This repository has been archived by the owner. It is now read-only.
Please sign in to comment.
client: Don't query unconnected machines for leader information
Before this commit, we queried all machines for leader information, regardless of whether they had connected. If a machine is not connected, it probably won't respond, so querying it only increases the amount of time required for the function to return. This resulted in `kelda ps` being extremely slow when a machine had not yet connected (which can happen when the cluster is first booting, or if a machine crashes). Sometimes, the error "context deadline exceeded" would be printed because the Query call to the daemon would timeout since obtaining the leader client took so long.
- Loading branch information...
Showing with 36 additions and 1 deletion.