Slow performance on ARM64 docker image #3740
Comments
|
Hi, this is not ARM related - I have spun up an instance yesterday and it is so slow, that thumbnails take hours to load or don't load at all. A click takes minutes... x64 system Edit: it also is the same on other instances which worked well until now |
|
Do you hourly restart your instance? |
I do not. I had been restarting the instance a few times to capture logs and data with different config settings. |
|
It does not matter if I start it fresh or it runs 5 hours - same thing happening. I have no useful logs for now either. After restarts it seems even slower. I have 32GB RAM, 2400G on Arch and 300MBit Downstream with 7ms latency. |
|
I'm in North America, so it doesn't seem like its a CDN issue. |
|
DB version from 13 to 14 (development yml has 13 in it!) and fetch the image, don't build it from scratch. Need to test some stuff and I'll come back in a few days. |
|
Working now quicker than public instances. I changed my DNS from 1.0.0.1 and 1.1.1.1 to ISP defaults (maybe also other DNS filters like AdGuard will interfere?). Apache settings: When switching from Postgresql 13 to 14 you need to delete the volume "postgresdata" volume because the data is not compatible. Do a full export of your subs and watches before and reimport later.
Invidious docker-compose: Request are now in the <500ms range instead of 30s + |
|
I completely removed all Postgres info, changed the DB to v14 and rebuilt it from scratch. The site is faster, but the images still don't load and video watching/downloading is terribly slow. |
|
Did you try what @blackandcold said? using a different DNS server |
|
Can it be the pubsubhub? 438 - 465 Since it requires a callback your server would also need to be reachable from the internet. |
I've also tried Quad9, Cloudflare and Google DNS with the same results. Example after changes: 023-04-15 19:11:58 UTC [info] jobs: running ClearExpiredItems job |
|
After enabling 'hmac_key' and 'use_pub_sub_feeds: true' the performance is a little better, but images still wont load and searching is slow. |
|
While searching, trending, popular and general operations are slow, downloads are much faster now after the recent change. It just takes a long time for the search and page load of the video with no thumbnail. |
|
well, all that is left are the logging block in the config and the thread counts. neither should influence search, thumbnails or general operations. |
|
Want to confirm after playing with some config settings. Still the same issue with slowness, but at least when I click to play the video, it plays flawlessly. Just still no thumbnails and terribly slow navigation between pages. Not sure what else to check. Logs aren't reporting anything really other than very long load times. |
I have confirmed that development is not in any logs. If I knew where to look at the code, I would investigate myself. |
|
I'd highly recommend running multiple docker containers in parallel. Here is as an example the docker-compose config I'm using on my instance. version: "3"
services:
invidious:
image: quay.io/invidious/invidious:latest
restart: unless-stopped
deploy:
mode: replicated
replicas: 7
endpoint_mode: vip
restart_policy:
condition: any
delay: 30s
window: 30s
resources:
limits:
cpus: '0.75'
memory: 1500M
reservations:
cpus: '0.15'
memory: 400M
ports:
- "127.0.0.1:3000-3006:3000"
# The rest of your config goes here |
|
I reinstalled the same OS on a new nvme and deployed the same exact docker-compose files and it worked flawlessly. Not sure what the original problem was, but all works great now. |
I have recently changed from an AMD64 docker image to an ARM64 for my invidious instance. Everything ran great on the x86 server deployment in docker using docker-compose. I've copied over the exact config, but changed the image to:
quay.io/invidious/invidious:latest-arm64I am experiencin terrible performance since the switch. The thumbnails don't really load, but it shows the text for latest and trending. Search function is also fairly slow. When I open a video I can eventually get it to load and play, but it buffers constantly or may never play at all. I have tried accessing it through my swag nginx reverse proxy or on my internal network with multiple devices and can replicate the same issue. The hardware should be more than enough to run invidious. Specs are below:
_Orange Pi 5 16G memory unit
RK3588 CPU
500G M.2 NVME
Docker version 23.0.3, build 3e7cbfd
ls_release -a
Distributor ID: Debian
Description: Debian GNU/Linux 11 (bullseye)
Release: 11
Codename: bullseye_
docker-compose.yml file:
invdious container log snippet:
postgresdb container log snippet:
Screenshot of experience:
The text was updated successfully, but these errors were encountered: