ has joined #openstack-oslo | 01:02 | |
| *** rpodolyaka1 has joined #openstack-oslo | 01:07 | |
| *** rpodolyaka has quit IRC | 01:08 | |
| *** rpodolyaka1 has quit IRC | 01:14 | |
| *** openstack has joined #openstack-oslo | 01:15 | |
| *** openstack has joined #openstack-oslo | 01:19 | |
| *** openstack has quit IRC | 01:24 | |
| *** openstack has joined #openstack-oslo | 01:25 | |
| *** openstack has joined #openstack-oslo | 15:09 | |
| *** matrohon has joined #openstack-oslo | 15:09 | |
| *** banix has joined #openstack-oslo | 15:10 | |
| ihrachys | markmcclain: I've checked cinder and nova porting patches to see how they solved the problem, and they didn't... because they don't create rpc connections in random places, instead relying on services to invoke proper methods [this can be my incorrect understanding though, I haven't checked their code for too long] | 15:11 |
|---|---|---|
| markmcclain | ihrachys: we probably should clean up how we start RPC | 15:11 |
| markmcclain | would it make sense to add something similar to init_db that arosen added to the plugins? | 15:12 |
| dhellmann | ihrachys: that's why I wondered if the plugin was subclassed from something like MessageHandlingServer | 15:12 |
| markmcclain | also for the unit tests we shouldn't be spawning RPC threads anyway | 15:12 |
| dhellmann | there is setup_rpc(), there's just nothing cleaning it up when the plugin is "done" | 15:13 |
| ihrachys | markmcclain: yes, probably, though oslo.messaging does not have fake executor at the moment anyway | 15:13 |
| markmcclain | ihrachys: here's what I'm thinking | 15:14 |
| ihrachys | markmcclain: whatis init_db? I failed to find it in source tree | 15:14 |
| *** yamahata has quit IRC | 15:14 | |
| *** yamahata has joined #openstack-oslo | 15:14 | |
| markmcclain | ihrachys: sorry wrong method name.. he folded it into the base class' init | 15:15 |
| *** rpodolyaka has joined #openstack-oslo | 15:15 | |
| markmcclain | https://github.com/openstack/neutron/blob/master/neutron/db/db_base_plugin_v2.py#L224 | 15:15 |
| markmcclain | ihrachys: so right now we're in the process of revising/defining a new version of the plugin's class API | 15:16 |
| markmcclain | ihrachys: might make sense to pause on your current work as we work out what the design plans | 15:17 |
| markmcclain | we can incorporate the design changes needs to enable you to port to oslo.messaging | 15:18 |
| ihrachys | markmcclain: dhellmann: also I've noticed that tests run a lot longer with oslo.messaging, so now I wonder whether it may be the case of unused threads spawned in tests... | 15:18 |
| markmcclain | ihrachys: I'm hoping we will be able to share a draft plan by the end of the month with community would that work? | 15:18 |
| markmcclain | ihrachys: likely | 15:18 |
| ihrachys | markmcclain: dhellmann: and 'a lot' is like '2.5 times longer' | 15:19 |
| markmcclain | ihrachys: ouch that's a painful delay | 15:19 |
| ihrachys | + memory usage is a lot higher (rpc servers unstopped?) | 15:19 |
| dhellmann | ihrachys: that seems possible, but we haven't seen that elsewhere so I wonder if it is related to not stopping them? | 15:19 |
| markmcclain | my guess is the clean-up between tests is leaving lots of objects around | 15:20 |
| markmcclain | we already had a few subtle leaks via cyclical obj relationships | 15:20 |
| ihrachys | actually, this is what took me here to discuss rpc connections in plugins :) I was in the middle of debugging slowliness issue when caught those greenlets in the logs. | 15:20 |
| ihrachys | dhellmann: that's exactly my convern - getting connections sitting there in memory | 15:21 |
| ihrachys | markmcclain: btw I've seen TestCases in the list of object types left after the run. why are they even left sitting there AFTER they are run? | 15:22 |
| ihrachys | markmcclain: yes, master also leaks, just not that much as with my patch | 15:22 |
| markmcclain | ihrachys: the garbage collector is unable to reap them | 15:23 |
| ihrachys | in 2012, we had this patch applied to reduce memory usage https://review.openstack.org/#/c/16735/3/quantum/tests/unit/test_db_plugin.py | 15:23 |
| ihrachys | and frankly, in the bug linked, a person claimed that it was enough to have ~200 MB RES RAM to run tests | 15:23 |
| markmcclain | right that was only a bandaid to the larger problem | 15:23 |
| ihrachys | now it uses like gigabytes of RAM... | 15:23 |
| markmcclain | yes that is problem that marun and others have been thinking about how to solve | 15:24 |
| markmcclain | ihrachys: so are you ok pausing on your current work for now and we can revisit at the end of the month when we start talking about API framework changes? | 15:27 |
| ihrachys | markmcclain: without solving this rpc cleanup issue I can't proceed to get the patch in review phase anyway, so probably we'll live with it :) | 15:28 |
| markmcclain | ihrachys: ok.. when the 1st design draft is published I'll ping you to get your feedback | 15:29 |
| seagulls | it probably would be best to collaborate to find a closer-to-optimal solution than beat our heads against the wall trying to hack our way through | 15:29 |
| ihrachys | this oslo.messaging port looked quite easy at first sight when I started :) [I've checked cinder and glance and nova patches at that moment and it looked quite easy to achieve the same in neutron] | 15:29 |
| ihrachys | markmcclain: thanks! | 15:29 |
| seagulls | it's a very worthwhile effort nonetheless, ihrachys has uncovered/exposed lots of things to discuss and improve | 15:29 |
| ihrachys | dhellmann: but still, it would be probably great to see fake executor available. it's not nice to run threads in tests. | 15:30 |
| dhellmann | ihrachys: if you can move the thread management out of the plugins, then the unit tests won't start threads | 15:31 |
| * seagulls nods | 15:31 | |
| seagulls | I've seen similar snafus in dealing with CORBA and other OOM systems. where folks put OOM initialization code in their servants | 15:33 |
| seagulls | it always ends in tears | 15:33 |
| seagulls | we wouldn't run into the same issues here, but there are issues.. obviously | 15:34 |
| ihrachys | dhellmann: moving rpc connections creation in separate function can be the way to go. but then something should call this separated method (and stop method too), and I think this is where we lack understanding how to achieve this | 15:34 |
| dhellmann | ihrachys: why does the plugin need to create those connections at all? why isn't the plugin being called by something that creates the connections, receives the messages, and then needs the plugin to do the work? | 15:35 |
| ihrachys | dhellmann: surely, this is how it should look like - separate setup_rpc/cleanup_rpc method called from outside by some manager | 15:38 |
| dhellmann | ihrachys: no, that's not what I mean at all | 15:38 |
| dhellmann | ihrachys: the plugin should not need to create or manage its own rpc connections | 15:38 |
| dhellmann | the plugin shouldn't even need to know it is being fed data from rpc | 15:39 |
| dhellmann | some common core piece should handle all of that | 15:39 |
| seagulls | +1 | 15:39 |
| seagulls | where possible business logic implementation, should not know or implement anything regarding the specifics of how it is being called | 15:40 |
| dhellmann | seagulls: right | 15:41 |
| seagulls | it can then be used in any manner as seen fit without extra consideration | 15:41 |
| seagulls | it is more testable, easier to write, etc. etc. etc | 15:41 |
| *** rpodolyaka has quit IRC | 15:42 | |
| ihrachys | ah, ok. plugins will just provide the list of endpoints, right? | 15:42 |
| dhellmann | ihrachys: right | 15:42 |
| *** hartsocks has quit IRC | 15:43 | |
| *** hartsocks has joined #openstack-oslo | 15:44 | |
| * markmcclain reads scrollback | 15:44 | |
| markmcclain | dhellmann: so one pattern is some plugins use RPC to talk to proprietary agents, so we'll need to have a interface for a plugin to register callbacks | 15:45 |
| *** drumkilla is now known as rustlebee | 15:47 | |
| seagulls | markmcclain, are you referring to a call flow of RPC-Layer->Endpoint->proprietary-agent or something else? | 15:48 |
| *** markmcclain has quit IRC | 15:50 | |
| openstackgerrit | Anastasia Karpinska proposed a change to openstack/taskflow: Move flow states to developers documentation https://review.openstack.org/80350 | 16:07 |
| *** salv-mobile has quit IRC | 16:17 | |
| openstackgerrit | Terry Howe proposed a change to openstack/cliff: Add value only output formattter https://review.openstack.org/80614 | 16:19 |
| *** rpodolyaka has joined #openstack-oslo | 16:25 | |
| *** openstackstatus has joined #openstack-oslo | 16:38 | |
| openstackgerrit | Cyril Roelandt proposed a change to openstack/oslo-incubator: Python3: pass bytes as 'contents' to create_tempfiles() https://review.openstack.org/80627 | 16:46 |
| *** rpodolyaka has quit IRC | 16:51 | |
| *** dims has joined #openstack-oslo | 16:54 | |
| *** matrohon has quit IRC | 16:55 | |
| *** saikrishna_ has joined #openstack-oslo | 16:56 | |
| *** rpodolyaka has joined #openstack-oslo | 16:58 | |
| *** dims has quit IRC | 17:00 | |
| *** markmcclain has joined #openstack-oslo | 17:00 | |
| *** markmcclain has quit IRC | 17:00 | |
| *** markmcclain has joined #openstack-oslo | 17:01 | |
| *** harlowja_away is now known as harlowja | 17:02 | |
| openstackgerrit | Brant Knudson proposed a change to openstack/oslo-incubator: Add connection trace for DB2 https://review.openstack.org/70163 | 17:21 |
| *** morganfainberg_Z is now known as morganfainberg | 17:21 | |
| beekneemech | Hmm, I think these Python 3 changes probably shouldn't go in during feature freeze. | 17:30 |
| beekneemech | It isn't something we support now, so it's really a new feature, and it's going to add noise to any critical syncs to other projects. | 17:31 |
| *** dshulyak has quit IRC | 17:34 | |
| *** mriedem has quit IRC | 17:37 | |
| *** mriedem has joined #openstack-oslo | 17:39 | |
| *** dims has joined #openstack-oslo | 17:40 | |
| *** rpodolyaka has quit IRC | 17:45 | |
| ekarlso | is it possible to do a rpc.call() towards a host ? | 17:45 |
| ekarlso | or is that only in .cast | 17:45 |
| *** saikrishna_ has quit IRC | 17:47 | |
| *** rpodolyaka has joined #openstack-oslo | 17:48 | |
| dhellmann | beekneemech: +1 | 17:52 |
| openstackgerrit | Anastasia Karpinska proposed a change to openstack/taskflow: Move flow states to developers documentation https://review.openstack.org/80350 | 18:17 |
| *** hartsocks is now known as hartbot | 18:19 | |
| openstackgerrit | Solly Ross proposed a change to openstack/oslo-incubator: Report: Add basic processes section https://review.openstack.org/78793 | 18:34 |
| openstackgerrit | Solly Ross proposed a change to openstack/oslo-incubator: Report: Propagate view_type through collections https://review.openstack.org/78792 | 18:34 |
| openstackgerrit | Solly Ross proposed a change to openstack/oslo-incubator: Report: Support Sequences and Mappings as Data https://review.openstack.org/78791 | 18:34 |
| openstackgerrit | Joshua Harlow proposed a change to openstack/taskflow: Move change state to after storage calls https://review.openstack.org/80657 | 18:36 |
| *** seagulls is now known as seagulls_coffee | 18:38 | |
| *** rpodolyaka has quit IRC | 18:39 | |
| *** rpodolyaka has joined #openstack-oslo | 18:47 | |
| *** seagulls_coffee is now known as seagulls | 18:59 | |
| openstackgerrit | Joshua Harlow proposed a change to openstack/taskflow: Add graph property to all flow patterns https://review.openstack.org/72532 | 19:08 |
| *** rpodolyaka has quit IRC | 19:14 | |
| *** rpodolyaka has joined #openstack-oslo | 19:17 | |
| *** rpodolyaka has quit IRC | 19:25 | |
| *** rpodolyaka has joined #openstack-oslo | 19:26 | |
| *** rpodolyaka has quit IRC | 19:47 | |
| *** Oslomann29 has joined #openstack-oslo | 19:48 | |
| *** rpodolyaka has joined #openstack-oslo | 19:49 | |
| *** yamahata has quit IRC | 20:07 | |
| *** jraim has joined #openstack-oslo | 20:09 | |
| *** rpodolyaka has quit IRC | 20:10 | |
| jraim | dhellmann: question about oslo.logging if you have a sec | 20:11 |
| dhellmann | jraim: sure | 20:11 |
| jraim | we had this PR put in | 20:11 |
| jraim | https://review.openstack.org/#/c/79011/ | 20:11 |
| jraim | which is complaining about a problem that logging is causing in heat templates and we're trying to figure out the right way to fix it | 20:12 |
| * dhellmann looks | 20:13 | |
| dhellmann | so the barbican client is registering command line options at import time? | 20:13 |
| *** Oslomann29 has left #openstack-oslo | 20:14 | |
| dhellmann | oh, no, it imports oslo.log and we do that | 20:14 |
| dhellmann | ok, here's my juno plan for oslo.log: | 20:14 |
| dhellmann | We are starting to move logic from the ContextAdapter to ContextFormatter. When that is done, projects won't need to ask oslo.log for a log handler, since the formatter will be registered by the application and won't need to be attached to each handle. | 20:15 |
| *** rpodolyaka has joined #openstack-oslo | 20:15 | |
| dhellmann | At that point, libraries can stop using oslo.log and just use python's logging module. Applications would need to use oslo.log to set up the formatter and handler, but they would only have to do that at the start of the process. | 20:16 |
| jraim | that makes sense | 20:16 |
| *** rpodolyaka has quit IRC | 20:16 | |
| dhellmann | so I think removing oslo logging from the barbican client is the right approach | 20:16 |
| jraim | alright, we'll remove it for now and add it back in later if needed | 20:17 |
| *** hartbot has left #openstack-oslo | 20:17 | |
| dhellmann | jraim: ok | 20:17 |
| jraim | thanks for the advice | 20:17 |
| dhellmann | jraim: any time! | 20:17 |
| *** rpodolyaka has joined #openstack-oslo | 20:18 | |
| *** rpodolyaka has quit IRC | 20:48 | |
| *** mriedem has left #openstack-oslo | 21:16 | |
| *** mriedem has quit IRC | 21:16 | |
| *** arnaud has joined #openstack-oslo | 21:20 | |
| *** openstackgerrit has quit IRC | 21:21 | |
| *** openstackgerrit has joined #openstack-oslo | 21:21 | |
| *** openstackgerrit has quit IRC | 21:32 | |
| *** openstackgerrit has joined #openstack-oslo | 21:32 | |
| *** dims has quit IRC | 21:34 | |
| openstackgerrit | Zane Bitter proposed a change to openstack/oslo-incubator: Config generator: handle options in multiple groups https://review.openstack.org/79738 | 21:38 |
| openstackgerrit | Terry Howe proposed a change to openstack/cliff: Add value only output formattter https://review.openstack.org/80614 | 21:42 |
| *** zhiyan is now known as zhiyan_ | 21:42 | |
| *** rpodolyaka has joined #openstack-oslo | 21:47 | |
| *** dims has joined #openstack-oslo | 21:53 | |
| openstackgerrit | Joshua Harlow proposed a change to openstack/taskflow: Adjust the exception hierachy https://review.openstack.org/80372 | 22:07 |
| openstackgerrit | Joshua Harlow proposed a change to openstack/taskflow: Adjust the exception hierachy https://review.openstack.org/80372 | 22:08 |
| * beekneemech -2's all the things | 22:12 | |
| beekneemech | I wasn't sure what to do with the Python 3 test-only changes though. There's technically an argument to be made that those don't fall afoul of the feature freeze restrictions since they aren't synced to other projects. | 22:12 |
| beekneemech | For the moment I just left them alone. | 22:13 |
| *** banix has quit IRC | 22:18 | |
| * beekneemech is fried and leaves for the weekend | 22:19 | |
| beekneemech | See y'all Monday :-) | 22:19 |
| *** rpodolyaka has quit IRC | 22:26 | |
| *** ihrachys has quit IRC | 22:26 | |
| openstackgerrit | Joshua Harlow proposed a change to openstack/taskflow: Adjust logging levels and usage to follow standards https://review.openstack.org/80497 | 22:48 |
| *** rpodolyaka has joined #openstack-oslo | 22:49 | |
| *** flaper87 is now known as flaper87|afk | 23:27 | |
| *** rpodolyaka has quit IRC | 23:32 | |
| *** openstackstatus has quit IRC | 23:36 | |
| *** openstackgerrit has quit IRC | 23:39 | |
| *** arnaud has quit IRC | 23:43 | |
| *** lbragstad has quit IRC | 23:46 | |
| *** seagulls has quit IRC | 23:46 | |
| *** bogdando has quit IRC | 23:46 | |
| *** sergmelikyan has quit IRC | 23:46 | |
| *** harlowja has quit IRC | 23:46 | |
| *** bknudson has quit IRC | 23:46 | |
| *** jd__ has quit IRC | 23:46 | |
| *** zhiyan_ has quit IRC | 23:46 | |
| *** dhellmann has quit IRC | 23:46 | |
| *** clarkb has quit IRC | 23:46 | |
| *** wingwj has quit IRC | 23:46 | |
| *** beav has quit IRC | 23:46 | |
| *** viktors_ has quit IRC | 23:46 | |
| *** yamahata_ has quit IRC | 23:46 | |
| *** flaper87|afk has quit IRC | 23:46 | |
| *** jeblair has quit IRC | 23:46 | |
| *** dteselkin has quit IRC | 23:46 | |
| *** SergeyLukjanov has quit IRC | 23:46 | |
| *** ruhe has quit IRC | 23:46 | |
| *** gilliard has quit IRC | 23:46 | |
| *** sileht has quit IRC | 23:46 | |
| *** jraim has quit IRC | 23:46 | |
| *** GheRivero has quit IRC | 23:46 | |
| *** openstack has joined #openstack-oslo | 23:46 | |
| *** rpodolyaka1 has joined #openstack-oslo | 23:47 | |
| *** openstackstatus has joined #openstack-oslo | 23:47 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!