Are there fundamental thresholds on responses to a single topic?


So, here’s the situation: When we livestream, we encourage folks to hangout in our community and respond to one particular topic / post and it always kills the site.

More specifically, during the livestream, we’ll do giveaways, creating a new topic in a random part of the community and then people just hammer that topic to death. Here’s an example from tonight.

We’ve moved to a HighCPU (“CPU Intensive Plan”) on DigitalO with a KeyCDN installed with this sizing:

52 PM

Regardless, it tanks the site every single time.

I suppose there is very little we can do at this point outside of investing in serious hardware and config to keep the site up when 300-400 people are trying to respond to a single post/topic at the same time?

My last thought is to do a Load Balancer via DO but that may be overly-complex for what we’re doing?

I suppose we just shouldn’t attempt a “live giveaway” via Discourse…!

1,000,000 pageviews in 2 weeks...! Why? Because Discourse. 🚀
(Alan Tan) #2

How many unicorn workers do you have configured?

Note that the high CPU droplet only has 20GB of disk space so you’ll probably want to attach a mounted volume to that droplet.

(Jeff Atwood) #3

How many replies are in that topic?

Also beware the external disks on digital ocean are quite slow, 2x or more slower than the local SSDs.


The default i believe…, which is 3. I’ll change this. moved to 8 workers and 512MB.

(ljpp) #5

You can go even higher. I beleive I have 9 workers, using an 6-core / 8GB plan. Our community also chats along with live events, so we have traffic peaks. This configuration gives us capacity to handle about 450+ concurrent users. Our biggest day of the year is the trade deadline of the local hockey league once a year, so I can only observe the maximum traffic once per 365 days.

How’s the single core performance of that DO plan? Their regular droplets are on the slow side, and UpCloud is nearly twice as fast on raw single core processing. They also have fast disks (MaxIOPS)