MAX_DURATION option does not count the Pod Status Start time about pod-reaper

[rogerioefonseca]

haha, I tried to debug but I'm not familiar with go so, now, I'm not able to investigate and help further more.

[brianberzins]

Interesting case! Given the configuration, I would have expected it to be removed at that 11 minute mark. So in this case the human readable "5 minutes after it gets evicted" doesn't translate well. I think that might require a different rule from what I originally had in mind with MAX_DURATION (which going from memory was designed to be "how long has this pod existed").

It might mean I need to implement a rule for MAX_POD_STATUS_DURATION.
I'm honestly not sure off the top of me head if that information is readily available in the pod, but if it is, the rule would be pretty easy to add.

On the side here, I'm working to rebuild the CI/CD system for this project based on some changes that Docker made that basically broke the CD system for me. Trying to get all that worked out ASAP! Then this'll be high on my "fix or implement new!"

[brianberzins]

Looking into this, I'm actually a bit surprised to find that there isn't (at least an obvious) timestamp about when pods shift phases at the top level. Right now, the check for eviction uses pod.Status.Reason and the duration check is looking at pod.Status.StartTime (both off the PodStatus object). When looking around for any sort of transition timestamp, I the only other timestamps I found were in an array (not 1:1 mapping) of PodCondition structures. That would make it pretty difficult to specify a rule around because it would probably have to be really specifically "when the pod status is X and there is a PodConditionStatus of Y and that transition happened over Z time ago" feels like it would be difficult to generalize in a consumable way.

Really surprised I didn't see something like "lastPodStatusChangedTime" or something like that...