from gyoo@lemmy.ca to selfhosted@lemmy.world on 10 Jun 23:30
https://lemmy.ca/post/66232628
I am trying to increase the upload limit on my matrix server. I would like to push it to 500MB, but it keeps failing the upload past ~180MB. There is very little information about the failure, the client just says “Unkown error”, nginx says “error 499”, which means the client disconnected to quickly. Additionally, trying to recreate the failure with curl just results in a success, so the issue seems to be with the client. I tried with element-web, element desktop, element x and cinny and they all have this issue.
I run a continuwuity server on a Raspberry Pi 4B, behind two nginx reverse proxies and a rathole tunnel. Continuwuity, and the two reverse proxies run inside docker compose projects. Here is my architecture :
__
.( ).
(`Cloud )
`-----`
|
V
+- Outside server -+
| +---------+ |
| | rathole | |
| +---------+ |
+--------|---------+
|
+- Home network ---|-------------------------+
| +- RasPi --------|-----------------------+ |
| | V | |
| | +- tunnel docker -+ | |
| | | +---------+ | | |
| | | | nginx 1 | | | |
| | | +---------+ | | |
| | +-----|-|---------+ | |
| | | | | |
| | | +- other services -->... | |
| | | | |
| | V | |
| | +- matrix docker --+ | |
| | | +---------+ | | |
| | | | nginx 2 | | | |
| | | +---------+ | | |
| | | | | | |
| | | +------V-------+ | | |
| | | | Continuwuity | | | |
| | | +--------------+ | | |
| | +------------------+ | |
| +----------------------------------------+ |
+--------------------------------------------+
Of course, I made sure to increase the upload limit in Continuwuity :
max_request_size = 500000000
And in both nginx instances :
client_max_body_size 500M;
From what I can gather from other issues on the internet, it seems it could be related to timeout errors. Although they all seem to be talking about 2-8Gb files, I am probably running much weaker hardware than they do, so it checks out.
I tried setting the following values in both nginx instances :
location / {
proxy_connect_timeout 600;
proxy_send_timeout 600;
proxy_read_timeout 600;
send_timeout 600;
proxy_next_upstream off;
proxy_buffering off;
proxy_pass ...;
}
And setting timeout values in Continuwuity :
client_receive_timeout = 75 client_request_timeout = 180 client_response_timeout = 120 client_shutdown_timeout = 10
Since that didn’t have any effect, I bumped them by 20x and 40x respectively :
location / {
proxy_connect_timeout 18000;
proxy_send_timeout 18000;
proxy_read_timeout 18000;
send_timeout 18000;
proxy_next_upstream off;
proxy_buffering off;
proxy_pass ...;
}
client_receive_timeout = 3000 client_request_timeout = 7200 client_response_timeout = 4800 client_shutdown_timeout = 400
But that did not have any effect. I also looked at modifying element-web’s settings, but there doesn’t seem to be anything related to timeouts.
I also saw that encrypted vs. unencrypted rooms might have an effect, but I got ~190MB max in encrypted rooms and ~180MB in unencrypted rooms.
Finally, I tried using curl :
curl -X POST https://server.domain/_matrix/media/v3/upload -H "Authorization: Bearer <token>" -H "Content-Type application/octet-stream" --data @200mb.file # success # returns a mxc:// url curl -X POST https://server.domain/_matrix/client/v1/media/download/server.name/file_id -H "Authorization: Bearer <token>" -o 200mb_copy.file # See that the two files are identical
I am at a loss. What can I do to solve this issue?
#selfhosted
threaded - newest
I’d try on the PI against the service port to remove all the extra complexity. If that doesn’t work, fix it and move on to the next layer, fixing as needed until you get to the end.
How would I do that? The curl command already works and the service is being handeld by the reverse proxy, which depends on the domain name to pass the requests correctly. Even if I succeeded, element-web is configured to send the file to the domain name, not the local address.
I darkly remember that there was another relevant size settin in matrix…
Relevant settings from my homeserver:
That’s synapse, not continuwuity, correct?
Yeah, I can’t find either of those settings in the configuration docs and I wasn’t just sending images.
Why is there two nginx containers? Can you do nginx1 --> continuwuity?
Does continuwuity show any logs? What about nginx? Check Element Web’s devtools, does the 499’d network request say anything of note?
Maybe have a chat in the support room
The idea is to have one nginx container responsible for the tls handling and request distribution (I have more than one service running on the device) and the other for actually handling the request. It keeps things organised.