UP node check needs to be made more resilient #360
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This fixes an issue with subcluster creation. We weren't running the rebalance_shards() to add subscriptions to the new subcluster. To hit this you had to have a few specific things: - had to use a v11 server - the database has to be migrated from enterprise to eon There was a server issue during migration, to be fixed separately, that set some wrong state for certain tables. Some tables were identified as being "shared", which to the server means we need active shard subscriptions to query them. If no subscriptions, then the query would fail with "ERROR 9099: Cannot find participating nodes to run the query". This was affecting queries the operator does to catalog tables -- key for this fix was it affected any query to the nodes table. Because of this, the operator deemed the new scaled out nodes as down. So, it would never run the rebalance. The fix is to align the UP node check with the livenessProbe by looking to see if the vertica process is running. We still query out the node state from node, but it is used for information purposes now.
roypaulin
approved these changes
Mar 24, 2023
| @@ -209,7 +209,7 @@ func (r *VerticaDBReconciler) constructActors(log logr.Logger, vdb *vapi.Vertica | |||
| // status updates after both of them. | |||
| MakeStatusReconciler(r.Client, r.Scheme, log, vdb, pfacts), | |||
| // Update the labels in pods so that Services route to nodes to them. | |||
| MakeClientRoutingLabelReconciler(r, vdb, pfacts, AddNodeApplyMethod, ""), | |||
| MakeClientRoutingLabelReconciler(r, vdb, pfacts, PodRescheduleApplyMethod, ""), | |||
There was a problem hiding this comment.
Just curious what are these *ApplyMethod?
There was a problem hiding this comment.
Good question, when I was testing this out I ran into an issue here. When we add a label for client routing and use AddNodeApplyMethod we don't do anything if the node we need to update doesn't have any subscriptions. But the reconciler for the rebalance happens after this. So, it was a bit of a chicken/egg problem. Using PodScheduleApplyMethod got me around that.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This fixes an issue with subcluster creation. We weren't running the rebalance_shards() to add subscriptions to the new subcluster. To hit this you had to have a few specific things:
There was a server issue during migration, to be fixed separately, that set some wrong state for certain tables. Some tables were identified as being "shared", which to the server means we need active shard subscriptions to query them. If no subscriptions, then the query would fail with "ERROR 9099: Cannot find participating nodes to run the query".
This was affecting queries the operator does to catalog tables -- key for this fix was it affected any query to the nodes table. Because of this, the operator deemed the new scaled out nodes as down. So, it would never run the rebalance.
The fix is to align the UP node check with the livenessProbe by looking to see if the vertica process is running. We still query out the node state from node, but it is used for information purposes now.
Closes #355