Transaction IDs start at zero and are incremented.

[michaelkaye]

This has a possible failure mode if:

12:00 : matrixmon is started, sending trx_id 0 through (say 720)
18:00 : matrixmon is restarted, losing the local trx_id, causing it to start again at trx_id 0.

The next 6 hours of events will be accepted with a 200 OK by the server (because their transaction ID matches one that has already been accepted) but will not be received by the receiver, showing up as an outage.

When the restarted matrixmon reaches transactionID 721, it will start functioning again.

We should consider initalizing the transaction ID to a timestamp based value here:

https://github.com/matrix-org/matrixmon/blob/master/mon.pl#L152

[michaelkaye]

See matrix-org/synapse#8762 for an example when this restart happens too soon.

[michaelkaye]

This probably became a problem when matrix-org/synapse#8476 was merged, as after that we moved from tnx_ids being cached in-memory for ~1h or so to being persisted in the database for 24h.