Use debounce for pod scanning #153
Conversation
Instead of syncing the pod list for a service at most once every 10 minutes, apply a debounce on changes so that we don't sync the pods list unless the service has been stable for 5 seconds or it has been more than 5 minutes since the last sync. This should help avoid cases where pods are restarted and kubefwd doesn't seem to allow connecting to them any more, because it will sync pods more often than before.
dobesv
force-pushed
the
debounce-again
branch
from
ad0e5ed to
26cb0dc
Compare
|
@dobesv thanks I'll review this soon |
|
@dobesv I have been testing this and noticing that my forwards seem to resync all services on occation when no pods have gone offline, this disrupts my work. We may need a better way of detecting a failure (or I need a better way of testing this). |
|
Hmm I wonder if I accidentally reverted some of your changes when I merged them into this PR. |
|
@dobesv I am looking into a problem with IP assignements on headless services. Once I clear that up I think this PR should be good. |
|
OK, let me know if you need my help with anything. |
|
@dobesv the whole service registry system has been problematic. Now when SyncPodForwards runs it tries to re-bind ports on the running services, which it can not and then removes the service. I don't think it has to do with your debounce but I don't see how to make this registry work properly. I might disable the sync since the new version basicaly becomes unusable to me after the re-sync interval. |
Use debounce for pod scanning
Instead of syncing the pod list for a service at most once every 10 minutes, apply a debounce on
changes so that we don't sync the pods list unless the service has been stable for 5 seconds or it has been more than 5 minutes since the last sync.
This should help avoid cases where pods are restarted and kubefwd doesn't seem to allow connecting to them any more, because it will sync pods more often than before.
Note I had previously made a PR for this exact change, which was merged, but most of the changes were lost in a later merge, maybe due to merge conflicts or somesuch.