nats-io / nats-streaming-server / 1050

94%

main: 92%

push

travis-ci

Various fixes to sent msg/ack replication

-Make the low level creation of channel not create the raft group.
 This so that in recovery, when processing the recovered channels,
 there is no race between the replay of raft log and update of
 subscriptions state from the recovered subs that we get from the
 store. The raft group is assigned when recovering subs for a
 channel is complete.
-Replication of sent message and ack is now non blocking (in that
 we don't wait for the consensus. This avoids deadlock.
-All functions invoked from channel.Apply() need to be idemptotent
 since we could be replaying raft log entries that have been
 already processed when recovering the store.
-In cluster mode, redelivery for queue group alwasy resend to same
 queue member (this is to avoid message sent/ack replication inside
 the redelivery code.
-In non cluster mode, the ack subscription for a given channel is
 now done in initSubscriptions(). We want to delay receiving data
 from clients after server set up reaches a certain point.
-Fix some clustering tests and added a test that ensure that
 if a node of the cluster is restarted, the raft log is replayed
 and subscriptions/msgs/acks are not duplicated.

148 of 148 new or added lines in 1 file covered. (100.0%)

7520 of 8044 relevant lines covered (93.49%)

211624.64 hits per line