hoo added a comment.
In https://phabricator.wikimedia.org/T111769#2043332, @jcrespo wrote:
> Yesterday at ~19:00UTC there was an increase 2x in the number of UPDATES on
almost all wikis. I tracked this down to large amounts of `UPDATE /*
jcrespo added a comment.
Yesterday at ~19:00UTC there was an increase 2x in the number of UPDATES on
almost all wikis. I tracked this down to large amounts of `UPDATE /*
Wikibase\Client\Usage\Sql\EntityUsageTable::touchUsageBatch` queries. Is there
anything that could have caused that? What
daniel added a comment.
@jcrespo That describes the problem as "When updating slow special pages from
terbium, such as Listredirects, certain rows of the querycache table are
deleted and then inserted." It mentions that updating slow query pages is
problematic when "Other writes are happening
jcrespo added a comment.
No more meetings, please! :-) Someone hanging out in #wikimedia-operations or
https://phabricator.wikimedia.org/tag/dba/ would be enough so I can ping
someone about ongoing issues, suggestions, etc.
Certain awareness about the work being done both in reducing lag and
daniel added a comment.
@jcrespo ok, will do! Conversely, you can find us at
https://phabricator.wikimedia.org/tag/wikidata/ on IRC if there is any trouble.
TASK DETAIL
https://phabricator.wikimedia.org/T111769
EMAIL PREFERENCES
jcrespo added a comment.
I created https://phabricator.wikimedia.org/T122429 one month ago.
TASK DETAIL
https://phabricator.wikimedia.org/T111769
EMAIL PREFERENCES
https://phabricator.wikimedia.org/settings/panel/emailpreferences/
To: hoo, jcrespo
Cc: JanZerebecki, hoo, daniel, jcrespo,
daniel added a comment.
@jcrespo Can you write down somewhere what problems you see with wikidata
updates on the database level? Perhaps it would be useful to establish a
communication channel for this kind of thing. This is the first I hear about
our updates being a blocker for switching to
jcrespo added a comment.
As per @hoo request, I am sharing my own opinion here:
This is blocking ROW based replication application, which should be rolled in
soon. In my own optinion (database-focused) I think wikidata updates (this is
one of them) are the worse mediawiki-related production
hoo added a comment.
I just looked into this very briefly and I see two cases where we do queries as
listed above:
- On `LinksUpdateComplete` which should also be throttled to not overwhelm the
infrastructure
- In `AddUsagesForPageJob`… maybe we should throttle that job in order to
reduce
