Page MenuHomePhabricator

CirrusSearch: Investigate removing stored from most fields in Elasticsearch
Closed, ResolvedPublic

Description

Right now we tag most fields as stored in Elasticsearch but on further reading we should really only tag the fields that we need for prefix searching. The reason for this is that during full text search we actually would load most of the _source document any way when you measure by size by virtue of highlighting the text. Once you load most of the _source you may as well get everything from it.

In other words, setting a field to stored only speeds up queries that *don't* highlight or return the text but *do* highlight or return the stored field. It actually slows down things that *do* highlight the text because it involves multiple loads. Also, it takes up extra disk space.


Version: unspecified
Severity: normal

Details

Reference
bz58098

Event Timeline

bzimport raised the priority of this task from to Medium.Nov 22 2014, 2:25 AM
bzimport added a project: CirrusSearch.
bzimport set Reference to bz58098.

Change 100625 had a related patch set uploaded by Manybubbles:
Stop 'store'ing fields

https://gerrit.wikimedia.org/r/100625

Change 100625 merged by jenkins-bot:
Stop 'store'ing fields

https://gerrit.wikimedia.org/r/100625