[05:50:30] cdanis: that ripe link is neat! [08:50:00] moritzm: jbond I have a puppet question :), we want to be able to use some information that resides in a service, in the manifests themselves. Currently we are manually creating hiera entries for it with the data, but I was thinking if using some custom fact, or something similar would be better. What do you recommend? [08:53:38] dcaro: sounds like a custom fact would be a good option but if you have a task or more info i can take a better look [08:57:52] jbond: in more detail, we create auth keys for ceph, we have a resource that creates them on the server side, but then you have to get the key data, and create a file on the client side. [08:58:48] the key data is stored from ceph itself [08:58:58] (a server with admin access can run a command to get the data) [12:30:35] XioNoX: yeah I want to play with the data more -- same technique could be interesting for anycast tweaking [12:33:27] dcaro: Pardon my joining in. This sounds like this might be a good use case for using exported resources: https://puppet.com/docs/puppet/5.5/lang_exported.html [12:35:44] btullis: no prob, yes, we were thinking about something like that, but given that the info is in the host, facts coupled to that seemed like a good idea, though we went instead for handcoding in hiera the values (given that they are also secrets) [12:35:53] cdanis: for the anycast stuff having also the chaos/nsid mappings is also really nice. see https://atlas.ripe.net/results/maps/root-instances/ for an example [12:36:18] yeah! [12:44:35] dcaro: Yes, I see. It just seems that if you already have a resource that creates them on the server side, you could just turn this into an exported resource and then use an exported resource collector whenever you need to write the key out. It would save duplication of copying things into hiera. [13:01:57] <_joe_> btullis: exported resources can't gather information from the agent, they happen on the master at compile time [13:02:16] <_joe_> so i'm not sure they'd work in this case [13:02:34] <_joe_> oh you mean to collect the fact [13:08:41] _joe_: regarding static, php files don't need to be served and there's a bug about them executing. Any other file types? Or do all non image files fail? [13:09:13] <_joe_> Krinkle: I only tested that and files without an extension [13:09:52] <_joe_> >>> StreamFile::contentTypeFromPath('README', false); [13:09:54] <_joe_> => null [13:15:53] ok, I don't think either of those are needed right now. Note that unlike /w/, static/current aren't used for programmatic or arbitrary access. Only explicit references in config. Opt-in essentially. But yeah, we could supper extension-less files I suppose, but realistically the only way those would be used if it all is via /w/. [13:16:14] Looks like all other file types work, the same way as /w/. [13:17:01] And when we opt in, we opt in by optimising from the default access method, which is /w/. [13:18:13] Afaik use of StreamFile isn't specific to static/current so any wild file types wouldn't work in the default manner either [13:18:21] <_joe_> yep [13:19:32] <_joe_> I'm just looking at things we were supporting in our http tests, which were just getting the status quo [13:56:59] godog: is there precedent for importing history from graphite to prom or thanos in some way? [13:58:15] Krinkle: not at wmf but my understanding is that nowadays it is possible to backfill prometheus/thanos [13:59:18] godog: ok, and the global aggregated prom is gone or deprecated in favour of thanos? [13:59:40] godog: oh really? historically prom has very much _not_ supported that [13:59:59] Krinkle: the latter yeah, https://phabricator.wikimedia.org/T288196 [14:00:02] I'm looking for what the story is for metrics where we want 5y retention [14:00:15] ack, meeting [14:00:24] godog: and indeed, it looks like a very recent addition: https://medium.com/tlvince/prometheus-backfilling-a92573eb712c [14:00:31] kormat: yeah! seems like it, you can plop prometheus blocks [14:52:32] kormat Krinkle we'd have to test thanos backfilling but in theory should work the same as prometheus