Feedback request for Message Flusher bug found in 3.0.17 (CASSANDRA-14855)
We recently found this issue on one of our 3.0.17 clusters, where the
Message Flusher falls off the event loop, eventually resulting in OOM on a
bunch of nodes. We saw this happen twice so far.
CASSANDRA-14855 <https://issues.apache.org/jira/browse/CASSANDRA-14855> has
all the details (including heap dump analysis). Backported ImmediateFlusher
from trunk as a fix for this. Would like to get feedback (on the JIRA) if
this fix is recommended, or if there is a suggestion for a better fix.