[{reqId}] {exception_url} Wikimedia\Assert\PreconditionException: Precondition failed: This Title instance does not represent a proper page, but merely a link target.| Title | Reference | Author | Source Branch | Dest Branch | |
|---|---|---|---|---|---|
| cirrusdoc: Handle unrenderable documents | repos/search-platform/cirrus-streaming-updater!180 | ebernhardson | work/ebernhardson/unrenderable | main | |
| sanity: Accept redirect-in-index without a link target | repos/search-platform/cirrus-streaming-updater!179 | ebernhardson | work/ebernhardson/sanity-unindexable-redirect | main |
Change #1127158 had a related patch set uploaded (by Ebernhardson; author: Ebernhardson):
[mediawiki/extensions/CirrusSearch@master] sanity: redirect-in-index should skip unindexable link targets
ebernhardson opened https://gitlab.wikimedia.org/repos/search-platform/cirrus-streaming-updater/-/merge_requests/179
sanity: Accept redirect-in-index without a link target
ebernhardson merged https://gitlab.wikimedia.org/repos/search-platform/cirrus-streaming-updater/-/merge_requests/179
sanity: Accept redirect-in-index without a link target
Change #1127595 had a related patch set uploaded (by Ebernhardson; author: Ebernhardson):
[operations/deployment-charts@master] cirrus: Update container image
Change #1127595 merged by jenkins-bot:
[operations/deployment-charts@master] cirrus: Update container image
As for how these problems got into the search index in the first place, I think it's via the saneitizer.
If we have a redirect:
> MediaWiki\Title\Title::newFromText('UDP Hole Punching')->isRedirect()
= true
> MediaWiki\Title\Title::newFromText('UDP Hole Punching')->getArticleID()
= 2632384And we ask cirrus-check-sanity if everything is correct about the way it's indexed it gets reported as an oldDocument which triggers a rerender:
https://en.wikipedia.org/w/api.php?action=cirrus-check-sanity&format=json&from=2632384&limit=1&sequenceid=0&rerenderfrequency=2&formatversion=2
{
"wikiId": "enwiki",
"clusterGroup": "chi",
"problems": [
{
"indexName": "enwiki_content",
"errorType": "oldDocument",
"pageId": 2632384,
"namespaceId": 0
}
]
}That rerender results in a document to be indexed, even though redirects shouldn't be indexed directly:
https://en.wikipedia.org/w/api.php?action=query&format=json&prop=cirrusbuilddoc&pageids=2632384&formatversion=2
{
"batchcomplete": true,
"query": {
"pages": [
{
"pageid": 2632384,
"ns": 0,
"title": "UDP Hole Punching",
"cirrusbuilddoc": {
...To resolve I'm putting together a few patches:
The main part I still don't understand is how we have so few redirects indexed. With the saneitizer on a 1:8 oldDocument, i would expect 1/8 of all redirects in all wikis to be in the index, and thus triggering redirectInIndex errors the following loop. But according to Saneitizer Bi-Weekly Fix Rate by Problem Type we only see 2k-3k of this problem type per loop.
Change #1127974 had a related patch set uploaded (by Ebernhardson; author: Ebernhardson):
[mediawiki/extensions/CirrusSearch@master] Don't build a document for redirects
ebernhardson opened https://gitlab.wikimedia.org/repos/search-platform/cirrus-streaming-updater/-/merge_requests/180
cirrusdoc: Handle unrenderable documents
Change #1127158 merged by jenkins-bot:
[mediawiki/extensions/CirrusSearch@master] sanity: redirect-in-index should skip unindexable link targets
ebernhardson merged https://gitlab.wikimedia.org/repos/search-platform/cirrus-streaming-updater/-/merge_requests/180
cirrusdoc: Handle unrenderable documents
Change #1128505 had a related patch set uploaded (by Ebernhardson; author: Ebernhardson):
[operations/deployment-charts@master] cirrus: Update container image
Change #1128505 merged by jenkins-bot:
[operations/deployment-charts@master] cirrus: Update container image
Change #1127974 merged by jenkins-bot:
[mediawiki/extensions/CirrusSearch@master] Don't build a document for redirects