UnitAssigner worker #3427
UnitAssigner worker #3427
Conversation
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Does not match ['fixes-1511771'] |
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Generating tarball failed |
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Generating tarball failed |
store the charm settings in the same transaction that adds the service This fixes part of https://bugs.launchpad.net/juju-core/+bug/1486553 **i/o timeout errors can cause non-atomic service deploys** - the part about settings not getting set atomically when the service is added, so you can have a service added without the settings set correctly. Now the setting and service add are done in the same transaction, so if there's a service, it has the right settings, period. This does not fix the second part of the problem, where the units may not get assigned, that can still happen (so if you deploy a service with non-zero units, you might get a service with zero units). (Review request: http://reviews.vapour.ws/r/2639/) Conflicts: state/state_test.go
…placement info on service creation
…e during AddService (since it's not yet). Also change AssignUnits to return a list of the results of the assignation.
refactor the idprefixwatcher to a generic collection watcher and refactor the watcher to be a strings watcher.
… the logic for converting an instance.Placement into the magic string that provicers want
Moved the code to run the unitassigner worker into a couple base suites, which fixed a ton of tests.
…emove the 'during add service' flag.
…s of unit agents if an assignment fails. Also converted the API client to take UnitTags, which makes its purpose more clear.
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Generating tarball failed |
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Tests failed |
|
|
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
|
Build failed: Tests failed |
|
this was a spurious test timeout, retrying. |
|
Status: merge request accepted. Url: http://juju-ci.vapour.ws:8080/job/github-merge-juju |
UnitAssigner worker This PR moves unit assignment during service deployment into a worker. Thus, when you deploy a service, we don't immediately create machines and assign units to them. Instead, we record what you'd like to have happen, and have the worker do that. This makes service deployment atomic, and allows for retrying of unit assignment as needed. However, it complicates tests, because now the assignment is not synchronous, so there's some changes to tests to poll for the expected units/machines while we wait for the worker to do its thing. (Review request: http://reviews.vapour.ws/r/2814/)
Revert unitassigner merge (PR #3427) Reverts PR #3427 (merge of unitassigner) This fixes https://bugs.launchpad.net/juju-core/+bug/1513552 (master cannot deploy charms) $$fixes-1513552$$ (Review request: http://reviews.vapour.ws/r/3083/)
This PR moves unit assignment during service deployment into a worker. Thus, when you deploy a service, we don't immediately create machines and assign units to them. Instead, we record what you'd like to have happen, and have the worker do that. This makes service deployment atomic, and allows for retrying of unit assignment as needed. However, it complicates tests, because now the assignment is not synchronous, so there's some changes to tests to poll for the expected units/machines while we wait for the worker to do its thing.
(Review request: http://reviews.vapour.ws/r/2814/)