[00:03:45] looks like an SSL issue...if I curl by wdqs1011's IPv4 addr from prom1006, I get an invalidate cert error [00:04:12] I guess we don't have the external domain on the internal servers' certs, which is fine [00:07:44] Anyway, I enabled puppet on the public hosts and left it off on the internal hosts...we can add some more puppet code to fix this tomorrow [00:08:16] also downtimed wdqs1011 [09:52:23] gehel: Hi! Can't make it 11, suggested a new time [10:55:12] lunch [11:01:15] pfischer: aci [11:01:18] Ack [14:23:28] o/ [16:36:14] hmm, will have to poke consumer logs. Had a couple pages that had delete events, but saneitizer found as not deleted. update topic had the deletes, running a backfill deleted the pages [17:03:19] Wednesday meeting has resumed in the original meeting [17:04:41] workout, back in ~40 [17:54:33] ebernhardson back [18:24:57] dinner [18:28:50] dr0ptp4kt ryankemper just added some mtg notes, feel free to add your own concerns here: https://docs.google.com/document/d/1yXifDY0499N2G-E3w84mIF5UR55cHjT3Ak2ZPVWDmMY/edit [18:28:58] Lunch, back in time for meeting [19:33:02] inflatador: dr0ptp4kt: omw back from pharmacy, 6m [19:33:16] ryankemper ACK, see you then [19:45:09] https://phabricator.wikimedia.org/T352253 inflatador [19:55:59] https://grafana.wikimedia.org/d/000000489/wikidata-query-service?orgId=1&refresh=1m&from=now-24h&to=now [20:14:40] hi! is it possible to downrank search results if a keyword only belongs to transcluded template content? e.g., a search for "foo" yields images 1 and 2. Image 1 has the word "foo" in the title. Image 2 has a template that transcludes the word "foo". I'd like image 1 to show up when searching for "foo" but not image 2. [20:15:37] (using `-hastemplate:{template that transcludes "foo"}` is probably not an option here, though that would solve the scenario I listed above) [20:36:35] kostajh: hmm, unfortunately in general we don't know what part of the content came from transclusions [20:37:19] ok [20:37:42] kostajh: the general solution would be identifying a css selector in the parser output that contains less useful content and add it to the auxiliary element selectors. That content essentially gets moved to a field with less importance than the main text [20:37:49] but it's not clear we could apply that to all transcluded content [20:38:05] usually it's things like thumbnail captions and tables [20:46:07] ack [21:15:05] dr0ptp4kt ah good, I was just going to ask about that [21:15:18] the CPU difference that is [21:26:17] quick break, back in ~20 [22:09:28] hmm, to know if a page needs to be deleted from the archive index we have to know if there are any remaining deleted revisions for this page :S [22:37:21] would almost need custom events...our needs are a bit odd. We use the page id as doc id, which also means we need the old page id and not the restored page id. And due to mediawiki being fun, an undelete can have multiple page_id's in the archive table :) [23:02:11] ryankemper looks like wdqs-internal still uses cleartext, based on hieradata/common/service.yaml . We added envoy in https://gerrit.wikimedia.org/r/c/operations/puppet/+/657913 but I guess we didn't update the plumbing [23:02:49] which explains at least part of why its new blackbox checks are failing ;) [23:32:25] ebernhardson do you happen to know what URL/port our internal wdqs consumers use? I can only seem to get it via cleartext port 80 . Just wondering if we need to notify anyone if/when we change it [23:36:08] hmm, [23:36:14] probably in the envoy config? looking [23:37:52] cirrus queries http://localhost:6009/bigdata/namespace/categories/sparql [23:38:10] per service.yaml in hiera, that goes to port 80 [23:39:18] i suppose with any hope, any other users are using the envoy proxy and will follow it's config- [23:42:13] Ah, thanks. I guess I don't understand the envoy magic [23:43:39] anyway, will mull this over tomorrow. thanks again for taking a look!