*** bobh has quit IRC | 01:33 | |
*** bobh has joined #openstack-mistral | 02:20 | |
*** jtomasek has quit IRC | 02:48 | |
*** jtomasek has joined #openstack-mistral | 03:03 | |
*** bobh has quit IRC | 03:03 | |
*** tonytan_brb has joined #openstack-mistral | 03:56 | |
*** tonytan4ever has quit IRC | 03:59 | |
*** bobh has joined #openstack-mistral | 04:04 | |
*** bobh has quit IRC | 04:09 | |
*** jtomasek has quit IRC | 04:12 | |
*** bobh has joined #openstack-mistral | 05:05 | |
*** bobh has quit IRC | 05:10 | |
*** tonytan_brb has quit IRC | 05:28 | |
*** bobh has joined #openstack-mistral | 06:06 | |
*** bobh has quit IRC | 06:11 | |
*** tonytan4ever has joined #openstack-mistral | 06:29 | |
*** tonytan4ever has quit IRC | 06:33 | |
rakhmerov | therve: I did not understand your point about patching level | 06:52 |
---|---|---|
rakhmerov | what do you mean? | 06:53 |
*** bobh has joined #openstack-mistral | 07:08 | |
*** bobh has quit IRC | 07:13 | |
*** bobh has joined #openstack-mistral | 08:09 | |
*** bobh has quit IRC | 08:13 | |
*** bobh has joined #openstack-mistral | 09:09 | |
openstackgerrit | Merged openstack/mistral: Updating mistralclient docs https://review.openstack.org/370821 | 09:13 |
*** bobh has quit IRC | 09:14 | |
rakhmerov | therve, ddeja: let's talk on Monday when you're available. It's a sensitive problem, we need to think twice before fixing it. But what Dawid suggested seems like a valid idea (using async in executor) | 09:20 |
rakhmerov | we can also make RPC executor configurable although I don't really like this idea. With 'eventlet' Mistral engine dies under certain conditions. I can explain why | 09:20 |
*** tonytan4ever has joined #openstack-mistral | 09:47 | |
*** tonytan4ever has quit IRC | 09:52 | |
*** bobh has joined #openstack-mistral | 10:10 | |
*** bobh has quit IRC | 10:15 | |
*** bobh has joined #openstack-mistral | 11:11 | |
*** bobh has quit IRC | 11:16 | |
therve | rakhmerov, I meant how much you patch stuff with evenlet | 11:42 |
therve | If you monkey patch threads, the 'threading' option ought to look a lot like the 'eventlet' one | 11:43 |
therve | If not, well it behaves differently | 11:43 |
rakhmerov | ok, I see | 11:43 |
rakhmerov | yes | 11:43 |
rakhmerov | eventlet.monkey_patch( | 11:43 |
rakhmerov | os=True, | 11:43 |
rakhmerov | select=True, | 11:43 |
rakhmerov | socket=True, | 11:43 |
rakhmerov | thread=False if '--use-debugger' in sys.argv else True, | 11:43 |
rakhmerov | time=True) | 11:43 |
rakhmerov | this is what we do | 11:43 |
rakhmerov | so yes, we patch threads | 11:44 |
rakhmerov | the reason I switched to 'blocking' is because I found some weird behavior when using 'eventlet' | 11:44 |
therve | OK, then it ought to be similar | 11:44 |
therve | Yeah I saw your comment, and your concerns are valid | 11:45 |
rakhmerov | green threads can easily get into a deadlock | 11:45 |
rakhmerov | yeah, ok | 11:45 |
therve | Except that it breaks several other things | 11:45 |
rakhmerov | what other things do you mean? Besides nested actions? | 11:45 |
therve | Well the latest reproducer doesn't involve nested actions AFAICT | 11:46 |
rakhmerov | hm.. maybe I didn't read carefully | 11:46 |
rakhmerov | ok | 11:46 |
rakhmerov | one of the things we can do I think pretty easily is configure RPC for engine and executor slightly differently | 11:47 |
rakhmerov | executor with 'eventlet' option (and it's recommended to use executors in separate processes) and engine with 'blocking' RPC | 11:47 |
rakhmerov | for now | 11:47 |
therve | I feel that still makes the engine a bottleneck, but I don't know all the details | 11:48 |
rakhmerov | bottleneck, yes, but if we can scale it it's no longer a bottleneck | 11:49 |
rakhmerov | well, anyway, we're moving in that direction to make it stable and scalable | 11:49 |
therve | Will we be able to do that in newton? | 11:49 |
rakhmerov | we'll get there | 11:49 |
rakhmerov | what exactly? | 11:49 |
therve | Run several engines | 11:49 |
rakhmerov | doing my best to make it happen | 11:50 |
rakhmerov | well, actually we can run multiple engines now | 11:50 |
rakhmerov | but | 11:50 |
rakhmerov | if you have many 'join' tasks in workflows and they are heavy (depends on deep subgraphs) then you'll be facing issues potentially | 11:50 |
rakhmerov | under load | 11:51 |
rakhmerov | if you don't have joins at all you can run them w/o problems (multiple engines) | 11:53 |
*** bobh has joined #openstack-mistral | 12:12 | |
*** bobh has quit IRC | 12:16 | |
openstackgerrit | Merged openstack/mistral-lib: Small changes like deletion of extra underline in the docs https://review.openstack.org/371571 | 12:44 |
*** tonytan4ever has joined #openstack-mistral | 12:49 | |
*** tonytan4ever has quit IRC | 12:53 | |
*** bobh has joined #openstack-mistral | 13:12 | |
*** bobh has quit IRC | 13:17 | |
*** tonytan4ever has joined #openstack-mistral | 13:49 | |
*** tonytan4ever has quit IRC | 13:54 | |
*** tonytan4ever has joined #openstack-mistral | 14:01 | |
*** bobh has joined #openstack-mistral | 14:18 | |
*** tonytan4ever has quit IRC | 14:30 | |
*** tonytan4ever has joined #openstack-mistral | 14:50 | |
*** tonytan4ever has quit IRC | 14:56 | |
*** tonytan4ever has joined #openstack-mistral | 14:56 | |
*** bobh has quit IRC | 15:19 | |
*** janki has joined #openstack-mistral | 15:22 | |
*** janki has quit IRC | 15:48 | |
*** bobh has joined #openstack-mistral | 16:46 | |
*** stevebaker has quit IRC | 17:27 | |
*** szaher has quit IRC | 17:27 | |
*** stevebaker has joined #openstack-mistral | 17:29 | |
*** szaher has joined #openstack-mistral | 17:29 | |
*** bobh has quit IRC | 20:58 | |
*** kong has joined #openstack-mistral | 21:44 | |
*** bobh has joined #openstack-mistral | 21:59 | |
*** bobh has quit IRC | 22:03 | |
*** bobh has joined #openstack-mistral | 23:00 | |
*** bobh has quit IRC | 23:05 | |
*** chlong has joined #openstack-mistral | 23:12 | |
*** bobh has joined #openstack-mistral | 23:29 |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!