[CURATOR-125] ConnectionStateListener is confused by READ_ONLY state

To reproduce:

• have a 3 nodes ZK ensemble with readonlymode.enabled


• shut down 2 of the 3 ZK servers ( we start in read only mode)

Then create a piece of code (see Test.java attached):

• a curator client (keep the timeout reasonably short)


• a NodeCache listener on '/'


• a separate ZooKeeper client

-> the connection goes into READ_ONLY/ConnectedReadOnly as expected

• start another ZooKeeper

-> the connection goes into SUSPENDED/Disconnected, then CONNECTED/SyncConnected, fine.

• stop one of the 2 ZooKeeper alive

-> the connection goes:

ZOOKEEPER STATE: Disconnected

CURATOR STATE: SUSPENDED

ZOOKEEPER STATE: ConnectedReadOnly

CURATOR STATE: READ_ONLY

CURATOR STATE: SUSPENDED

CURATOR STATE: READ_ONLY

CURATOR STATE: SUSPENDED

CURATOR STATE: READ_ONLY

So it's flaky. Sometimes it doesn't switch back and forth, sometimes twice only, sometimes a lot more.

Depending on the timeout on the client, it might take more time to appear.

I attached a sample code that would reproduce it in 20-30 seconds.

Note that the problem may ultimately be on the ZooKeeper side but at that point I just don't know.

---
Originally reported by benjamin.jaton, imported from: ConnectionStateListener is confused by READ_ONLY state

• status: Open
  
• priority: Major
  
• resolution: Unresolved
  
• imported: 2025-01-21