Using HBase Coprocessors to implement Prospective Search - Berlin Buzzwords - June 2012
25
CC 2.0 by William Brawley | http://flic.kr/p/7PdUP3
-
Upload
christian-guegi -
Category
Technology
-
view
2.830 -
download
0
description
With the HBase 0.92 release it is now possible to embed code directly into HBase server processes. Coprocessors offer a very flexible extension model to move functionality closer to the data.This presentation explains the key concepts and shows how coprocessors have been used in our social media monitoring system to accomplish the alert feature implemented with prospective search. It allows users to formulate a query by which they will be instantly notified via email about every newly detected document matching the terms.This session aims to help people interested in coprocessors understand how coprocessors can be adapted to their own use cases and how to develop coprocessor extensions.
Transcript of Using HBase Coprocessors to implement Prospective Search - Berlin Buzzwords - June 2012
CC 2.0 by William Brawley | http://flic.kr/p/7PdUP3
2
Agenda
• Why Hadoop and HBase? • Social Media Monitoring • Prospective Search and Coprocessors
• Challenges & Lessons Learned • Resources to get started
5. Juni 2012
3
About me
Software Architect @ sentric
Co-founder and organizer of the Swiss HUG
Contact: [email protected] http://www.sentric.ch @chrisgugi
5. Juni 2012
4
About sentric
• Spin-off of MeMo News AG, the leading provider for Social Media Monitoring & Analytics in Switzerland
• Big Data expert, focused on Hadoop, HBase and Solr
• Objective: Transforming data into insights
5. Juni 2012
CC 2.0 by Pete Reed | h"p://flic.kr/p/KS9kf
6
Social Media Monitoring Process
Why Hadoop and HBase?
5. Juni 2012
Information Gathering
Information Processing
Analysis & Interpretation
Insight Presentation
7
Requirements
Why Hadoop and HBase?
5. Juni 2012
SMM
Cost effective
High scalable
RT Alerting Analytical capabilities
Reliable
8
Technology Stack
Why Hadoop and HBase?
5. Juni 2012
HBase /HDFS Storage
Hadoop Mahout Analytics
Solr Search
HBase RowLog Event mechanism (MQ)
Prospective search Real-time alerting
CC 2.0 by nolifebeforecoffee | http://flic.kr/p/c1UTf
10
Overview
Social Media Monitoring
5. Juni 2012
Search Agents
Downloaded Articles
Output
match?
RT Alerts Reports Web-UI
Icons by http://dryicons.com
11
Solution Architecture
Social Media Monitoring
5. Juni 2012
REST
n Crawler
MySQL Solr
Web-UI
RT Alerts
RowLog Coprocessor
HBase
Icons by http://dryicons.com
12
Overview
• Inspired by Google Bigtable coprocessors
• HBase version 0.92 • Embed code directly into server
processes • High-level call interface for clients • Automatic scaling, load balancing,
request routing
Short Primer on Coprocessors
5. Juni 2012
13
Observer Classes
• Like a database trigger • Provides event based hooks
• Concrete Implementations • RegionObserver • CRUD or DML type operations
• MasterObserver • DDL or metadata operations and cluster
administration
• WALObserver • Write-ahead-log appending and restoration
Short Primer on Coprocessors
5. Juni 2012
14
Observer Execution
Short Primer on Coprocessors
5. Juni 2012
Client:Get()
CP1:preGet() CP2:preGet() CP3:preGet()
Hregion:Get()
CP1:postGet() CP2:postGet() CP3:postGet()
RegionServer
client response
15
Endpoint Classes
• Comparable to stored procedures • Custom RPC protocol, used between
client and region server
• Loaded in region server • Client call APIs over single row or a
row range • Framework translates row keys to region
location • Parallel execution
Short Primer on Coprocessors
5. Juni 2012
16
Endpoint Call Routine
Short Primer on Coprocessors
5. Juni 2012
Client code
Batch.Call<CountProtocol,int>
int call(CountProtocol p) { return p.getRowCount();
} .
Map<byte[], Integer> countsByRegion
HTable
coprocessorExec()
CountProtocol
CountProtocol
CountProtocol
CountProtocol
Region Server 1
table,,12345678
table,bbb,12345678
Region Server 2
table,ccc,12345678
table,ddd,12345678
17
Use Cases
• HBase Security (Version 0.94) • Aggregate operations avg(), sum() • AggregatorProtocol
• HBASE-3529: Embedded search
Short Primer on Coprocessors
5. Juni 2012
18
Prospective Search with Coprocessors
Social Media Monitoring
5. Juni 2012
Processing
HRegionServer
HRegion
Put operations
Prospective Search
RT Alerts
Icons by http://dryicons.com
19
Testing Setup
• Standard, virtualized test cluster: 4RS/DN, 1HM, 1NN, 3ZK
• Test dataset created from 2h of live index (1GB)
• Drive load on RS/DN
Social Media Monitoring
5. Juni 2012
20
Test Results
Social Media Monitoring
5. Juni 2012
0
200
400
600
800
1000
1200
1400
1600
1800
0 10 50 100 200 400 800
Wri
tes/
sec
# of agents
CC 2.0 by Sean Maurik | h"p://flic.kr/p/JUduu
22
Challenges
• Everyone is still learning • Some issues only appear at scale • Production cluster configuration • Hardware issues • Tuning cluster configuration to our work
loads
• HBase stability • Monitoring health of HBase
Challenges & Lessons Learned
5. Juni 2012
23
Lessons
• Be careful with expensive operations in coprocessors
• At scale, nothing works as advertised • Monitoring/Operational tooling is
most important • Play with all the configurations and
benchmark for tuning
Challenges & Lessons Learned
5. Juni 2012
24
Resources to get started
• https://blogs.apache.org/hbase/entry/coprocessor_introduction
• http://hbase.apache.org/apidocs/index.html
• http://www.lilyproject.org/lily/about/playground/hbaserowlog.html
• http://www.github.com/sentric/HBasePS
5. Juni 2012
