My Discourse is doing great, but I’ve been checking my logs and sometimes the connection times out. I know this is not a Discourse problem per se, and more of a Docker one, but it might be useful for newbies (like me) to know how to tune our Nginx, Docker or Discourse to avoid this. I already remov…

[image] codinghorror: Not sure message bus polling timeouts really matter though. Message bus is used to poll for updates (e.g. new posts, threads) and update the page?

Yes but a “failure” there would never be visible to the user, and may not matter. Since the bus is always active, more messages will arrive later…

Based on your feedback I added a line to clarify that Unicorns should scale with CPU cores, when possible ## TODO: How many concurrent web requests are supported? ## With 2GB we recommend 3-4 workers, with 1GB only 2 ## If you have lots of memory, use one or two workers per logical CPU core #UNICOR…

(This is an old topic, but it still seems like where this belongs) [image] codinghorror: one worker per CPU core (real or logical). I would allow at least 512mb ram per worker So the “ram per worker” is db_work_mem? That seems right, but the comments there don’t seem to say what you said abo…

[image] work_mem is used several times per query. So a complex query (like suggested topics) can use 5~10 times that, and this for each query running and you are fast into swapping territory. The allow memory per work that Jeff cited, is more on don’t set 10 unicorn workers if you have a 10CPU s…

You’re back! OK, I get that I’m wrong. Is this more on target? so with 8CPUs and 12GB db_work_mem: 40MB UNICORN_WORKERS: 8 and with 4 CPUs and 8GB db_work_mem: 40MB UNICORN_WORKERS: 4 And then I think also involved is the size of the database itself, which we want to be able to get buffered.…

[image] pfaffman: You’re back! Packing my stuff and took a minute to check Meta :blush: [image] pfaffman: so with 8CPUs and 12GB db_work_mem: 40MB UNICORN_WORKERS: 8 and with 4 CPUs and 8GB db_work_mem: 40MB UNICORN_WORKERS: 4 Sounds good! [image] pfaffman: With a huge datab…

db_work_mem: 40MB UNICORN_WORKERS: 8 and with 4 CPUs and 8GB db_work_mem: 40MB UNICORN_WORKERS: 4 Hi @Falco , If we traspolate those values to use Puma, what is the recommended number of Threads per worker you suggest? In the case we are talking about a site with 1M requests per day. Cheers, …

Is this a site that your are looking at moving to discourse? For 1M/day you may need to move to a more complex setup with haproxy or traefik reverse proxying to multiple backends (which is not supported here). How big is your database?

You can do multiple backend with nginx as well and it is well within the spec. Just serve static via a CDN

How to avoid upstream timeouts?

Support

Falco (Falco) February 21, 2019, 6:44pm 20

Well, that’s what it means.

We run some instances that autoscale to hundreds of Unicorns, but we don’t really use Puma in production.

Topic		Replies	Views
Due to extreme load, this is temporarily being shown to everyone... when it's not really the case Self-hosting server-resources	19	1984	July 21, 2023
Recommendation for number of workers: cores × 2? Self-hosting	16	693	April 29, 2025
High CPU usage (Ruby) Support server-resources	20	1040	February 3, 2025
Best configurations for speeding up standalone discourse Self-hosting server-resources	25	4607	February 14, 2024
502 Bad Gateway for only two Topics Self-hosting arm	19	2246	November 17, 2022

How to avoid upstream timeouts?

Related topics