Page MenuHomePhabricator

Searches fail on en.wp: "An error has occurred while searching: The search backend returned an error:"
Closed, DeclinedPublic

Description

Searches at enWP fail, returning the error

An error has occurred while searching: The search backend returned an error:

action has been occurring > 10 minutes, multiple attempts. Via IRC, another in #wikimedia-tech another can confirm the issue


Version: wmf-deployment
Severity: normal
Whiteboard: cirrus-fixed

Details

Reference
bz56936

Event Timeline

bzimport raised the priority of this task from to Low.Nov 22 2014, 2:31 AM
bzimport set Reference to bz56936.

Still not for me. Pinging enWP tells me that I am using 198.35.26.96 so is it something related to the West Coast servers?

European ESAMS here (91.198.174.192) and no issues, of course. :-/

Source code would provide info such as

<!-- Served by mw1061 in 0.553 secs. -->

but that won't help much here.

If that's still an issue, maybe somebody in #wikimedia-ops might also have an idea.

Working for me even when I put
198.35.26.96 en.wikipedia.org
in /etc/hosts

I'll check some logs.

Served by mw1164 in 6.653 secs

Didn't know that there was commentary in html comment

search term: history of the united states flag

still getting it

Hmm, it seems to be choking with the "of the" in the search. Now if I trim it, it returns something.
<!-- Served by mw1170 in 2.623 secs. -->

When I forward my requests directly to mw1164 I still don't get it (with or without "of the".) I confirmed the forwarding using the html comment.

This log should appear twice: once in mwsearch's debug log and once in the fatals log. It looks like mwsearch's debug log has been turned off and I can't find this particular error in the fatals section. Someone more adept and error scraping will have to take over from here.

Changing priority to 'High' since search engine not working is quite important :-) I would've changed it to 'highest', but I have no idea how many people are affected by it.

@nik: Any possibility you can assign that more adept person to the bug, or have you already informed them using a different medium? It's generally nice to know people are actively working on resolving this kind of bugs, thanks :-)

We've finally got the logs cranked up enough that we should be able to really see these failures. If anyone watching this bug has the magic touch to reproduce the problem, then please do so now.

Also, assigning to myself because I'm working on it any way.

For posterity the errors that lucenesearch logs for (pretty frequently) without an explanation look like they are caused by queries almost matching a url but lucene search not doing the right thing with them:

w:.@.
.:80

It seems to hate :s.

Searching for a single "." causes the "empty search" error.

I can replicate the error and it almost seems like a cached query error, as I can change the order in the query from "of the" to be "the of" and the second query works, and the first will fail. It is only the permutation of "history of the united states flag" that is failing.

Last tests run at 0955 GMT

As I cannot get this to fail in any other combination, I am moving it all back to normal and normal.

https://en.wikipedia.org/w/index.php?title=Special:Search&search=small%20a&fulltext=Search&profile=help&redirs=1 may be related (from OlEnglish on IRC).

As far as I can tell, this is Lucene-specific and will be fixed when new search becomes default. The short-term workaround is probably to enable new search for your account using Special:BetaFeatures, if it's available on your wiki.

Got this today when searching for "(-)" - https://en.wikipedia.org/w/index.php?title=Special:Search&search=(-)&fulltext=Search&profile=help&redirs=1

No error when switching to the new search feature in beta.

Known bug in MWSearch/lsearchd, not a problem in Cirrus/MWSearch. Moving and marking WONTFIX as appropriate.