You mention that "1.16.3 deploys wordpress and mysql and verifies that they are up".
However, looking at the log I see:
369:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju.worker.upgrader upgrader.go:134 desired tool version: 1.17.0
370:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju.worker.machiner machiner.go:52 "machine-1" started
371:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju runner.go:262 worker: start "lxc-watcher"
372:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju runner.go:262 worker: start "deployer"
373:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju.worker.deployer deployer.go:106 checking unit "wordpress/0"
374:juju-test-release-hp-machine-1:2013-11-26 15:09:23 INFO juju.worker.deployer deployer.go:146 deploying unit "wordpress/0"
That would indicate that the agent already thinks it wants to run Juju 1.17.0 *before* it has gotten to the point of deploying anything. I don't actually see any steps where the tool is clearly 1.16.3 (we should add a log message on startup of what version the agent is running, though we do have that in the RPC log on machine-0):
You mention that "1.16.3 deploys wordpress and mysql and verifies that they are up". test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju.worker. upgrader upgrader.go:134 desired tool version: 1.17.0 test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju.worker. machiner machiner.go:52 "machine-1" started test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju runner.go:262 worker: start "lxc-watcher" test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju runner.go:262 worker: start "deployer" test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju.worker. deployer deployer.go:106 checking unit "wordpress/0" test-release- hp-machine- 1:2013- 11-26 15:09:23 INFO juju.worker. deployer deployer.go:146 deploying unit "wordpress/0"
However, looking at the log I see:
369:juju-
370:juju-
371:juju-
372:juju-
373:juju-
374:juju-
That would indicate that the agent already thinks it wants to run Juju 1.17.0 *before* it has gotten to the point of deploying anything. I don't actually see any steps where the tool is clearly 1.16.3 (we should add a log message on startup of what version the agent is running, though we do have that in the RPC log on machine-0):
388:juju- test-release- hp-machine- 0:2013- 11-26 15:09:23 DEBUG juju.state. apiserver apiserver.go:102 <- [4] machine-1 {"RequestId" :7,"Type" :"Upgrader" ,"Request" :"SetTools" ,"Params" :{"AgentTools" :[{"Tag" :"machine- 1","Tools" :{"Version" :"1.17. 0-precise- amd64"} }]}}
Where machine-1 is reporting to the API server that it is running 1.17.0.
And I'll note that it isn't until:
1107:juju- test-release- hp-machine- 2:2013- 11-26 15:09:38 INFO juju charm.go:61 worker/ uniter/ charm: downloading cs:precise/mysql-29 from https:/ /region- a.geo-1. objects. hpcloudsvc. com/v1/ 17031369947864/ juju-pavlova- snags-- chiko-roll- meat-pie- pasty-sauce- 2013-10- 10/cs_3a_ precise_ 2f_mysql- 29
a full 800 messages and 16 seconds later that machine-2 actually starts downloading the Mysql charm.
so this doesn't seem to be "upgrading mysql fails" but "installing mysql using 1.17.0" fails.
Can you confirm?