Default sync_response_cache_duration can lead to clients never being able to initial sync

[turt2live]

Description

Large accounts or slow servers can mean that the client will retry /sync (without ?since) until it gets a result. However, the server's default cache time for sync responses seemingly went from some number of minutes to zero. This means that the server processes large /sync requests in the background but immediately throws away the result if the client gave up waiting and tried a second request.

There should be a non-zero default on the config value.

Steps to reproduce

• Be in thousands of rooms
• Have a slow-ish server
• Try to sync your account

Homeserver

t2l.io & t2bot.io

Synapse Version

1.60.0

Installation Method

pip (from PyPI)

Platform

N/A

Relevant log output

N/A

Anything else that would be useful to know?

No response

[reivilibre]

This is sounding similar to (maybe the cause of) #12971.

[erikjohnston]

However, the server's default cache time for sync responses seemingly went from some number of minutes to zero.

I don't believe it has ever been non-zero ftr, at least I can't see any evidence of that in the code.

[chagai95]

This was bothering me for a good few months, so happy I found this, thx so much for reporting!

[turt2live]

However, the server's default cache time for sync responses seemingly went from some number of minutes to zero.

I don't believe it has ever been non-zero ftr, at least I can't see any evidence of that in the code.

I have vague memory of it being 5 minutes, though that could have been a matrix.org hotfixes thing. Prior to the config setting it was easier to get a successful sync, at least.