https://bugzilla.wikimedia.org/show_bug.cgi?id=64171
ssas...@wikimedia.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |ssas...@wikimedia.org
--- Comment #9 from ssas...@wikimedia.org ---
I think the instrumentation in the WIP patch above should shed some light by
breaking the save-pipeline time into parsoid-serialization time + everything
else.
In addition to the instrumentation in the VE extension to measure Parsoid
serialization time, on the Parsoid side, we could instrument our serialization
pipeline into (a) dom-diff times (b) varnish hit/miss info (c) serialization
time.
In roundtrip testing, on slower servers, where there is no dom-diff, and there
is full serialization (unlike faster selective serialization in production), we
found html2wt times to be consistently fast (< 3s in almost all cases).
Unfortunately, those numbers are right now buried in our rt-server db which has
grown too big. Once the current round of rt-testing finishes, if I can access
those times and post some numbers here.
On my laptop, a full serialization of 3.8M Barack Obama page completed in ~4.2
s. And, dom-diff of a lightly-edited (~5-6 text edits) version of BO html
completed in ~2 s.
All that said, dom-diff could still be a bottleneck in some heavy-edit
scenarios -- we haven't really done any real perf tests on that component. If
this first-level breakdown of time to get Parsoid HTML reveals that we need
more breakdown of Parsoid numbers as above, we can prioritize that and add
instrumentation in Parsoid (or repurpose Ori's existing patch).
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
Wikibugs-l mailing list
Wikibugs-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikibugs-l
