>From what I've read, gathered and pieced together (this never seems to be made 
>explicit):
There are 2 types of crashes and recoveries:

1)
Connection loss (but both client and broker running as normal -- links states 
still in memory)
In this case the broker and client still maintain their links states. (In 
memory delivery state of each inflight message). Upon network reconnection: 
these links are restored and both broker and client can recover/resume message 
send/receive where it was left off.

2)
Client or broker crashes or both
In this case the crashed party will need to resend all un-ACKed messages from 
scratch (which may result in duplicates). Clients (consumers / producers) need 
to be idempotent (intelligent enough to detect and deal with duplicates).

Is this correct?

Reply via email to