federation issue to instance
federation issue to instance
I cant seem to get timely post federation to my instance (moist.catsweat.com) from lemmy.world. I might see a new post every few hours instead of every minute or so.
posts from other lemmy instances (lemmy.ca, lemmy.ml, dbzer0, etc) have no problem.
i see a ton of lemmy.world traffic in my nginx logs... it seems very specific to 'new posts'. voting/comments seem to come through no problem once the post shows up.
ive checked a few other instances similar to mine, and they are not having this issue. so it seems something specific to lemmy.world and my instance.
i do recall a similar issue several months back, and it involved a lemmy.world admin resetting some outbound queue.
any ideas?
You're viewing a single thread.
I can't tell you why you're lagging but you're clearly lagging quite a bit behind.
which country is your instance located in?
did you (or someone else on your instance) recently subscribe to a bunch of high traffic communities on lemmy.world, which would make lemmy.world send more activities to you?lemmy by default only sends activities in a community to another instance if there's at least one subscriber to the community on that instance. if you're located far from finland, where lemmy.world is located, you might have been able to keep up just enough before this, although this isn't the first time as the graphs above show.
Hey, sorry for the late reply, but I'm trying to figure out who runs this graph, because if it needs something like a bit of funding to store data for longer periods I'd like to help. Do you know, or know where you heard about it?
Hi, I run this.
What benefit do you expect from longer retention periods and how much time did you have in mind?
The way data is currently collected and stored keeps the same granularity for the entire time period, which currently uses around 60 GiB for a month of retention across all monitored instances.
Mainly, I'm interested in how Lemmy is growing and changing as a whole. If there was a way to store activity just weekly or even monthly that would help.
so all you're looking for is the amount of activities generated per instance?
that is only a small subset of the data currently collected, most of the storage use currently comes from collecting information in relation to other instances.
Yep. As far as I know nobody else collects that information, though.
I'll probably have to look at another storage than prometheus, aiui it's not really well suited for this task.
maybe something with influxdb+telegraf, although i haven't looked at that yet.
Thanks for the attention to this! If I can help in any way just say so.
do you happen to have experience with setting up influxdb and telegraf? or maybe something else that might be better suited?
the metrics are currently in prometheus metrics format and scraped every 5 minutes.
my idea was to keep the current retention for most metrics and have longer retention (possibly with lower granularity for data older than a month).
the current prometheus setup is super simple, you can see (and older copy of) the config here.
if you want to build a configuration for influxdb/telegraf that i can more or less just drop in there without too many adjustments that would certainly be welcomed.
the metric that would need longer retention is
lemmy_federation_state_last_successful_id_local.I don't have experience, but I can make a project out of it nonetheless!
its running on AWS, US. you can see on 4/24 i finally solved for a long running bottleneck (disk writes), and it was smooth sailing until recently.
im not sure if anyone else has subscribed to any new communities, but ive been subbed to most of the big ones for awhile.
i don't see any saturation in the server network activity, connections, etc. my queues are not overloaded, and as i mentioned other instances are very performative.
maybe i just need to be patient and let it catch up. those graphs are awesome
lemmy currently only sends one activity per receiving instance at a time, so there is a round trip for every single post, comment, vote, etc., before the next activity will be sent. you wouldn't see any increased number of connections, as there's only a single one.
do you have access logs for
/inboxwith the lemmy.world's user agent? you might be able to derive some information from that if requests increased over time or something, maybe also response status codes?