Using a 2 node cluster against 6.6.0 build 7873.
When running the gocb test suite at some point during the TestQuery test the SDK connection is closed by the server and then all kv tests thereafter fail with temp fail. Looking in the server logs (attached) it looks like memcached crashes and doesn't recover. In the UI the bucket shows as "1 node pending".
To repro create a 2 node cluster with kv,index,n1ql on one node and kv,index,n1ql,fts,cbas on the other. Create a default bucket. Clone https://github.com/couchbase/gocb and run tests with
go test -race -v ./ --server <address> --user <user> --pass <password> --bucket default --version 6.5.0
During the TestQuery test the logs will output an EOF and then the SDK will be unable to recover (and tests will take a long time to complete) due to memcached being down on one node. Unfortunately I've been unable to distill this down into a smaller repro case and I can't tell what is causing the crash. I can reliably reproduce this via our SDK jenkins and locally using our cbdyncluster infrastructure. This was not happening on build 7692 and does not happen against 6.5.1.