Issues with clean session (clearSession)

[jbwdevries]

When connecting to a server without a clean session, the application does not seem to connect properly.

On the server side, I see:

1571925100: New connection from 192.168.48.135 on port 443.
1571925100: New client connected from 192.168.48.135 as session-f931762b-97c6-49b6-a18a-e2daaf62dc57 (c0, k100, u'stresstest').
1571925102: Client session-f931762b-97c6-49b6-a18a-e2daaf62dc57 disconnected.
--
1571925130: New connection from 192.168.48.135 on port 443.
1571925130: Client session-f931762b-97c6-49b6-a18a-e2daaf62dc57 disconnected.
1571925130: New client connected from 192.168.48.135 as session-f931762b-97c6-49b6-a18a-e2daaf62dc57 (c0, k100, u'stresstest').

As you can see, the server first logs a disconnect event, and then a connect event. But it also earlier logged a disconnect event for that same session ID.

The client will PINGREQ and PINGRESP properly, but not receive any messages. Possibly because the subscriptions aren't restored?

I can't really understand it. I've attached a test case scenario: Stresstest.hs.gz. I've compiled it by adding this to the cabal file:

executable stresstest
  hs-source-dirs:      app
  main-is:             Stresstest.hs
  ghc-options:         -threaded -rtsopts -with-rtsopts=-N
  build-depends:       base
                     , bytestring
                     , c-mosquitto
                     , containers
                     , data-default-class
                     , options
                     , uuid
  default-language:    Haskell2010

This scenario creates a clean session, then spams a bunch of messages to this session from a different client, and then reconnects with the session.

https://github.com/dustin/mqtt-hs does not have the same problem, so the bug is probably not in the server: mqtt-hs-stresstest.gz

I'm running mosquitto 1.4.15, there is no mention in the changelog of any bug fixes related to clean session since that version. And I'm having a hard time imagining that this would be a bug in the C library. But I can't be sure.

Any suggestions on why this might be?

[tolysz]

I do not see ( https://github.com/tolysz/c-mosquitto/blob/master/app/Main.hs ):

 M.onConnect m $ \c -> do
           print c
           M.subscribe m 0 "#"

I might have some time later to test it a bit.

[jbwdevries]

Right, the .hs file is missing a subscribe call, not sure how that happened. Below a fixed version:
Stresstest.hs.gz

Same bug still shows, though.