DCP replication does not recover after SIGKILL/restart to memcached (dupe PREPARE?) [ETA 2019/7/19]

Build: 6.5.0-3527

Scenario:

1. Two node cluster, single Couchbase bucket (replica=1)
2. Start doc_ops
3. While sync_write creates are running, kill Memcached on node-1 (kill -9)

Observations:

Memcached restarts on node-1, all doc_ops are failing without any valid response from the server. In the packet dump, able to see the response status as "Unknown (0x00a3)". No exception was thrown from SDK end as well.

Expected behavior:

Once Memcached restarts, doc_ops should be handled without any errors. Even in case of some errors, SDK should report an exception to the user.

PCAP with Unknown status message:

TAF test case:

crash_test.crash_process.CrashTest.test_crash_process,nodes_init=2,replicas=1,num_items=10000,process=memcached,service=memcached,sig_type=sigkill,target_node=replica,durability=MAJORITY

Also attaching test.log and unknown_response_ops_failed.pcapng