Without producer-flow-control, the memory limits have no meaning. That's how
they take effect - by blocking producers.
So, without producer flow control, the symptoms described can happen
normally as there is nothing to prevent the JVM from running out of memory.
When a broker drops out-of-servi
If the broker is getting into GC purgatory, as you describe, I wouldn't
bother trying to figure out why its operation is sub-optimal - that's normal
for any java application in GC purgatory.
Look for a memory leak or other errant operation that would explain the
constant GC activity.
The results
It is possible for this to happen normally. For example, when using the
failover transport and temporarily losing connection to the broker, the
client might not realize that the consumer removal was already processed on
the broker due to the lost connection, and therefore may resend it on
reconnec