Fixing parlai onboarding

[JackUrb]

Overview

#421 included a bug for specific cases of ParlAIBlueprint onboarding. This fix resolves the issue, and also resolves an issue where the timeout status wasn't being pushed forward in onboarding.

Details

#421 introduced a bug specifically for ParlAIBlueprint onboarding worlds, as the ParlAIBlueprint is the only type of live task wherein either the backend (the World) or the frontend (via handleSubmit) can submit the onboarding. In the case that the world exited cleanly, this would lead to a message to the frontend telling it to submit.

Unfortunately, #421 improved the onboarding flow to ensure that the transition from running onboarding to registering would happen in the same order every time. Specifically, the requirement is that _on_submit_onboarding now has to trigger the termination of the onboarding runner. In the case where the backend triggered this for parlai onboarding by just exiting an onboarding world, that wasn't the order things happened in.

This updates the ParlAChatITaskRunner to first mark the agent as done, then wait for the submit packet (to come through _on_submit_onboarding before terminating.

Testing

Launched a task world with onboarding worlds that submit on the backend, was able to transition into the main world with no refreshes, no disconnects, and no other behavior.
Launched a task world with onboarding, timed out, then ensured the timeout was pushed to the frontend.

[mojtaba-komeili]

question: could we have this as a variable in case we wanted to find the optimal value?

[mojtaba-komeili]

The code looks good to me. I tried this in mock, sandbox and live. Mock worked well. In sandbox, when it was transitioning it looked like it was refreshing a bit too long, but it was generally good. The live (prod) version is going fine but a bit too slow (that might be related to other factors though).

[codecov-commenter]

Codecov Report

Merging #439 (352aa28) into mturk-hit-sync-speedup (d222f4d) will increase coverage by 0.11%.
The diff coverage is 46.15%.

@@                    Coverage Diff                     @@
##           mturk-hit-sync-speedup     #439      +/-   ##
==========================================================
+ Coverage                   65.95%   66.07%   +0.11%     
==========================================================
  Files                          78       78              
  Lines                        7084     7097      +13     
==========================================================
+ Hits                         4672     4689      +17     
+ Misses                       2412     2408       -4     

Impacted Files	Coverage Δ
.../blueprints/parlai_chat/parlai_chat_task_runner.py	28.00% <14.28%> (-0.68%)	⬇️
mephisto/operations/supervisor.py	79.48% <83.33%> (+1.47%)	⬆️
...tractions/architects/channels/websocket_channel.py	80.23% <0.00%> (+1.16%)	⬆️
mephisto/abstractions/architects/mock_architect.py	91.50% <0.00%> (+2.61%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d222f4d...352aa28. Read the comment docs.