Email or username:

Password:

Forgot your password?
Top-level
Darius Kazemi

Wow. Stats on federated events consumed by our server this week:

Nov 3-5: 280k events each day
Nov 6: 410k events
Nov 7: 764k events
Nov 8 (which is not even over yet): 1.7 million events

This is a sustained increase over the course of about a week. I have never seen anything like it. Usually what we see in an influx is a one or two day bump and then dropoff. I have never seen a week of solid rampup before. This feels very different.

35 comments | Expand all CWs
trendless

@darius excuse my n00bness; is there any similarity or comparison to 'jobs' that would be monitored/tabulated by sidekiq?

Darius Kazemi

@trendless that's what I am counting here! more or less each job is a single federated event -- a favorite, a follow, a post, etc

trendless

@darius that's a great way of articulating it. Thanks! We went from steady 5k prior to Oct 26 to 800k yesterday. Not what we expected, but a good problem to have 🙌

Anarcho Urbanist

@darius what does friend.camp run on btw?

Darius Kazemi

@anarchistbicycleclub Hometown v1.0.6+3.5.2 running on a Linode "Dedicated CPU" 4GB plan as specified here linode.com/products/dedicated-

Anarcho Urbanist

@darius how have you found that in terms of speed this last week, is federation slowing down a lot for you?

Anarcho Urbanist

@darius thank you! How have you found that in terms of speed this last week, is federation slowing down a lot for you?

Darius Kazemi

@anarchistbicycleclub @anarchistbicycleclub totally fine until today's huge jump. Now we have about 700 items floating the sidekiq queue (up from 0). Looking into ways to mitigate this (or if it even is really a problem. timelines seem to be realtime still (aka we're not getting posts minutes or hours after posted)

Darius Kazemi

@anarchistbicycleclub consider tweaking the settings in config/sidekiq.yml -- the number of queues for "push", the concurrency, and also the stuff in mastodon-sidekiq-push.service (DB_POOL and the "bundle exec sidekiq -c 10 -q push`, tweaking those values). I am not an expert in this stuff I just try different numbers and google to make sure I'm not being a total idiot

It_me_greg

@darius I came over from the bird site just a few days ago and I've spent maybe 15 minutes total back there since. Haven't been on it at all today. This is better.

Alek

@darius Hi Darius. I'm curious about overall sustainability / scaling of the Mastodon / Fediverse network. Do you know whether anyone is looking at this in aggregate? Is anyone calculating the server costs (increase)? I haven't seen any mention of economics of #Fediverse, but my server's admin is already mentioned sudden, additional costs of running the instance.

Darius Kazemi

@tarkowski Costs will go up across the network for sure. The economics vary from software to software (Pleroma is a lot more efficient than Mastodon, for example). On my end, this unprecedented jump looks like it's going to cost us an extra $10/mo. My server hosts 50 active users. I think the economics ultimately work out better for small servers than big ones: individuals are more likely to contribute money when they belong to a tight community of shared principles.

Darius Kazemi

@tarkowski This of course assumes that web hosting costs stay more or less constant. This might not be the case, and could be a massive disaster.

Alek

@darius thanks for the data point! it would be great, by the way, if Fediverse architecture was fully transparent in terms of traffic, costs, at best also carbon footprint.

Vicki Ziegler

@darius What constitutes an event? Is it any interaction, such as a boost, favourite, bookmark, etc.? Does it also include posts, replies, server moves? Is a new account setup an event?

Darius Kazemi

@bookgaga basically anything is an event. Any interaction (posts, boost, favs, bookmarks, replies, literally anything), plus some behind the scenes things users don't see (but these make up the vast minority). I imagine a new account setup probably triggers some small number of events, less than 5.

Vicki Ziegler

@darius Thanks for the explanation - much appreciated!

[DATA EXPUNGED]
Darius Kazemi

Something that can happen is:

- increase in fediverse usage
- numbers go up
- small servers crash/grind
- numbers go up more slowly since small servers are down
- unresponsive servers leads to servers that *are* up sending duplicate copies of messages at least a few times until they back off
- numbers go up fast
- bigger servers are stuck retrying lots of messages
- the retries themselves make big servers slow
- etc

I suspect this converges to stability over time though

Something that can happen is:

- increase in fediverse usage
- numbers go up
- small servers crash/grind
- numbers go up more slowly since small servers are down
- unresponsive servers leads to servers that *are* up sending duplicate copies of messages at least a few times until they back off
- numbers go up fast
- bigger servers are stuck retrying lots of messages
- the retries themselves make big servers slow
- etc

tim 🍓

@darius is there a point where it is unviable to run a smaller server of any kind? is that just directly correlated to the number of external follows?

Darius Kazemi

@objelisks I think we are going to have a bumpy period where Mastodon is going to need to seriously improve its overall performance. But I do think it will happen.

it's correlated (significantly though not exclusively) with number of external follows for sure. I am imagining an economics of this where if someone on a small server has a million followers, maybe the admin has to be like "hey... pay more money for your fair share of the resources here or migrate elsewhere please"

Darius Kazemi

@objelisks my big fear is that if web hosting costs go up across the board significantly for whatever reason it will just be not viable to run a small server. Fortunately there are a lot of forces in place to prevent that (namely it would probably make billionaires less rich if that were to happen, lol)

Mark Shane Hayden

@darius I suspect there is a bit of "instance churn" that contributes to this as well...

By that I mean new people who signed up for a big/general-purpose/well-publicized instance find it too slow, or has the wrong vibe or whatever decide to move instances (maybe the smaller surge back in April are now moving over during this surge? Or jsut new people instance-hopping?).

I know that instance moves can cause a brief blast of ActivityPub transactions and thus sidekiq jobs, and perhaps there are enough happening out there to make a significant contribution.

@darius I suspect there is a bit of "instance churn" that contributes to this as well...

By that I mean new people who signed up for a big/general-purpose/well-publicized instance find it too slow, or has the wrong vibe or whatever decide to move instances (maybe the smaller surge back in April are now moving over during this surge? Or jsut new people instance-hopping?).

marius

> bigger servers are stuck retrying lots of messages

@darius isn't there a back-off strategy ? I remember being frustrated that Mastodon doesn't respect when another fediverse service returns a "405 Method Not Allowed" to signal that it does not federate, but I was hoping that after a number of tries it stops.

Darius Kazemi

@mariusor yes there is a back off strategy but it still seems to clog up anyway for some reason

Matt Schneider

@darius just saw this and I’m super curious what they’re looking like now?

Matt Schneider

@darius interesting—so the surge is passed but there’s a decent amount of folks who’ve stuck with it?

Darius Kazemi

@mattlaschneider yeah we essentially doubled our prior daily average in what seems like a sustained way

Matt Schneider

@darius *wanders over to patreon* 😅

Go Up