Now at 1700 new accounts on the day @infosec.exchange. This is what it looks like on the load balancer:
And this is what's happening at the CDN:
And this is what the aggregate CPU on the PUMA servers look like (note, almost no change)
And this is the aggregate load on the sidekiq servers (also, nearly no change)
Transaction growth on the postgresql database:
CPU on the redis/elasticsearch server. I can barely find it:
final graph (I don't have much else that would be interesting) is the sidekiq graph:
I lied. One more. I feel bad that I am just now realizing I didn't tell Prometheus to monitor the load balancer, so here is a snapshot of nmon from the load balancer:
ok.... one more for the sidekiq nerds:
@jerry that failure count 🥺
@vmstan there are an amazing number of other instances that are down. some for good
@jerry that's a lot of failed jobs. What's the main cause of those? Failed requests to other servers?
@postmodern yes - 98% (give or take) are other instances that are dead/offline/not responding/error 500/etc
@jerry wow how many users do you have to require this much sidekiq threads ?
@nicdex 37k. I went through some rough times with Sidekiq a month ago. I never want my queues higher than 0 😂

@jerry kinda surprised you dont have a giant linux terminal with:
- htop
- nload
- iftop
- atop
- ioping

and prolly asciiquarium and cmatrix for good measure :D

Glances - An Eye on your system

Glances is a cross-platform curses-based system monitoring tool written in Python.

@pixelnull @Viss oh that looks amazing

@jerry @Viss it has a package avail for most distros.

i wouldn't do a github or pip install for it, they can get a little experimental with releases in my exp

@pixelnull @jerry oh yes! glances! i always forget about glances. You need a big terminal for it tho
@Viss @pixelnull I have two 43" monitors 🎉​
@jerry @pixelnull heh, i excitedly await the screenshots :D
@Viss @jerry nono take a picture of the monitors with your phone... make sure to brush your teeth in front of them tho. they need a bunch of gunk on them

@pixelnull @jerry

brew install glances

@Viss @pixelnull @jerry brew is always like "ah, if you wish to install an apple pie you must first install the universe"
@aburka @Viss @pixelnull @jerry Yep, though I would 1-up you and say that trying to deploy software that still uses Python 3.6 & Pipenv is much worse....trust me!

@compuguy @aburka @Viss @jerry

python packaging is a dumpster fire, nestled in a tire fire, that is then placed in a high-rise apartment building fire.

a fire-turducken as it were

@pixelnull @aburka @Viss @jerry Yep, but that's what the developers I work with use. That and Node/NPM. I call it "Dependency Hell"....
@jerry whoa those are some really interesting spikes. can you tell what they are?
@Viss @jerry lots and lots of INSERT INTO?
@Viss Not really. I have neglected to instrument the basic linux stats on the load balancer. That is simply queries that nginx is responding to, so I assume that it is probably related to something that causes lots of other instances to hit the server all at once, like a large account mentioning something here.
@jerry interesting. yeah cuz the 'average load' looks to be considerably less than those spikes, so something is causing.. i guess a shitload of queries to happen all at once, in bursts?
@Viss exactly. If someone on an instance with followers on thousands of different instances posts a link that references this site, all those instances reach out. This is what the attack of last week was all about - creating many accounts that caused instances to attempt to DDOS some other sites.
@jerry ah ha - illumnation of burst-style DDOS scenarios. too many signups spike psql, psql chokes, front end throws 500s
@Viss @jerry probably every time Elon posts on twitter. Lol.
@throAU @jerry well, banning journalists and letting nazis on the site were two big spikes, for sure
@jerry i wonder how soon there's gonna be a patch to replace sidekiq with something else.
@Viss @jerry Whenever that is, it wouldn't be too soon for my taste.
@nullagent @jerry right? the biggest improvement to twitter back in the day was when they got the fuck off of ruby on rails
@Viss perhaps. Now that I have experience with it, it's not the worst thing in the world.

@jerry so it looks like other folks have figured out how to migrate from sidekiq to rabbitmq https://stanko.io/rabbitmq-is-more-than-a-sidekiq-replacement-b730d8176fb

(we run rabbitmq. its way more burly. we dumped redis to go to this)

RabbitMQ is more than a Sidekiq replacement | Stanko K.R.

A rabbit dressed up as the karate kid - a cross of RabbitMQ and Sidekiq logos I've had gripes with Sidekiq because of which I switched to RabbitMQ. Here a...

@Viss interesting - will have to look at it
@jerry I imagine that its not gonna be super easy or straightforward to migrate from sidekiq to rabbitmq, but I would imagine that rabbit will treat your cpus and ram way better than sidekiq does
@jerry @Viss oooh nice.
@danhon @jerry this post doesnt have anything to do with mastodon specifically tho - but, you gotta start somewhere :D
@Viss @jerry I doubt it. I've followed a guide and spit off the scheduler and mailer queues to a separate container. The rest can be scaled up to meet demand.
@jerry wow, that is quite an increase in traffic.
@jerry when you moved to object storage, was that a massive i/o improvement?
@Viss I had the media on their own cluster, which were pushing ~500Mbps each at times
@jerry whoof, thats nutty. ~17kish active users, was it? or was it 12k? i remember your older post saying you had like 27k users or something but a lot of those signups were like 'landgrabs to secure their handle' then they never post
@jerry fuck i'm glad you're a good admin
@jerry that looks like quite the traffic, the sine wave looks beautiful though.
@jerry I am having a hard time getting a decent list of who’s really new on the directory page for some reason. I want to greet folks…