Details
-
Bug
-
Resolution: Unresolved
-
Major
-
7.6.0
-
Untriaged
-
0
-
Unknown
Description
When the CC is failed over, the behavior varies depending on the state of the NC:
- for NCs which were connected to the CC at the time of failover, they will start reporting 503s for requests.
- While the CC is failed over, the NCs will constantly attempt to reestablish a connection to the CC in vain- or worse if the failed-over CC is network partitioned from the ns_server orchestrator, but otherwise reachable by the NC. This will cause the NC to send requests to a CC which should be failed over, leading to inconsistent results.
- for NCs whose java or cbas process restarts for any reason, they do not start the driver, resulting in connection refused for any requests
It seems to be the correct behavior should be:
- 503 for requests when CC is failed over
- this should be the case irrespective of the when the NC process started
- NCs should not keep spamming failed-over CCs in vain- they should wait for a topology update indicating a restored or replacement CC.