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

Tue Aug 8 00:26:49 2017  krut_:hmm yeah haven't heard anything back from infra about 8-14... surely it doesn't take that long to run puppet -_-
Tue Aug 8 00:28:19 2017  adamholmberg:Joined the channel
Tue Aug 8 00:29:49 2017  jeffj_:halp halp how i pssh puppet-agent
Tue Aug 8 00:32:13 2017  kohlisankalp:Joined the channel
Tue Aug 8 00:33:51 2017  krut_:who am i kidding puppet is a joke
Tue Aug 8 00:34:13 2017  krut_:they are probably still debugging why it's trying to deploy the wrong modules
Tue Aug 8 00:34:15 2017  clohfink:Joined the channel
Tue Aug 8 00:55:28 2017  kvaster:Joined the channel
Tue Aug 8 02:14:35 2017  jmckenzie:Joined the channel
Tue Aug 8 02:22:25 2017  clohfink:Joined the channel
Tue Aug 8 02:31:02 2017  kohlisankalp:Joined the channel
Tue Aug 8 03:28:30 2017  clohfink:Joined the channel
Tue Aug 8 04:06:48 2017  jeffj:Joined the channel
Tue Aug 8 04:11:56 2017  clohfink:Joined the channel
Tue Aug 8 04:22:14 2017  neil_:Joined the channel
Tue Aug 8 04:48:44 2017  damons:Joined the channel
Tue Aug 8 05:46:02 2017  jeffj:Joined the channel
Tue Aug 8 06:00:39 2017  clohfink:Joined the channel
Tue Aug 8 06:06:47 2017  ostefano:Joined the channel
Tue Aug 8 06:08:52 2017  jeffj:Joined the channel
Tue Aug 8 06:52:59 2017  spodkowinski:Joined the channel
Tue Aug 8 07:42:38 2017  cosql:Joined the channel
Tue Aug 8 07:43:41 2017  cosql:Joined the channel
Tue Aug 8 08:41:17 2017  jeffj:Joined the channel
Tue Aug 8 08:42:53 2017  clohfink:Joined the channel
Tue Aug 8 10:18:19 2017  marcuse:Joined the channel
Tue Aug 8 10:18:39 2017  zanson:Joined the channel
Tue Aug 8 11:25:06 2017  clohfink:Joined the channel
Tue Aug 8 13:08:43 2017  adamholmberg:Joined the channel
Tue Aug 8 13:16:19 2017  clohfink:Joined the channel
Tue Aug 8 13:16:39 2017  clohfink:Joined the channel
Tue Aug 8 13:32:32 2017  kohlisankalp:Joined the channel
Tue Aug 8 14:35:12 2017  kohlisankalp:Joined the channel
Tue Aug 8 15:08:06 2017  clohfink:Joined the channel
Tue Aug 8 15:42:08 2017  jeffj:Joined the channel
Tue Aug 8 15:53:22 2017  zaller:Joined the channel
Tue Aug 8 16:12:33 2017  damons:Joined the channel
Tue Aug 8 16:37:02 2017  zaller:Joined the channel
Tue Aug 8 16:40:53 2017  jeffj:Joined the channel
Tue Aug 8 16:44:50 2017  jayzhuang:Joined the channel
Tue Aug 8 16:50:40 2017  jayzhuang:Joined the channel
Tue Aug 8 16:59:49 2017  cosql:Joined the channel
Tue Aug 8 17:10:36 2017  adamholmberg:Joined the channel
Tue Aug 8 17:13:25 2017  adamholmberg:Joined the channel
Tue Aug 8 17:16:37 2017  adamholmberg:Joined the channel
Tue Aug 8 17:42:47 2017  bbromhead:Joined the channel
Tue Aug 8 17:54:26 2017  bbromhead:just bouncing an idea around, for clusters that have vnodes enabled bootstrapping multiple nodes at once is impossible if you want to maintain consistent range movements
Tue Aug 8 17:55:08 2017  bbromhead:Reading https://issues.apache.org/jira/browse/CASSANDRA-11005 you can do consistent range movements for multiple ranges (multiple nodes) if they are far enough apart in the ring
Tue Aug 8 17:55:40 2017  bbromhead:e.g. the moving ranges are not replicas for other changing ranges
Tue Aug 8 17:56:49 2017  bbromhead:so theoretically you should be able to do multiple range movements with vnodes, as long as you chose only tokens that did not have over lapping range movements
Tue Aug 8 17:57:57 2017  bbromhead:say by working out the tokens and range movements before hand and then assigning them to the new bootstrapping nodes via initial_token (this would require to removing the vnode safety check from CASSANDRA-11005)
Tue Aug 8 17:57:57 2017  CassBotJr:https://issues.apache.org/jira/browse/CASSANDRA-11005 (Resolved; Fixed; 2.1.13, 2.2.5, 3.3): "Split consistent range movement flag"
Tue Aug 8 17:58:23 2017  bbromhead:am I missing something here that would make this approach unsafe?
Tue Aug 8 18:01:13 2017  adamholmberg:Joined the channel
Tue Aug 8 18:24:06 2017  zaller:Joined the channel
Tue Aug 8 18:25:33 2017  bbromhead:Joined the channel
Tue Aug 8 19:01:10 2017  bbromhead:Joined the channel
Tue Aug 8 19:02:30 2017  jayzhuang1:Joined the channel
Tue Aug 8 19:02:32 2017  bbromhead:Joined the channel
Tue Aug 8 19:22:08 2017  jeffj:bbromhead: probably works, but with 256 tokens/node, it becomes pretty unlikely pretty quickly
Tue Aug 8 19:22:30 2017  jeffj:would need to be a very large cluster, or very few tokens/node
Tue Aug 8 19:22:56 2017  bbromhead:aimed at large clusters, as generally they are the ones that need to add multiple nodes at a time to keep up
Tue Aug 8 19:23:40 2017  bbromhead:we are starting to come across a few 200+ vnode clusters in the wild that are still using 256 tokens/node
Tue Aug 8 19:23:49 2017  bbromhead:*200+ node
Tue Aug 8 19:25:52 2017  bbromhead:with a 16 node test cluster, using random token allocations I am able to get 3 to 4 nodes worth of tokens on average
Tue Aug 8 19:27:23 2017  bbromhead:obviously at that size the end goal would be to get to a cluster that uses a smaller amount of tokens / no vnodes, but sometimes reality gets in the way
Tue Aug 8 20:17:27 2017  RussSpitzer:Joined the channel
Tue Aug 8 20:20:58 2017  adamholmberg:Joined the channel
Tue Aug 8 21:05:26 2017  cosql:Joined the channel
Tue Aug 8 21:09:52 2017  jayzhuang:Joined the channel
Tue Aug 8 21:42:12 2017  RussSpitzer_:Joined the channel
Tue Aug 8 21:51:44 2017  jayzhuang:Joined the channel
Tue Aug 8 22:06:39 2017  adamholmberg:Joined the channel
Tue Aug 8 22:07:33 2017  jayzhuang1:Joined the channel
Tue Aug 8 22:17:02 2017  jayzhuang:Joined the channel
Tue Aug 8 23:03:46 2017  jayzhuang:Joined the channel
Tue Aug 8 23:07:38 2017  jayzhuang:Joined the channel

Comments