[EPIC] Parsoid Performance
Open, Needs TriagePublic
Actions

Assigned To

None

Authored By

	cscott
	Apr 17 2025, 2:58 PM

Description

Work on performance issues potentially blocking Parsoid Read Views roll out to large wikipedias, including:

performance outliers where parsoid performance diverges greatly from legacy
OOM and other crashers
reconciling various limits (page size, max parsing time, transclusions, etc) so that Parsoid and legacy agree on which pages we should refuse to parse
ensuring DOS limits are in place and consistent between parsoid and legacy
adding metrics to track other issues that might affect us as we scale up

Related Objects
Search...

Status	Subtype	Assigned	Task
Open		None	T392227 [EPIC] Parsoid Performance
Resolved		Jgiannelos	T392079 Investigate open telemetry support for parsoid
Resolved		ssastry	T392260 Investigate performance outliers
Resolved		ssastry	T392261 Investigate crashers (out of memory, timeouts)
Open		None	T392262 Make limit reports match between legacy and Parsoid
Declined		None	T392263 Look at [[Special:LongPages]] and try to extract a performance benchmark
Resolved		ihurbain	T348255 Parser cache infrastructure for OutputTransform
Resolved		ihurbain	T348253 Create MediaWiki\ParserOutputTransform\DefaultOutputTransform
Resolved		cscott	T348254 Add ParserOutput::getContentHolder()
Resolved		ihurbain	T347062 Create ContentHolder interface
Resolved		cscott	T346829 JsonCodec should be a standalone library
Resolved		ihurbain	T404459 Use ContentHolder to speed up Parsoid-related OutputTransform stages
Open		None	T348853 Deprecate null skin on DefaultOutputTransform
Open		cscott	T380517 Make Parsoid language conversion into an OutputTransform pass
Open	BUG REPORT	None	T400140 Existed links render as redlinks in Template
Resolved		cscott	T415435 Add temporary URL request parameter to opt-in to the new Parsoid LanguageConverter implementation
Open		None	T267067 Make language variant a parser option
Open		None	T346996 Move page view language detection logic / functions out of LanguageConverter class and stop using global state
Resolved		Krinkle	T299369 Consider removing global $userLang from onPageContentLanguage hook
Resolved		Krinkle	T341244 ParserOptions and Title::getPageViewLanguage may disagree on the lang/dir
Resolved		DAlangi_WMF	T350806 Phase out Title::getPageViewLanguage in favour of ParserOutput metadata
Open		cscott	T416104 Use proper variant in ParsoidLocalization on messages localized into the page language
Open		ihurbain	T410916 Allow requesting postprocessed output for some/many/most/all APIs which accept a ParserOption (in some form) and emit a ParserOutput (in some form)
Resolved		ihurbain	T410919 Allow ParserOutputAccess to generate (and cache) postprocessed output for use in Article read views
Resolved		ihurbain	T410923 Clone ParserOutput in Article before post-processing
Resolved		ihurbain	T412803 Tweak unsafe post-processing cache keys
Resolved		cscott	T413227 Section edit links appear in wrong user interface language on Parsoid Read View wikis
Resolved		ihurbain	T414359 Use postprocessing cache in FlaggedRevisions
Open		None	T411401 Eliminate double-parse and double-postprocess in CommonsMetadata
Resolved	PRODUCTION ERROR	cscott	T412959 Logstash poolcounter warnings "Usage error: You may only aquire a single non-nowait lock" on wikis with post-processing cache enabled
Open		ihurbain	T413795 Use ParserOptions in the post-processing pipeline rather than the options array