The first request after a server node becomes unavailable sometimes fails with RequestCanceledException

The first UpsertAsync after a node becomes unavailable sometimes fails immediately with RequestCanceledException

• Steps to reproduce:
• Set up 2 node cluster.
• call UpsertAsync with a long timeout (should succeed immediately)
• make one of the CB nodes unavailable (I am doing this by stopping one of the CB containers)
• all UpsertAsync with a long timeout (should succeed immediately or later depending on which CB node it writes to)
• make the CB node available again (I am starting the container )
• the above UpsertAsync  should soon succeed if it was waiting
• repeat

Most of the time UpsertAsync will work as expected, but sometimes it will fail immediately with RequestCanceledException

Call stack:

 at Couchbase.Core.ClusterNode.ExecuteOp(Func`4 sender, IOperation op, Object state, CancellationTokenPair tokenPair)
at Couchbase.CouchbaseBucket.SendAsync(IOperation op, CancellationTokenPair tokenPair)
at Couchbase.Core.Retry.RetryOrchestrator.RetryAsync(BucketBase bucket, IOperation operation, CancellationTokenPair tokenPair)
at Couchbase.KeyValue.CouchbaseCollection.UpsertAsync[T](String id, T content, UpsertOptions options)
at NewClientLibTest.Program.Main(String[] args) in C:\source\CBTests\NewClientLibTest\Program.cs:line 133