You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
And all these LB policies throw logs for events like creation, new state from above, creating a child policy, new state from the child policy etc. But the format of these logs are different in subtle ways and it makes the whole process of debugging painful.
It would be nice to standardize the format and the events which every LB policy should be logging (at minimum). We are going to be running into issues where we are going to have to look at logs with 10+ LB policies and doing this asap is going to help us a great deal.
The text was updated successfully, but these errors were encountered:
From what I understand about this issue, xds/balancer have several logs like
b.logger.Infof("Created child policy %p of type %s", b.childLB, clusterresolver.Name)
b.logger.Infof("Received update from resolver, balancer config: %+v", pretty.ToJSON(s.BalancerConfig))
b.logger.Infof("Created")/Updated/Shutdown
They do have prefixes to differentiate between each other.
How can more structure to these logs can be added ? If you can direct me , ill be able to help implementing it
@easwars is out on parental leave atm. There is a possibility that the logging could have been improved since this issue was posted. I will try to see if we can get some more info (if possible) on the original intent of this issue
meanwhile, please peruse through the other set of issues marked "Hacktoberfest" or "Status:Help wanted" if you'd like to get started while waiting on this issue.
When I was debugging one of the interop test failures where the LB policy tree on the channel looked something like this:
cluster_manager
-->rls
-->cds
-->cluster_resolver
-->priority
-->cluster_impl
-->ring_hash
The above is only part of the tree.
And all these LB policies throw logs for events like creation, new state from above, creating a child policy, new state from the child policy etc. But the format of these logs are different in subtle ways and it makes the whole process of debugging painful.
It would be nice to standardize the format and the events which every LB policy should be logging (at minimum). We are going to be running into issues where we are going to have to look at logs with 10+ LB policies and doing this asap is going to help us a great deal.
The text was updated successfully, but these errors were encountered: