docs/page_lifecycle.txt
A page from MediaWiki, indexed into CirrusSearch, has a lifecycle of birth
and death. This only covers the traditional job-queue based page updates,
some installations use cirrus-streaming-updater to perform updates based on
event streams rather than embedded into the job queue.
The steps are basically:
=== New page is created ===
* After some core handling of the page edit via the job queue the
LinksUpdateComplete hook fires. This inserts a CirrusSearch\Job\LinksUpdate job.
* The LinksUpdate job follows redirects from edited page to the final target,
generates a full document for indexing, and writes that document to
Elasticsearch. Additionally calls OtherIndex which dupes some file
information to a shared file index, and inserts CirrusSearch\Job\IncomingLinkCount
jobs for each link added/removed in this revision
=== Existing page is edited ===
* Same code path as new page
=== Page is imported ===
* Same code path as new page
=== Page is moved ===
* TitleMoveComplete hook fires, if page moved between index types (content, general)
the page is deleted from the old index and the same code path as new page is run
to index at the new location.
=== Revision of page is deleted ===
* RevisionDelete hook fires, inserts CirrusSearch\Job\LinksUpdate
* Technically nothing needs to happen, "current" revision which we index cannot
be deleted. Indexes anyways just in case to be sure we have that current version
and not the older version being deleted.
=== Existing page is deleted ===
* ArticleDelete hook fires, inserts CirrusSearch\Job\LinksUpdate only if the
page is a redirect. This reindexes the target to remove the redirect.
* ArticleDeleteComplete hook fires, inserts CirrusSearch\Job\DeletePages
* Issue delete-by-id to page type of appropriate index
* Send archive docs to general index archive type
* ???
* ???
=== Page is un-deleted ===
* ArticleUndelete hook fires, inserts CirrusSearch\Job\DeleteArchive
* Delete documents from archive type under some (?) constraints
** A single DeleteArchive job works with a single Title
** If that title has archived rows ????
* Article itself is indexed via standard LinksUpdateComplete hook