journey of implementing solr at target: presented by raja ramachandran, target
TRANSCRIPT
/
About Target
1,806 stores in the United States 133 stores in Canada Global locations in India and Canada
Source:h+ps://corporate.target.com/about/
/
About Target.com
3 years old*
4th most visited website in US retail 334 million visits per quarter
* target.com website has been opera8onal for over 10 yrs
/
“Search* handles 65% of traffic and drives nearly a billion dollars in revenue”
* Includes all grid pages – search+browse
/
Search - Architecture
Ra8ngs and Reviews
Popularity
Availability
Item
WCS Catalog ITL/Indexing
Business Configura8ons
Search mWeb
Apps
Target.com
Content Sources e-‐Commerce Pla1orm
Search Pla1orm Consumers
IBM
Oracle-‐Endeca
Index
/
• More suited to handle our scale of data
• Flexible relevancy control
• Active community
Why Solr?
/
• Flexible infrastructure and deployment options
• Ease of signal acquisition and relevancy
• Deliver a delightful experience to
our guests
Guiding Principles
/
Test drive
Target.com
Solr
Endeca
WCS Catalog
Same Catalog Extract
• Easy to install and run
• Low moving parts
• Simplified schema
Handling guest traffic in 2 sprints
/
Test drive 2 – To the cloud
typeahead
Endeca
Popular Search Terms
WAS
Solr Tomcat
Solr Tomcat
Current Solu8on
Public Cloud
/
Trends Sales Loca8on
Avail. Price Catalog
Context
Relevancy in eCommerce is no longer just textual. It is contextual.
Seasons
SOLR
/
Synchronized designer and product launches Step 1: Create two cores with the same named configset in zookeeper
catalog, catalog_shadow
Step 2: Set Collection Aliases catalog= CATALOG_LIVE catalog_shadow = CATALOG_SHADOW
Step 3: Import into CATALOG_SHADOW, validate, cache warm up scripts Step 4: Temporarily swap collection aliases
catalog_shadow = CATALOG_LIVE catalog = CATALOG_SHADOW
Step 5: Repeat import on CATALOG_SHADOW and validate
/
Atomic Updates
Collection Aliases
LIVE/SHADOW Cores
Foundational elements of an eCommerce search platform