| 6 Feb 2022 |
hexa | that is otherwise idle | 20:12:28 |
hexa | this is why I need workers. | 20:12:33 |
hexa | Dandellion, Sumner Evans care to share? | 20:12:47 |
Dandellion | In reply to @hexa:lossy.network Dandellion, Sumner Evans care to share? https://gist.github.com/dali99/4bbca68b14dcf312f22d776acc9075ad | 20:20:55 |
hexa | that is quite something 😀 | 20:33:36 |
Sumner Evans | In reply to @hexa:lossy.network Dandellion, Sumner Evans care to share? https://git.sr.ht/~sumner/nixos-configuration/tree/master/item/modules/services/matrix/synapse/default.nix | 21:18:16 |
hexa | uh, what's the difference? 😀 | 21:18:40 |
Sumner Evans | Mine is probably worse. That is the main difference haha | 21:19:48 |
hexa | thanks for the honesty 🙂 | 21:20:12 |
Dandellion | mine is,, also pretty bad,,, | 21:58:12 |
hexa | ok, I think I've migrated to the module and things are working (without workers) | 21:50:38 |
| 7 Feb 2022 |
Ralith | I really don't understand how synapse manages to be so incredibly CPU-intensive tbh | 18:54:33 |
| 6 Feb 2022 |
hexa | and I've apparently lost my very own avatar | 21:50:48 |
| 7 Feb 2022 |
Ralith | also I/O-intensive | 18:54:42 |
| 6 Feb 2022 |
hexa | ah, different media store paths | 21:52:55 |
| 7 Feb 2022 |
Moritz Hedtke | Do you have experience with dendrite? Because I think lots of it is because of bad implementation and some of it is because decentralized means you have to cope with many huge rooms and lots of slow other federated servers | 19:58:49 |
| 6 Feb 2022 |
hexa | Dandellion: can you share your worker and reverse proxy configuration? | 22:02:52 |
| 7 Feb 2022 |
Dandellion | it's a lot of connections, since federation is per room and fully meshed | 20:05:49 |
| 6 Feb 2022 |
hexa | also … wondering how you handle the worker_shared_secret | 22:03:13 |
| 7 Feb 2022 |
Ralith | I'll muck with dendrite when it's mature and has a migration story and not before | 20:06:18 |
| 6 Feb 2022 |
hexa | * also … wondering how you handle the worker_replication_secret | 22:03:22 |
hexa |
Extracting a user ID from the access token or Authorization header is currently left as an exercise for the reader.
| 22:14:03 |
hexa | awesome, so from university I know that this is trivial. | 22:14:22 |
hexa | hm ok, feels much snappier | 23:24:41 |
hexa | still having issues with the client endpoints sadly | 23:24:52 |
hexa | the requests correctly got forward to the client workers | 23:28:02 |
hexa | but resulted in http 400s | 23:28:29 |
hexa |
SynapseError: 400 - Unrecognized request
| 23:28:34 |
hexa | # cat /nix/store/0mi0bv1r1gnxa53k3ljhcfk4jzlx1vcf-matrix-synapse-worker-client1-config.yaml
worker_app: synapse.app.generic_worker
worker_listeners:
- bind_addresses:
- localhost
port: 8010
resources:
- compress: false
names:
- client
tls: false
type: http
x_forwarded: true
worker_name: client1
worker_replication_host: 127.0.0.1
worker_replication_http_port: 8009
| 23:29:12 |
hexa | not sure why a generic_worker with a client resource would not be able to handle that request | 23:29:35 |