I seem to remember that all Wikimedia wikis now share a single search index, and per-wiki searches are filtered through a tag for the respective wiki.
If that is indeed the case, is there an API to search all wikis, but omitting that tag?
See also related bug: https://phabricator.wikimedia.org/T71489
On Fri, Sep 11, 2015 at 12:45 PM, Magnus Manske <magnusmanske@googlemail.com
wrote:
I seem to remember that all Wikimedia wikis now share a single search index, and per-wiki searches are filtered through a tag for the respective wiki.
If that is indeed the case, is there an API to search all wikis, but omitting that tag? _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Unfortunately there is not a single index that is then filtered by tags, in WMF production we currently have 1824 different search indexes. Due to these being all independent search indexes the scores generated by queries in one in index are not directly comparable to the scores generated by querying another index (we can't just naively merge the result set's together). By having these as separate indexes our busiest index (enwiki_content) is able to be kept down to only querying against 150GB of data, rather than querying the full cross-wiki data set which represents 2.5TB of data. This ends up being very important as we serve 1.5-3k queries per second against this index.
On Fri, Sep 11, 2015 at 4:25 AM, Eran Rosenthal eranroz89@gmail.com wrote:
See also related bug: https://phabricator.wikimedia.org/T71489
On Fri, Sep 11, 2015 at 12:45 PM, Magnus Manske < magnusmanske@googlemail.com
wrote:
I seem to remember that all Wikimedia wikis now share a single search index, and per-wiki searches are filtered through a tag for the
respective
wiki.
If that is indeed the case, is there an API to search all wikis, but omitting that tag? _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
I should add we are currently also looking into https://phabricator.wikimedia.org/T109715 which would make the full data set available for raw queries (directly to elasticsearch) to labs users. When issuing raw queries you can query every index in elasticsearch. its not performant at all, but it would work.
On Fri, Sep 11, 2015 at 7:56 AM, Erik Bernhardson < ebernhardson@wikimedia.org> wrote:
Unfortunately there is not a single index that is then filtered by tags, in WMF production we currently have 1824 different search indexes. Due to these being all independent search indexes the scores generated by queries in one in index are not directly comparable to the scores generated by querying another index (we can't just naively merge the result set's together). By having these as separate indexes our busiest index (enwiki_content) is able to be kept down to only querying against 150GB of data, rather than querying the full cross-wiki data set which represents 2.5TB of data. This ends up being very important as we serve 1.5-3k queries per second against this index.
On Fri, Sep 11, 2015 at 4:25 AM, Eran Rosenthal eranroz89@gmail.com wrote:
See also related bug: https://phabricator.wikimedia.org/T71489
On Fri, Sep 11, 2015 at 12:45 PM, Magnus Manske < magnusmanske@googlemail.com
wrote:
I seem to remember that all Wikimedia wikis now share a single search index, and per-wiki searches are filtered through a tag for the
respective
wiki.
If that is indeed the case, is there an API to search all wikis, but omitting that tag? _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
wikitech-l@lists.wikimedia.org