sync v2 implicit todevice acks considered incredibly harmful.

Nobody has claimed this yet.

I think we’ve tracked the main final reliability problem on matrix-video-chat to the fact that people leave instances lying around in bg tabs, which sit there running a sync loop which races to consume msc matrix-org/matrix-spec-proposals#3401 todevice messages.

I think the race is that if one sync loop calls /sync before the other tab’s /sync returns, it can “steal” the todevice messages out from the other loop by acknowledging and thus deleting them. As a result, half your todevice messages go missing (hence all my accusations in https://github.com/matrix-org/synapse/issues/9533 about unreliable todevice msgs).

@kegsay - any suggestions on a backwards compatible change to v2 sync to fix this so we’re not blocked waiting for sliding sync to land?

See also https://github.com/matrix-org/synapse/issues/6450 as another instance of this wreaking havoc.

Start with the v2 sync and to-device message behavior described in this issue, then read MSC3401 and the linked Synapse issues #9533 and #6450. A useful outcome would be an agreed, backwards-compatible change that prevents concurrent sync loops from losing to-device messages; the issue does not name implementation files or tests.

Written by the indexing model from the issue text.