1

A client of mine asked a question that I'm not really sure how to answer correctly. His question is: In case of one node falls out of the cluster, stays offline for a few hours or days and then comes back: how fast the new data will be migrated to the node and will this affect the rest of the nodes performance?

TL;DR: They are asking what is the 3rd node comes back to the cluster and the app is asking the 3rd node for information, but the 3rd node cannot answer since it is still just rejoining the cluster and does not have the relevant data. Will the other 2 nodes provide the answer for sure 100% or is there a chance for the application to respond with a 'shut up I'm busy' answer and breaking the web app?

Now I have no clue about the answer and I've found no meaningful answer either. Can any of you tell me maybe this? Let's assume all the nodes are on the same VLAN in the same server room, so network latency is not an issue.

Bert
  • 1,028
  • 1
  • 16
  • 33
  • Possibly a question more suitable for https://dba.stackexchange.com/ but IIRC the simplified answer is: clients connect to the cluster and not to a random (and failed) node and the cluster is aware of the state of all connected nodes. Queries shouldn't be directed by the cluster to nodes that are not in a synchronised state. -|- Recovery depends on how that failed node gets re-initialised - with Incremental State Transfer (IST) or State Snapshot Transfer (SST) https://galeracluster.com/library/documentation/sst.html and that depends on how long the outage was – HBruijn Aug 28 '23 at 13:13
  • Correct me if I'm wrong please. I have HAProxy to connect to by my application and HAProxy decides where the query goes to and gget's executed, right? Does HAProxy know if like Node3 is busy (joining ATM) so it does not send there any queries or it does, but because it is in a cluster, Node3 will forward the query to Node1 or 2 and tell them to respond via HAProxy? – Bert Aug 28 '23 at 13:57

0 Answers0