| aisrael | I've done something slightly dumb; moved a juju controller/model, on lxd, to a new bridge. The containers have the right ip now, but the juju controller isn't accessible. It looks like the api websocket isn't listening on the new ip. | 01:36 |
|---|---|---|
| === frankban|afk is now known as frankban | ||
| kjackal | Good morning Juju world! | 07:43 |
| admcleod | kjackal: hello! do you have a good example for peer relations? | 08:00 |
| kjackal | hello admcleod, I guess the best would be ZK or spark | 08:00 |
| admcleod | spark please | 08:01 |
| kjackal | admcleod: Give me a sec to spot the respective | 08:01 |
| kjackal | code | 08:01 |
| admcleod | interface-spark-quorum? | 08:01 |
| kjackal | admcleod: Here is the interface: https://github.com/juju-solutions/interface-spark-quorum/blob/master/peers.py | 08:05 |
| kjackal | admcleod: Here is the interface: https://github.com/juju-solutions/interface-spark-quorum/blob/master/peers.py | 08:05 |
| admcleod | thanks | 08:05 |
| kjackal | admcleod: updating peers ends up here: https://github.com/juju-solutions/layer-apache-spark/blob/master/lib/charms/layer/apache_spark.py | 08:06 |
| kjackal | admcleod: after going though the reactive part: https://github.com/juju-solutions/layer-apache-spark/blob/master/reactive/spark.py#L143 | 08:07 |
| admcleod | kjackal: i wonder, why use unit scope and not global for peering? | 08:07 |
| kjackal | admcleod: https://github.com/juju-solutions/layer-apache-spark/blob/master/lib/charms/layer/apache_spark.py#L294 | 08:07 |
| kjackal | admcleod: that is a good question! | 08:07 |
| kjackal | admcleod: not sure I have a good answer | 08:08 |
| admcleod | or service | 08:09 |
| admcleod | kjackal: looks like UNIT is what it should be. im trying to do something really simple, just setting a state (conv.set_remote('thing', True) and then trying to read it back (a = conv.get_remote('thing') | 08:33 |
| admcleod | kjackal: but it keeps returning None and im sure its something simple. got a good example of this? | 08:34 |
| kjackal | admcleod: you want all nodes to agree on a remote 'thing'? | 08:35 |
| admcleod | kjackal: yep | 08:35 |
| kjackal | admcleod: have you consedered using the leadership layer? | 08:36 |
| admcleod | yes, but this should work no? | 08:36 |
| kjackal | yes, peers should work as well but I feel you will be coding the same logic that you can find in leadership | 08:37 |
| admcleod | thats fine | 08:37 |
| kjackal | admcleod: let me look at the zk layer, It might be easier | 08:37 |
| admcleod | kjackal: i want the leader to be able to set 'is_leader, True', which all other peers should be able to see - that bit i could do with leadership. but i also want all other peers to set a state, "non-leader ready" | 08:39 |
| kjackal | admcleod: looking at the zk quorum there is not magic: https://github.com/juju-solutions/interface-zookeeper-quorum/blob/master/peers.py#L55 | 08:40 |
| kjackal | admcleod: give me a sec for the non-leader state, I have seen an example | 08:41 |
| kjackal | admcleod: here is how kubernetes-master figures out if it is the leader: https://github.com/kubernetes/kubernetes/blob/master/cluster/juju/layers/kubernetes-master/reactive/kubernetes_master.py#L145 | 08:42 |
| kjackal | admcleod: and here is what happens if it is not: https://github.com/kubernetes/kubernetes/blob/master/cluster/juju/layers/kubernetes-master/reactive/kubernetes_master.py#L191 | 08:43 |
| kjackal | I think you should have something like "when_not('leadership.is_leader') | 08:44 |
| kjackal | def nonleader(): | 08:44 |
| kjackal | set_state('non-leader') | 08:44 |
| kjackal | status_set('active', 'non-leader ready')" | 08:44 |
| kjackal | admcleod: ^ | 08:44 |
| admcleod | kjackal: right.. yeah... but | 08:45 |
| admcleod | kjackal: when_not leader, i want to set something.. like "ok im ready". | 08:45 |
| admcleod | kjackal: then once all the other nodes have set ready, the leader can do action | 08:46 |
| kjackal | admcleod: I see... this is tricky... how would you know that all your non-leaders are up? What if you have some non-leader pending provision? | 08:48 |
| admcleod | kjackal: it gets a list of all other nodes (it knows its the leader). it checks if those other nodes had set a state (ready). if it hasnt, it waits | 08:48 |
| === firl_ is now known as firl | ||
| === stormmore is now known as Budgie^Smore | ||
| admcleod | kjackal: anyway. i guess the thing here is... for me, set_remote and get_remote dont appear to be working as id expect | 08:57 |
| kjackal | admcleod: The thing that needed some care when we were doing these interactions was that we would have to use conv.set_state('my_state') to signal the other side that something has changed https://github.com/juju-solutions/interface-zookeeper-quorum/blob/master/peers.py#L31 | 09:00 |
| admcleod | ah. | 09:01 |
| kjackal | admcleod: after that the other end would have to go and do a get_remote('whatever') | 09:01 |
| kjackal | admcleod: basicaly you should model the message enchange "protocol" using conversation states | 09:02 |
| admcleod | kjackal: told you it was somethign simple. thanks | 09:02 |
| kjackal | admcleod: :) | 09:02 |
| admcleod | kjackal: also, how do you force an update-status now? | 09:58 |
| admcleod | kjackal: nvm | 10:04 |
| joedborg | hey all! is the Joyent cloud still supported? | 11:02 |
| joedborg | hey all! is the Joyent cloud still officially supported? | 11:03 |
| Diabelko | is there anything special I should know about using Juju with MAAS? | 11:21 |
| Diabelko | used to work just fine up until I removed the model and tried to redeploy the environment | 11:22 |
| Diabelko | now it doesn't boot the servers and I don't see anything in the logs | 11:22 |
| joedborg | Diabelko: can you still play with the nodes via MaaS? | 11:48 |
| kjackal | joedborg: Joyent should be still supported | 12:37 |
| joedborg | kjackal: seems that apt fails now | 13:15 |
| kjackal | joedborg: how does it fail? At which point? | 13:25 |
| === lazyPower is now known as lp|swap | ||
| === freyes__ is now known as freyes | ||
| admcleod | kjackal: turns out its a problem with the function im calling after i set the state (with set_remote) - it seems as if, since its in 'executing' state, the state i sent hasnt been committed.. and there doesnt seem to be a flush that applies to set_remote | 14:57 |
| cory_fu | tvansteenburgh, petevg, stokachu (and anyone else who feels like reviewing): https://github.com/juju/python-libjuju/pull/102 when you have a chance | 16:03 |
| stormmore | lp|swap, what you swaping today? | 17:15 |
| === frankban is now known as frankban|afk | ||
| mbruzek | stormmore: Yes he is. Is there something I can help you wiht? | 17:45 |
| stormmore | mbruzek, not at the moment, thanks though :) just wanted to let him know that I think I figured out the last "bug" I came across with regards to seeing old replicasets for heapster | 17:58 |
| mbruzek | Was there any code change needed? (hint: I am working on our next release) | 17:59 |
| stormmore | mbruzek, no I think based on what I am seeing elsewhere, it is working as designed. | 18:00 |
| stormmore | mbarnett, the "problem" (hence the confusion) is that heapster ends up with 3 replicasets but only 1 which has running containers | 18:00 |
| stormmore | mbruzek, no code needed I think based on what I am seeing elsewhere, it is working as designed. | 18:01 |
| mbruzek | stormmore: OK glad to read that. If you need anything in the next release please let me know | 18:01 |
| stormmore | mbruzek, in testing deployments of my own services, it would appear that this is the design so that it makes it easier to roll back | 18:02 |
| mcapsali1 | hi guys | 18:03 |
| mcapsali1 | i have a problem with python-jujuclient and juju ha and maybe you can help me | 18:03 |
| derekcat | jamespage: Hey, thanks for the reply! Could you resend that message? I didn't have chat logging turned on until now >_< [I got a notification that said, "Actually that's a charm bug that's been...", but I didn't have enough scrollback] | 18:04 |
| mcapsali1 | we run juju 2.0.1 with 3 state machines in ha | 18:04 |
| mcapsali1 | we have a custom python code that uses python-jujuclient to deploy a bundle and then use juju.watcher to monitor the deployment untill it is finished | 18:05 |
| mcapsali1 | the problem is that sometimes after a bundle deployment, jujuclient tries to reconnect to a watcher | 18:06 |
| mcapsali1 | https://paste.ubuntu.com/24307892/ | 18:06 |
| mcapsali1 | here is a paste of the error | 18:06 |
| mcapsali1 | iḿ not that good with python but as i understand there is no ´connector´ instanciated in jujuclient code when the reconnect occurs | 18:06 |
| mcapsali1 | there is a bug opened in december 2016 with the same problem, it seems this only occurs in a juju ha env | 18:07 |
| mcapsali1 | this bug https://bugs.launchpad.net/python-jujuclient/+bug/1639104 is similar to what we expirience | 18:08 |
| mup | Bug #1639104: can no longer deploy to model: 'error': 'shared state watcher was stopped' <canonical-bootstack> <canonical-is> <landscape> <python-jujuclient:New> <https://launchpad.net/bugs/1639104> | 18:08 |
| mcapsali1 | any suggestion will be greatly appreciated :) | 18:12 |
| tvansteenburgh | mcapsali1: the best suggestion i can give is to submit a patch for the bug | 18:16 |
| tvansteenburgh | mcapsali1: or you can try using the new python client | 18:17 |
| tvansteenburgh | mcapsali1: python-jujuclient isn't gonna get much love going forward | 18:17 |
| tvansteenburgh | mcapsali1: https://github.com/juju/python-libjuju | 18:18 |
| cory_fu | mcapsali1: I should note, though, that I think that also has an issue with reconnects (issues #98 and #99) | 18:19 |
| mcapsali1 | iĺl take a look into that | 18:20 |
| mcapsali1 | did not know there was a new python client | 18:20 |
| mcapsali1 | ty everyone | 18:20 |
| mcapsali1 | i will get back if i run into same problem is the new client :) | 18:21 |
| cory_fu | mcapsali1: It seems like the connector call ought to be picking up https://bazaar.launchpad.net/~juju-deployers/python-jujuclient/trunk/view/head:/jujuclient/juju2/environment.py#L21 | 18:24 |
| tvansteenburgh | cory_fu: it's a Watcher though | 18:24 |
| tvansteenburgh | i think there are just missing connector properties on the juju1 and juju2 Watcher subclasses | 18:24 |
| cory_fu | tvansteenburgh: Hrm. I figured the Watcher would need to be based off the correct Environment for the correct version of Juju? | 18:25 |
| cory_fu | I see | 18:25 |
| cory_fu | So it might be enough to copy that connector method to the Watcher class there? | 18:25 |
| tvansteenburgh | yeah | 18:26 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!