delete_local_events for purge_room_history

[richvdh]

Add a flag which makes the purger delete local events

[richvdh]

retest this please

[ara4n]

isn't this too much whitespace?

[richvdh]

yes.

[ara4n]

lgtm other than stupid whitespace query.

I assume the new 'allow_empty_body' thing is necessary because previously we just crashed on trying to parse '' as JSON.

finally: i always feel a bit scared of the purge API; when we first implemented it there were dire warnings about "this will lock your synapse solid" or "you must restart once you've purged" etc. If this is no longer true, we might want to call it out in the doc.

[richvdh]

I assume the new 'allow_empty_body' thing is necessary because previously we just crashed on trying to parse '' as JSON.

we'd return a 400 NOT_JSON error or something. Sending an empty body seems like a thing that people would have been likely to do, so I'm trying to keep that working.

finally: i always feel a bit scared of the purge API; when we first implemented it there were dire warnings about "this will lock your synapse solid" or "you must restart once you've purged" etc. If this is no longer true, we might want to call it out in the doc.

It is, evidently, still true. There's an argument that should be called out in the doc...

[ara4n]

i stared at this stupidly thinking that "event persistence loop" meant the loop immediately above which is purging and persisting the new events... rather than the general BAU activity of persisting events.

If you can clarify that you're talking about "synapse takes out an exclusive lock on room_depth whenever it persists events..." it would avoid confusing bears of very little brain such as I

[ara4n]

re-lgtm other than the comment-which-i-found-confusing