Nil pointer dereference in bbolt 1.4.0-alpha.0 detected by robustness tests about bbolt HOT 3 CLOSED

Might be an etcd issue with how bbolt is used, haven't investigated yet.

from bbolt.

It's similar to etcd-io/etcd#17223.

During etcd shutting down, etcd may panic due to race condition (e.g. the backend transaction has already been closed) if

• a client request is received; e.g etcd-io/etcd#17223.
• the applying workflow is in progress. e.g. this issue

High level solution/idea... when terminating an etcd instance,

• Firstly stop receiving any client & peer requests (e.g stop all the gRPC servers and http servers)
• Wait for all the existing requests (either from client or from peer) to finish. We may also need a timeout, e.g. 30s.
• Shutdown all internal services, e.g. the backend.

from bbolt.

Right, looks like duplicate of etcd-io/etcd#17223

from bbolt.