docker can not delete endpoint when allocateNetwork fail

[Ace-Tang]

Description

when allocate network for a new container, after CreateEndpoint in daemon/containers_operations.go, docker will call ep.Join to join sandbox, in sbJoin function, according to logic now, it will update endpoint's sandbox info in datastore, if any step fail(in my position, it fails at allocate ports), it just set ep.sandboxID = "", but bot update this info into datastore.

update endpoint info with sanbox in function sbJoin

if err = n.getController().updateToStore(ep); err != nil {
         return err
  }

set ep.sandboxID="" in function sbJoin after any step fails

defer func() {
       if err != nil {
            ep.Lock()
            ep.sandboxID = ""
            ep.Unlock()
        }
  }()

Describe the results you received:
in my test-integration-cli, this problem cause stale endpoint leave in datastore.
for eg, a container named fail got this problem, and leave endpoint info in datastore, when i want to create a container with name fail, it will got this error

Error response from daemon: service endpoint with name fail already exists

my fix:

i modify my code in function sbJoin , it will fix my problem

defer func() {
       if err != nil {
            ep.Lock()
            ep.sandboxID = ""
            ep.Unlock()
+          if err := n.getController().updateToStore(ep); err != nil {
+               log.Warnf("updateToStore failed: %v", err)
           }
        }
  }()

I wonder if i am right , can someone help me

Output of docker version:
i do some modify in my docker, but allocate network logic not have lots different with moby

[sam-thibault]

This is an old issue. I will close it as stale.