Building a real time, big data analytics platform with solr

Trey Grainger
Manager, Search Technology Development
@
Building a Real-time, Big Data
Analytics Platform with Solr
Lucene Revolution 2013 - San Diego, CA

Overview
• Intro to Analytics with Solr
• Real-world examples & Faceting deep dive
• Solr enhancements we’re contributing:
– Distributed Pivot Faceting
– Pivoted Percentile/Stats Faceting
• Data Analytics with Solr… the next frontier

My Background
Trey Grainger
Manager, Search Technology Development
@ CareerBuilder.com
Relevant Background
• Search & Recommendations
• High-volume, Distributed Systems
• NLP, Relevancy Tuning, User Group Testing, & Machine Learning
Other Projects
• Co-author: Solr in Action
• Founder and Chief Engineer @ .com

About Search @CareerBuilder
• Over 2.5 million new jobs each month
• Over 50 million actively searchable resumes
• ~300 globally distributed search servers (in
the U.S., Europe, & Asia)
• Thousands of unique, dynamically generated
indexes
• Over ½ Billion actively searchable documents
• Over 1 million searches an hour

Our Search Platform
• Generic Search API wrapping Solr + our domain stack
• Goal: Abstract away search into a simple API so that any engineer
can build search-based product with no prior search background
• 3 Supported Methods (with rich syntax):
– AddDocument
– DeleteDocument
– Search
*users pass along their own dynamically-defined schemas on each call
• Building out as Restful API so search can be used “from anywhere”

Paradigm shift?
• Originally, search engines were designed to
return a ranked list of relevant documents

Paradigm shift?
• Then, features like faceting were added to augment
the search experience with aggregate information…

Paradigm shift?
• But… what kinds of products could you build if you
only cared about the aggregate calculations?

//Range Faceting
&facet.range=years_experience
&facet.range.start=0
&facet.range.end=10
&facet.range.gap=1
&facet.range.other=after
Faceting Overview
/solr/select/?q=…&facet=true
//Field Faceting
&facet.field=city
"facet_ranges":{
"years_experience":{
"counts":[
"0",1010035,
"1",343831,
…
"9",121090
], …
"after":59462}}
"facet_fields":{
"city":[
"new york, ny",2337,
"los angeles, ca",1693,
"chicago, il",1535,
… ]}
"facet_queries":{
"0 to 10 km":1187,
"10 to 25 km":462,
"25 to 50 km":794,
"50+":105296
},
//Query Faceting:
&facet.query={!frange key="0 to 10 km" l=0 u=10 incll=false}geodist()
&facet.query={!frange key="50+" l=50 incll=false}geodist()
&sfield=location
&pt=37.7770,-122.4200

Multi-select Faceting with Tags / Excludes
*Slide Taken from Ch 8 of Solr in Action
Faceting with No Filters Selected:
&facet=true
&facet.field=state
&facet.field=price_range
&facet.field=city
Faceting with Filter of state:California selected: &facet=true
&facet.field=state
&facet.field=city
&fq=state:California
Multi-Select Faceting with state:California selected:
&facet=true
&facet.field={!ex="stateTag"}state
&facet.field=city
&fq={!tag="stateTag"}state:California

Why Solr for Analytics?
• Allows “ad-hoc” querying of data by keywords
• Is good at on-the-fly aggregate calculations
(facets + stats + functions + grouping)
• Solr is horizontally scalable, and thus able to handle
billions of documents
• Insanely Fast queries, encouraging user exploration

Supply over Demand (Labor Pressure)

/solr/select/q=...&facet=true&facet.field=state
/solr/select/?q=…&facet=true&facet.field=month*
/solr/select/?q=…&facet=true&
facet.field=military_experience
Building Blocks…
*string field in format 201305

Building Blocks…
/solr/select/?
q="construction worker"&
fq=city:"las vegas, nv"&
facet=true&
facet.field=company
/solr/select/?
q="construction worker"&
fq=city:"las vegas, nv"&
facet=true&
facet.field=lastjobtitle

Building Blocks…
/solr/select/? q=...&
facet=true&facet.field=experience_ranges
/solr/select/?q=...&facet=true&
facet.field=management_experience

That’s pretty basic… what else can you do?

Query 1:
/solr/select/?...
fq={!geofilt sfield=latlong pt=37.777,-122.420 d=80}
&facet=true&facet.field=city&
"facet_fields":{
"city":[
"san francisco, ca",11713,
"san jose, ca",3071,
"oakland, ca",1482,
"palo alto, ca",1318,
"santa clara, ca",1212,
"mountain view, ca",1045,
"sunnyvale, ca",1004,
"fremont, ca",726,
"redwood city, ca",633,
"berkeley, ca",599]}Query 2:
/solr/select/?...
&facet=true&facet.field=city&
fq=( _query_:"{!geofilt sfield=latlong pt=37.7770,-122.4200 d=20} " //san francisco
OR _query_:"{!geofilt sfield=latlong pt=37.338,-121.886 d=20} " //san jose
…
OR _query_:"{!geofilt sfield=latlong pt=37.870,--122.271 d=20} " //berkeley
)
Geo-spatial Analytics

Customer’s Performance vs. Competition

A/B Testing
1. Hash all users based upon user identifier in
application stack
2. Flow user event streams into Solr with user
identifier
3. Implement custom hashing function in Solr
4. Facet based upon custom function
5. Profit!

//Custom Solr Plugin
public class ExperimentGroupFunctionParser extends ValueSourceParser {
public ValueSource parse(FunctionQParser fqp) throws SyntaxError{
String experimentName = fqp.parseArg();
ValueSource uniqueID = fqp.parseValueSource();
int numberOfGroups = fqp.parseInt();
double ratioPerGroup = fqp.parseDouble();
return new ExperimentGroupFunction (experimentName, uniqueID, numberOfGroups, ratioPerGroup);
}
}
public class ExperimentGroupFunction extends ValueSource {
…
@Override
public FunctionValues getValues(Map context, AtomicReaderContext readerContext) throws IOException {
final FunctionValues vals = uniqueIdFieldValueSource.getValues(context, readerContext);
return new IntDocValues(this) {
@Override
public int intVal(int doc) {
//returns an deterministically hashed integer indicating which test group a user is in
return GetExperimentGroup(experimentName, vals.strVal(doc), numGroups, ratioPerGroup);
}
}
…
}
//SolrConfig.xml:
<valueSourceParser name=”experimentgroup"
class="com.careerbuilder.solr.functions. ExperimentGroupFunctionParser" />
A/B Testing – Adding Function to Solr

//Group 1 Time Series Facet Query:
/solr/select?
facet.range=eventDate&
facet.range.start=2013-04-01T01:00:00.000Z&
facet.range.end=2013-04-07T01:00:00.000Z&
facet.range.gap=+1Hour&
fq={!frange l=1 u=1 key="Group 1"}experimentgroup("EXPERIMENT_NAME", userIdField, 4, 0.25)
//repeat for groups 2-4
…fq={!frange l=2 u=2 key="Group 2"}experimentgroup("EXPERIMENT_NAME", userIdField, 4, 0.25)
…fq={!frange l=3 u=3 key="Group 3"}experimentgroup("EXPERIMENT_NAME", userIdField, 4,
0.25)…fq={!frange l=4 u=4 key="Group 4"}experimentgroup("EXPERIMENT_NAME", userIdField, 4,
0.25)
A/B Testing – Faceting on Function

SOLR-2894: “Distributed Pivot Faceting”
Status: We have submitted a stable patch (including distributed
refinement) which is working in production at CareerBuilder.
Note: The community has reported Issues with non-text fields (i.e.
dates) which need to be resolved before the patch is committed.

SOLR-3583: “Stats within (pivot) facets”
Status: We have submitted an early patch (built on top of SOLR-2894,
distributed pivot facets), which is in production at CareerBuilder.

/solr/select?q=...&
facet=true&
facet.pivot=state,city&
facet.stats.percentiles=true&
facet.stats.percentiles.averages=true&
facet.stats.percentiles.field=compensation&
f.compensation.stats.percentiles.requested=10,25,50,75,90&
f.compensation.stats.percentiles.lower.fence=1000&
f.compensation.stats.percentiles.upper.fence=200000&
f.compensation.stats.percentiles.gap=1000
"facet_pivot":{
"state,city":[{
"field":"state",
"value":"california",
"count":1872280,
"statistics":[
"compensation",[
"percentiles",[
"10.0","26000.0",
"25.0","31000.0",
"50.0","43000.0",
"75.0","66000.0",
"90.0","94000.0"],
"percentiles_average",52613.72,
"percentiles_count",1514592]],
"pivot":[{
"field":"city",
"value":"los angeles, ca",
"count":134851,
"statistics":{
"compensation":[
"percentiles",[
"10.0","26000.0",
"25.0","31000.0",
"50.0","45000.0",
"75.0","70000.0",
"90.0","95000.0"],
"percentiles_average",54122.45,
"percentiles_count",213481]}}
…
]}]}
SOLR-3583: “Stats within (pivot) facets”

Real-world Use Case
Stats Pivot Faceting (Percentiles)Stats Pivot
Faceting (Average)
Field
Facet

Data Analytics with Solr… the next frontier…

Solr Analytics Wish List…
• Faceting Architecture Redesign:
– “All” Facets should be pivot facets
(default pivot depth = 1)
– Each “Pivot” could be a field, query, or range
– Meta information (like the Stats patch) could be nested in each pivot
– Backwards compatibility with current facet response format for a while…
• Grouping
– Grouping should also support multiple levels (Pivot Grouping)
– If “Pivot Grouping” were supported… what about faceting within each of the pivot
groups?
• Remaining caches should be converted to “per-segment” to enable NRT
• These kinds of changes would enable Solr to return rich Data Analytics
calculations in a single search call where many calls are required today.

Recap
Billions of documents
+
Ad-hoc querying by keyword
+
Horizontal Scalability
+
Sub-second query responses
+
Facet on “anything” – field, function, etc.
+
User friendly visualizations/UX
--------------------------------------------------------
Building a real-time, Big Data Analytics Platform with Solr

Contact Info
Yes, we are hiring @CareerBuilder. Come talk with me if you are interested…
 Trey Grainger
trey.grainger@careerbuilder.com
@treygrainger
Other presentations:
http://www.treygrainger.com http://solrinaction.com

Building a real time, big data analytics platform with solr

More Related Content

Similar to Building a real time, big data analytics platform with solr (20)

More from lucenerevolution (20)

Recently uploaded (20)

Building a real time, big data analytics platform with solr