Comment 8 for bug 1739380

John A Meinel (jameinel) wrote :

Looking at the code to dial Mongo, I think it is correct:
  c, err := net.DialTimeout("tcp", addr, opts.Timeout)
  if err != nil {
   logger.Warningf("mongodb connection failed, will retry: %v", err)
   return nil, err
  if tlsConfig != nil {
   cc := tls.Client(c, tlsConfig)
   if err := cc.Handshake(); err != nil {
    logger.Warningf("TLS handshake failed: %v", err)
    if err := c.Close(); err != nil {
     logger.Warningf("failed to close connection: %v", err)
    return nil, err
   c = cc

The only question is if DialTimeout returns an error, would you ever need to close the connection? I would assume that 'c' is nil at that point.
And the error we're seeing in the log file is:
2018-01-11 02:26:20 WARNING juju.mongo open.go:165 TLS handshake failed: read tcp> read: connection reset by peer

"handshake failed"

looks like it should be closing the file handle correctly.

The other error that we are seeing, though, is "TLS handshake error" coming from the http package. Which is more the API server layer, than the Mongo layer.

I haven't tracked down where those are actually coming from. What is really weird is that it is saying it is happening in "log.go line 168", but we don't have a 'log.go' file in the Juju code base.

I do see several places that are creating an "apiserver" logger:
./juju/apiserver/apiserver.go:48:var logger = loggo.GetLogger("juju.apiserver")
./juju/cmd/jujud/agent/machine.go:1509: logger := loggo.GetLogger("juju.apiserver")

And the issue with the latter one is that it is part of the observer factories, thus is likely to be passed to something else.

I haven't found any obvious log.go files that are using a logger object on line 168.