Feature/rolling update #66
Conversation
|
Would it add too much complexity to allow maxSurge and maxUnavailable on the same config like deployments do? |
I'd rather not change any behaviour and/or scheduler config at this moment @andrehp. Do you think this is a must have? |
There was a problem hiding this comment.
I agree @cscatolini. Maestro already has the query string maxsurge which suffices many cases
controller/utils.go
Outdated
| db, clientset, configYAML, scheduler) | ||
| if err != nil { | ||
| if err == nil || !strings.Contains(err.Error(), "not found") { | ||
| logger.WithField("pod", pod.GetName()).Debugf("pod still exists, deleting again") |
There was a problem hiding this comment.
log the error here if not nil
I wouldn't say it's a must have, but it might be nice to speed up deploys, though I do agree it might not be the best idea to change this behavior right now.
I did a quick search on the docs about maxsurge but didn't find anything. But I think maestro's maxsurge is different from a k8s deployment's, isn't it? |
controller/utils.go
Outdated
| @@ -276,51 +271,83 @@ func waitTerminatingPods( | |||
| exit := true | |||
| select { | |||
| case <-ticker.C: | |||
| for _, pod := range deletedPods { | |||
| if operationManager.WasCanceled() { | |||
There was a problem hiding this comment.
shouldn't you have the same check here?
if operationManager != nil && operationManager.WasCanceled() {
controller/utils.go
Outdated
| } | ||
|
|
||
| if len(createdPod.Status.Phase) == 0 { | ||
| //HACK! Trying to detect if we are running unit tests |
There was a problem hiding this comment.
can't this have any unintended side effects? Maybe add a logger here?
You can talk to @lftakakura. We don't want major changes but we believe that creating 25% new pods before deleting previous versions may be a big burden in our infrastructure. He is thinking about some way to speed the process and will update this PR soon.
Yes. It is a global maestro's config instead of per scheduler. The default we're using in prod is 25%. |
* Create rooms before deleting old ones * Remove unnecessary comments * Fix rolling-update flow * Create pod and delete old one after in each chunk * Remove waitCreatingAndDeleteOldPods * Log error if not able to get pod in waitTerminatingPods and waitCreatingPods * WIP: Use goroutines to perform rolling update * Use goroutines to perform rolling update
* Create rooms before deleting old ones * Remove unnecessary comments * Fix rolling-update flow * Create pod and delete old one after in each chunk * Remove waitCreatingAndDeleteOldPods * Log error if not able to get pod in waitTerminatingPods and waitCreatingPods * WIP: Use goroutines to perform rolling update * Use goroutines to perform rolling update
* Create rooms before deleting old ones * Remove unnecessary comments * Fix rolling-update flow * Create pod and delete old one after in each chunk * Remove waitCreatingAndDeleteOldPods * Log error if not able to get pod in waitTerminatingPods and waitCreatingPods * WIP: Use goroutines to perform rolling update * Use goroutines to perform rolling update
Deploy pods before deleting them in a rolling-update respecting maxSurge