Appservice worker appears to die when an AS catches up

[turt2live]

Three times now I've run into an issue where the appservice worker process exits after the Telegram bridge catches up. This has only started happening since my 0.34+python3 upgrade (both happened at the same time).

The last couple lines in the log each time are:

appservice_1 - 2018-12-29 00:22:00,414 - synapse.appservice.scheduler - 171 - INFO - as-recoverer-telegram-7- Successfully recovered application service AS ID telegram
appservice_1 - 2018-12-29 00:22:00,415 - synapse.appservice.scheduler - 172 - INFO - as-recoverer-telegram-7- Remaining active recoverers: 0

For perspective, the Telegram bridge stacks up hundreds of transactions after a restart for ~15 minutes. After a while, it catches up and starts streaming normally, assuming the appservice sender is still working.

I have not seen the worker survive the Telegram bridge restarting in the 3 times I've done it.

[richvdh]

Hopefully fixed by #5885