| 3 Jun 2025 |
 hexa | updating the macos builders to 15.5 | 18:40:46 |
 Arian | Most of the HTTP 1.1 requests seem to be coming from Fastly (??)
https://github.com/NixOS/infra/issues/711#issuecomment-2936898473
| 19:43:15 |
 edef | oh, that is the shield datacenter mechanism i suspect | 19:43:37 |
| edef | you have to filter the internal requests out of the logs | 19:43:48 |
| Arian | ah | 19:43:53 |
| edef | iirc we don't have a flag for accurately selecting them | 19:44:06 |
| edef | * iirc we don't have a flag for accurately selecting them by | 19:44:32 |
| Arian | yeh so all the user agents are Nix in the top-most amount of requests | 19:47:40 |
| Arian | So I feel like we might have a nix bug that makes some request use HTTP 1.1 for whatever reason or really a lot of people are behind weird corpo proxies | 19:48:11 |
| hexa | also Lix from AS8881 is Versatel | 19:49:28 |
| hexa | which is an eyeball network | 19:49:31 |
| hexa | * also Lix from AS8881 (Versatel 1&1) | 19:49:41 |
| Arian | oh the top result is actually no user agent | 19:50:54 |
| hexa | so if anything something before nix 2.19 and nothing newer | 19:50:57 |
| hexa | Redacted or Malformed Event | 19:51:37 |
| hexa | pennaes first suspicion is that h1 would be used if the h2 connection setup failed | 19:52:21 |
| hexa | there is no explicit fallback, so either users explicitly configuring it or a curl thing | 19:53:23 |
| Arian | so the no-user-agent requests from Amazon are only requesting narinfo files it seems | 19:57:31 |
| Arian | so I guess some scraper | 19:59:59 |
| Arian | other theory: some Fastly PoPs fail to negotiate HTTP 2.0 for some reason | 20:07:23 |
| hexa | uh, we only use the shield pop | 20:07:38 |
| hexa | https://github.com/NixOS/infra/issues/212#issuecomment-1187503882 is a great post that explains the setup | 20:08:10 |
| hexa | I don't think much has changed since it was potsed | 20:08:18 |
| hexa | * I don't think much has changed since it was posted | 20:08:25 |
| edef | note that S3 only speaks HTTP/1.1 | 20:14:13 |
| edef | are you sure you are measuring client requests to the CDN and not origin requests? | 20:14:36 |
| edef | because curling narinfos definitely gives me h2 | 20:15:18 |
| edef | and like, we might just be logging this wrong, this data may not exist, i am not sure | 20:19:18 |
| Arian | yeh im pretty sure these are client requests to the cdn | 20:19:38 |
| edef | the logs are literally some json we sprintf together | 20:19:46 |
