If an event takes ages to persist it blocks other events from being notified

[erikjohnston]

This is because we assign a stream ordering fairly early on, and we block notifying /sync streams / replication / etc about later events until the slow event has finished persisting.

One mitigation could be to assign a stream ordering as late as possible inside the transaction, to minimise the number of queries that could block future events.

[clokep]

@erikjohnston did one of your recent PRs fix this?

[erikjohnston]

@erikjohnston did one of your recent PRs fix this?

Alas not, no. (I think we fixed an underlying cause of slow persists, maybe? But not this issue itself)