Skip to content
This repository has been archived by the owner on Apr 26, 2024. It is now read-only.

Presence is worse on 0.99.3 #5081

Closed
turt2live opened this issue Apr 19, 2019 · 4 comments
Closed

Presence is worse on 0.99.3 #5081

turt2live opened this issue Apr 19, 2019 · 4 comments

Comments

@turt2live
Copy link
Member

Possibly related: #4713
Related: #3971

https://snapshot.raintank.io/dashboard/snapshot/7FO0m6GfHbrA183bQ60ukgl7VQU8lJy2

At roughly 19:00 UTC in that graph I upgraded Synapse from 0.99.0 to 0.99.3 on t2bot.io - shortly after the server started melting due to presence EDU spam. Traffic in rooms does not appear to be any different, but going from under 1Hz of EDUs outbound to 25Hz+ is a bit drastic.

@turt2live
Copy link
Member Author

turt2live commented Apr 19, 2019

This might actually be a change in how the metrics are populated? It lines up perfectly with the outgoing transaction rate after 19:00, but not before. 40Hz+ of outgoing transactions is normal for t2bot.io, but there is an unexplained gap between the number of PDUs and EDUs going out versus the transaction rate.

Edit: The graph in question being https://snapshot.raintank.io/dashboard/snapshot/80FmwXBJSMHBDeq01EKhB8KqHFAbNcPr

@erikjohnston
Copy link
Member

I'd be interested in what synapse_federation_client_sent_edus_by_type metric looked like in that time

@turt2live
Copy link
Member Author

Was m.presence, which has the same line as EDUs from the snapshots.

I think it's just recording it differently now, which means that presence gets counted.

@turt2live
Copy link
Member Author

after some research, it is indeed the metrics being recorded differently. Not sure when it changed, but the outbound traffic is the same - it just looks different in grafana.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants