Pathological test case for Parsoid wt->html
Open, HighPublic
Actions

Assigned To

None

Authored By

	ssastry
	Oct 23 2019, 4:06 PM

Description

On this page with about ~37K links ( https://ce.wikipedia.org/wiki/Декъашхо%3ATakhirgeran_Umar%2FНах_беха_меттигаш ), when you switch off the resource limit constraints, curl of this page on scandium times out with Parsoid/PHP whereas Parsoid/JS finishes in 24 seconds.

Flamegraph locally with parse.php shows WTUtils::reinsertFosterableContent and ProcessTreeBuilderFixups::deleteShadowMeta being the functions that take the most time.

Like some O(n^2) behavior.

Related Objects

Mentioned In: T239806: Parsoid/PHP errors

Event Timeline

ssastry triaged this task as Medium priority.Oct 23 2019, 4:06 PM

ssastry created this task.

Restricted Application added a subscriber: Aklapper. · View Herald TranscriptOct 23 2019, 4:06 PM

Looks like both these functions have a $node->parentNode->replaceChild(...) and in this case, the parent node has 37K children ... so, this is what seems to be causing the blowup.

ssastry moved this task from Backlog to Performance on the Parsoid-PHP board.Oct 23 2019, 5:38 PM

ssastry raised the priority of this task from Medium to High.Oct 25 2019, 2:14 PM

Does this mean that, unlike Domino, libxml rebuilds the children data-structure for replaceChild?

ssastry mentioned this in T239806: Parsoid/PHP errors.Dec 4 2019, 4:59 PM

Aklapper edited projects, added Parsoid; removed Parsoid-PHP.Apr 10 2020, 4:27 PM

ssastry moved this task from Needs Triage to Performance on the Parsoid board.Apr 10 2020, 4:50 PM

Pathological test case for Parsoid wt->htmlOpen, HighPublicActions

Description

Related Objects

Event Timeline

Pathological test case for Parsoid wt->html
Open, HighPublic
Actions