Part of #809.
Why?
This opens the door for fixing the Oauth2 implementation which solves problems for many (including us).
Current state
Although it is termed a NiFi cluster, cluster members are not exactly replicas, and hold local state.
One example of local state is a Site-to-Site flowfile transfer to another cluster. Currently, to get the status of a transfer, you must connect to the node that you initiated the transfer on.
This might be the reason for the following:
- A JWT issuer per cluster memeber (tokens cannot be used on other cluster members)
- SNI hostname checks
Future state
All cluster state should be shared to cluster members so that cluster members can actually be replicas, and therefore sit behind a load balancer without needing sticky-sessions.
Recommended approach
- Try to find any reason why some state cannot be shared across cluster members.
- Start a conversation with NiFi maintainers to gauge interest in accepting this fix upstream.
- Fork NiFi and try to share the currently unshared state across cluster members.
Unknowns
- Which other state is local only? Perhaps the answer can be found in NIFI-7246
Part of #809.
Why?
This opens the door for fixing the Oauth2 implementation which solves problems for many (including us).
Current state
Although it is termed a NiFi cluster, cluster members are not exactly replicas, and hold local state.
One example of local state is a Site-to-Site flowfile transfer to another cluster. Currently, to get the status of a transfer, you must connect to the node that you initiated the transfer on.
This might be the reason for the following:
Future state
All cluster state should be shared to cluster members so that cluster members can actually be replicas, and therefore sit behind a load balancer without needing sticky-sessions.
Recommended approach
Unknowns