Add calls to /namespace/ready if supported by coordinator #245
Conversation
| // NB(nate): Marking namespaces ready (without force) involves checking dbnodes for namespace | ||
| // availability. External coordinators do not have connections to dbnodes so don't attempt this | ||
| // step for now. | ||
| if cluster.Spec.ExternalCoordinator != nil { |
There was a problem hiding this comment.
FYI, support for this is something we'll be focusing on after 1.0
| } | ||
| err := c.adminClient.namespaceClientForCluster(cluster).Ready(req) | ||
| if pkgerrors.Cause(err) == m3admin.ErrNotFound { | ||
| c.logger.Info("coordinator does not yet support the ready endpoint. " + |
There was a problem hiding this comment.
This is probably valuable enough context to keep at Info, but just a heads up this will log every time a cluster is processed.
Codecov Report
@@ Coverage Diff @@
## master #245 +/- ##
=======================================
Coverage 76.29% 76.29%
=======================================
Files 30 30
Lines 2168 2215 +47
=======================================
+ Hits 1654 1690 +36
- Misses 380 388 +8
- Partials 134 137 +3 Continue to review full report at Codecov.
|
| // NB(nate): Due to bug in coordinator API routing logic, missing routes may | ||
| // also be returned as 405s (i.e. method not allowed). So check for that in addition | ||
| // to 404s. | ||
| if cause == m3admin.ErrNotFound || cause == m3admin.ErrMethodNotAllowed { |
There was a problem hiding this comment.
This was a fun little discovery during testing. Depending on the route, misses will either return a 404 or a 405. Almost certainly a bug in our routing logic, but since these coordinators will be out there in the wild, we need to account for it.
(cc @schallert since this is a small update after the LGTM)
Add support for transitioning namespaces to a ready state after they've been created. This operates as a signal to coordinators pulling namespace information from etcd that the namespace is ready for use.