I made this based on the gripe about some of the silent failures with federation. Might help users choose other servers. Might help admins troubleshoot. Open to comments and criticisms!
Oooohhh … Nice!! I’m repeatedly impressed at how many hackers are going ahead and just getting some stuff done here!!
Questions/thoughts:
- What instance is used as a reference for the delay? One you self-host (lemmy.management)?
- Sooo … what’s the deal with lemmy.ml … that seems to have gone beyond lag and is basically falling over … seems like the devs have neglected their own instance’s health?
- What’s that
Redash? Is it a plotly thing or some other product that just uses their graphing library? How have you found it?
What instance is used as a reference for the delay? One you self-host (lemmy.management)?
Yes. lemmy.management. It is purposefully updating subscribed communities to as many as possible (via automation.) This doesn’t correct for network lag, but the idea was to capture the “federation” lag. There’s no code I’m aware of that allows admins to prioritize outbound federation traffic. I could be wrong though.
Sooo … what’s the deal with lemmy.ml … that seems to have gone beyond lag and is basically falling over … seems like the devs have neglected their own instance’s health?
I just collect the data.
What’s that Redash? Is it a plotly thing or some other product that just uses their graphing library? How have you found it?
https://redash.io I don’t remember how I found it. Probably an “awesome” list on github.
On mobile, when touching the “Federation Lag-o-meter (now - 1h)” statistics, the page is hard to scroll. Other than this the page is gold
When I saw the bar looking like the Burj Khalifa, I assumed it was .world instead of .ml. Interesting.
Props to Ruud@lemmy.world for dealing admirably with the Rexxit hug of death.
I’m expecting that JSON parsing is a huge overhead with the fediverse. I work on a SAAS that needs to do all its internal processing in under 10 ms, and serializing/deserializing ends up being a sizable chunk of server time. I saw a 40% reduction in runtime using simdjson for deserializing, and there exists a rust crate for it, but I haven’t had time to look the Lemmy code over.
Can anyone with an overloaded instance get on their command line and gather a decent flamegraph so the performance folks can aim optimizations in the right direction?
Graph should remove the outlier as it is skewing the results for every other instance and not letting to see smaller numbers show up.
Or we should move to log scale so that it can be displayed correctly.
Great idea. I was trying to figure out if it was lemmy.world trying to deal with new users or a bug with Memmy app that caused random errors
Is it possible to have the lag metrics by instances in a table format? Its so hard to view your site on mobile
I didn’t even load it on mobile. I will check it out tonight and maybe just create a separate “mobile friendly” dashboard.
