Don't fetch state for missing events that we fetched

[erikjohnston]

After fetching missing events we didn't update the list of events that were missing, so while we did successfully fetch the missing events it still thought it had a hole. This meant that it fetched the state, which adds delays, CPU time and increased DB usage (since we don't delta encode those state sets)

[richvdh]

For the record, this was introduced in synapse 0.20, in 3406333

@erikjohnston: the intention was that _get_missing_events_for_pdu returns the new value for have_seen. It might well make more sense for the call to have_events to be where you have put it, but in that case you can remove the redundant call in _get_missing_events_for_pdu.

[erikjohnston]

I think its more explicit to have it done in the outer function. WIll remove from _get_missing

[richvdh]

there are other calls to defer.returnValue in this function; please update them too

also please update the docstring.

[richvdh]

while you're in there, the docstring claims that prevs should be a list, which looks like a lie. my fault, but could you fix it?

[erikjohnston]

Done

[richvdh]

haven't we already got all this at https://github.com/matrix-org/synapse/pull/2170/files#diff-3f6de39b112e0c501c3c2925a571b908R199 ?

[erikjohnston]

no?

[richvdh]

well, it looks very similar to me. what's the difference?

[erikjohnston]

have_seen has been updated the line above, and have_seen needs to be rechecked as we just fetched some missing events. Is that what you mean? Or do you mean we should wrap the call to have_seen and seen in a function?

[richvdh]

use prevs here?

[richvdh]

still unclear why it looks like we're duplicating code here

[richvdh]

(here = https://github.com/matrix-org/synapse/pull/2170/files/9b147cd73049c6222fa6c3b45dfb2a360de93cc4#diff-3f6de39b112e0c501c3c2925a571b908)

[erikjohnston]

I am completely failing to understand what you mean I'm afraid :(

[richvdh]

this is a horrible, horrible bit of code

but I think it looks ok now