3 cluster nodes
member-01：172.20.14.162
member-02：172.20.14.163
member-03：172.20.14.164

odl-version:Oxygen-sr4(0.8.4)

1,config cluster
2,cluster nodes start
3,install feature:odl-mdsal-all
4,add reject route on node01 as:
route add -host 172.20.14.163 reject
route add -host 172.20.14.164 reject

5,few minutes later delete reject route as :
route del -host 172.20.14.163 reject
route del -host 172.20.14.164 reject

6, log always print "is still unreachable or has not been restarted. Keeping it quarantined."
7, node did not restart

cluster config use default settings,such as node01:
such as node01:
odl-cluster-data {
akka {
remote {
artery

netty.tcp

1. when under load we might trip a false positive on the failure detector
2. transport-failure-detector { # heartbeat-interval = 4 s # acceptable-heartbeat-pause = 16s # }

   }

cluster

persistence {

1. By default the snapshots/journal directories live in KARAF_HOME. You can choose to put it somewhere else by
2. modifying the following two properties. The directory location specified may be a relative or absolute path.
3. The relative path is always relative to KARAF_HOME.

1. snapshot-store.local.dir = "target/snapshots"
2. journal.leveldb.dir = "target/journal"

journal {
leveldb

}
}
}
}

when test on n-sr2, node quarantine and automatic restart is possible.
I read the log notice that on n-sr2 log will print "Got quarantined by akka.tcp://opendaylight-cluster-data@xx.xx.xx.xx:2550" before auto-restart.
so i check the o-sr4 code and add log found that node did not received any message and the restart method will not be called.

method "onReceive":
QuarantinedMonitorActor

Now I'm trying to solve this problem, hope I can get some help about this problem, thanks.

I have tested Fluorine-SR2， still has this problem.
Here are the problems I found：
odl-controller subscribe the “ThisActorSystemQuarantinedEvent” and call karaf.restart method.
Oxygen-sr4 dependency new akka-version and akka do not publish “ThisActorSystemQuarantinedEvent” during the operation process. so I change subscribe event from “ThisActorSystemQuarantinedEvent” to RemotingLifecycleEvent(super class) and found akka publish "AssociationErrorEvent" when restore network with the detail log:

_onReceive AssociationErrorEvent AssociationError [akka.tcp://opendaylight-cluster-data@172.20.14.162:2550] -> [akka.tcp://opendaylight-cluster-data@172.20.14.163:2550]: Error [Invalid address: akka.tcp://opendaylight-cluster-data@172.20.14.163:2550] [
akka.remote.InvalidAssociation: Invalid address: akka.tcp://opendaylight-cluster-data@172.20.14.163:2550
Caused by: akka.remote.transport.Transport$InvalidAssociationException: The remote system has a UID that has been quarantined. Association aborted._

I read akka code about publish event, publish business has changed, not much has been discovered.
I have already made a repair on my odl-project based on the "AssociationErrorEvent", and I will write out the details of the modification recently. But I do not intend to submit it to the community, because I feel this solution is not good, just for your reference.

Here is my solution(based on stable/oxygen):
when receive "AssociationErrorEvent" and message contains "The remote system has a UID that has been quarantined", ready to call restart method.
Before restart , it will count the number of remote addresses to restart the Isolated single node. Previous designs may restart two other nodes, and my approach guarantees business uninterrupted.
This solution works only for three-nodes clusters, more nodes may have problems. I've tested it several times in a three-nodes environment and it's stable.

I have submitted my codes to my github：
change-diff

just for your reference. Looking forward to better solutions.

CONTROLLER-1904 have already been solved

https://jira.opendaylight.org/browse/CONTROLLER-1941