Displaying #cassandra-dev/2017-08-16.log:

Wed Aug 16 00:14:26 2017  kohlisankalp:Joined the channel
Wed Aug 16 00:17:56 2017  RussSpitzer_:Joined the channel
Wed Aug 16 01:22:13 2017  RussSpitzer:Joined the channel
Wed Aug 16 01:22:49 2017  RussSpitzer_:Joined the channel
Wed Aug 16 02:12:22 2017  kohlisankalp:Joined the channel
Wed Aug 16 02:32:36 2017  jayzhuang:Joined the channel
Wed Aug 16 02:47:15 2017  RussSpitzer:Joined the channel
Wed Aug 16 03:33:22 2017  kohlisankalp:Joined the channel
Wed Aug 16 04:43:08 2017  adamholmberg:Joined the channel
Wed Aug 16 04:53:27 2017  adamholmberg:Joined the channel
Wed Aug 16 05:32:23 2017  jayzhuang:Joined the channel
Wed Aug 16 05:37:49 2017  kvaster:Joined the channel
Wed Aug 16 06:14:10 2017  kohlisankalp:Joined the channel
Wed Aug 16 06:49:47 2017  spodkowinski:Joined the channel
Wed Aug 16 08:45:52 2017  cosql:Joined the channel
Wed Aug 16 09:04:03 2017  kvaster:Joined the channel
Wed Aug 16 09:11:53 2017  RussSpitzer:Joined the channel
Wed Aug 16 09:24:36 2017  kvaster:Joined the channel
Wed Aug 16 11:26:32 2017  ostefano:I am working on https://issues.apache.org/jira/browse/CASSANDRA-13043
Wed Aug 16 11:26:38 2017  ostefano:Basically trying to solve a race where a counter mutation is assigned to a loader which has yet no full visibility over the cluster
Wed Aug 16 11:26:44 2017  ostefano:Is there a way for a node to know whether another endpoint is actually accepting client requests (or alternatively its gossiping actually settled)?
Wed Aug 16 11:27:30 2017  ostefano:The snitch can't really help me here
Wed Aug 16 11:31:18 2017  ostefano:s/loader/leader :)
Wed Aug 16 12:47:36 2017  adamholmberg:Joined the channel
Wed Aug 16 12:50:16 2017  RussSpitzer_:Joined the channel
Wed Aug 16 12:56:31 2017  RussSpitzer:Joined the channel
Wed Aug 16 14:01:46 2017  iamaleksey:ostefano: I don't think there is a way, off the top of my head. all we have is UP
Wed Aug 16 14:02:20 2017  iamaleksey:ostefano: but I'm no gossip expert. driftx might know
Wed Aug 16 14:13:40 2017  clohfink:post 2.2 it at least holds on announcing its up to cluster until after the native interface has started providing its enabled
Wed Aug 16 14:16:07 2017  jeffj:Joined the channel
Wed Aug 16 14:17:37 2017  clohfink:CASSANDRA-8236
Wed Aug 16 14:17:38 2017  CassBotJr:https://issues.apache.org/jira/browse/CASSANDRA-8236 (Resolved; Fixed; 2.2.0 beta 1): "Delay "node up" and "node added" notifications until native protocol server is started"
Wed Aug 16 15:40:26 2017  adamholmberg:Joined the channel
Wed Aug 16 15:41:25 2017  adamholmberg:Joined the channel
Wed Aug 16 15:55:54 2017  adamholm_:Joined the channel
Wed Aug 16 15:58:54 2017  jeffj:Joined the channel
Wed Aug 16 16:04:16 2017  ostefano:clohfink, can that happen before the gossip settles? I see the leader receiving those mutations way before it finishes updating the toopology
Wed Aug 16 16:07:39 2017  jeffj:Joined the channel
Wed Aug 16 16:16:35 2017  adamholmberg:Joined the channel
Wed Aug 16 16:19:25 2017  kohlisankalp:Joined the channel
Wed Aug 16 16:20:57 2017  clohfink:i think its more of an issue in that it can send notification to the clients that the host is up, but the cql interface not being up means the client fails to connect
Wed Aug 16 16:39:53 2017  ostefano:clohfink, those mutations are not coming from the client, but from another node. When the mutations are received, the consistency check fails because the node itself has a partial view of the cluster (its gossip didn't settle yet)
Wed Aug 16 16:40:07 2017  jayzhuang:Joined the channel
Wed Aug 16 16:40:51 2017  clohfink:ssl cluster by chance?
Wed Aug 16 16:48:25 2017  adamholmberg:Joined the channel
Wed Aug 16 16:50:42 2017  adamholmberg:Joined the channel
Wed Aug 16 17:01:50 2017  kohlisankalp:Joined the channel
Wed Aug 16 17:18:32 2017  adamholm_:Joined the channel
Wed Aug 16 17:23:15 2017  zaller:Joined the channel
Wed Aug 16 17:32:50 2017  driftx:ostefano: there's a RPC_READY state in gossip to let you know
Wed Aug 16 17:33:09 2017  ostefano:clohfink, nope :/
Wed Aug 16 17:33:36 2017  ostefano:driftx , that might do actually, thx!
Wed Aug 16 17:34:05 2017  driftx:cool
Wed Aug 16 17:34:42 2017  ostefano:I'am going to try to replicate the race locally. dtest is the way to go afaik, anything I should know?
Wed Aug 16 17:35:01 2017  driftx:python? ;)
Wed Aug 16 17:36:28 2017  ostefano:hehe, fair enough :) I meant as examples/howto approach it
Wed Aug 16 17:53:11 2017  driftx:plenty of existing dtests for examples
Wed Aug 16 18:02:10 2017  kohlisan_:Joined the channel
Wed Aug 16 18:39:34 2017  kvaster:Joined the channel
Wed Aug 16 18:49:55 2017  cosql:Joined the channel
Wed Aug 16 18:52:19 2017  echoreply:Joined the channel
Wed Aug 16 19:56:29 2017  kohlisankalp:Joined the channel
Wed Aug 16 20:02:19 2017  jeffj:ostefano: grep for byteman in dtests, it's one of the only ways to make races deterministic (by artificially injecting pauses in one instance or another)
Wed Aug 16 20:03:07 2017  ostefano:thx jeffj , will do
Wed Aug 16 20:32:58 2017  kohlisankalp:Joined the channel
Wed Aug 16 20:35:43 2017  jeffj:Joined the channel
Wed Aug 16 20:36:21 2017  jeffj_:Joined the channel
Wed Aug 16 20:54:15 2017  jeffj:Joined the channel
Wed Aug 16 20:57:16 2017  jeffj_:Joined the channel
Wed Aug 16 21:07:16 2017  ynazarov:Joined the channel
Wed Aug 16 22:01:44 2017  adamholmberg:Joined the channel
Wed Aug 16 22:09:03 2017  kohlisankalp:Joined the channel
Wed Aug 16 22:09:09 2017  iamaleksey:do we still shoot out RPC_READY transtition if rpc is disabled?
Wed Aug 16 22:21:21 2017  kohlisankalp:Joined the channel
Wed Aug 16 22:48:21 2017  kohlisankalp:Joined the channel
Wed Aug 16 23:29:45 2017  jayzhuang:Joined the channel

Comments