Apache HBase

For Architects
Nick Dimiduk Member of Technical Staff, HBase Strata/Hadoop World, 2013-10-29

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 1

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 2

Agenda
! Background
! (how did we get here?)

! TL;DR
! (dont waste my time!)

! High-level Architecture
! (where are we?)

! Anatomy of a RegionServer
! (how does this thing work?)

! By Example
! (how do I use it?)

! Resources
! (where do we go from here?)
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 3

Background

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 4

So what is HBase anyway?

! BigTable paper from Google, 2006, Dean et al.
! Bigtable is a sparse, distributed, persistent multi-dimensional sorted map. ! http://research.google.com/archive/bigtable.html

! Key Features:
! Distributed storage across cluster of machines ! Random, online read and write data access ! Schemaless data model (NoSQL) ! Self-managed data partitions

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 5

Apache Hadoop Dependencies

! Apache Hadoop Distributed Filesystem (HDFS)
! Distributed, fault-tolerant, throughput-optimized data storage ! The Google File System, 2003, Ghemawat et al. ! http://research.google.com/archive/gfs.html

! Apache Zookeeper (ZK)

! Distributed, available, reliable coordination system ! The Chubby Lock Service !, 2006, Burrows ! http://research.google.com/archive/chubby.html

! Apache Hadoop MapReduce (MR)

! Distributed, fault-tolerant, batch-oriented data processing ! MapReduce: !, 2004, Dean and Ghemawat ! http://research.google.com/archive/mapreduce.html
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 6

TL;DR

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 7

So what is HBase anyway?

C1 tree C0 tree

Disk

Memory

Figure 2.1. Schematic picture of an LSM-tree of two components

Figure 2.1 reproduced from ONeil, Patrick, et al. "The log-structured merge-tree (LSM-tree)." Acta Informatica 33.4 (1996): 351-385.
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 8

So what is HBase anyway?

C1 tree C0 tree

DataNode

RegionServer

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk

Memory

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 9

cessing is also cessing a DataNode part is also of cessing the a DataNode part workloads, is also of the a part workloads, TaskTrackers of the w T DataNode RegionServer RegionServer DataNode RegionServer RegionServe DataNode So what is HBase anyway? Servers canServers run together. canServers run together. can run together.
C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

DataNode
Disk

Memory

RegionServer DataNode
Disk

Memory

RegionServer DataNode DataNode RegionServer RegionServe DataNode

Disk Memory

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Figure 3.7

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase DataNode and RegionServer HDFS processes DataNo and
Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk Memory Disk Memory Disk Memory

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 10

So what is HBase anyway?

C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree

Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro Figure 3.6 that AFigure table consists 3.6 AFigure table ofthe multiple consists 3.6 A smaller Figure table ofstored multiple consists 3.6 chunks A smaller Figure table called ofstored multiple consists 3.6 chunks regions. A smaller Figure table called ofstored multiple consists 3.6 chunks regions. A smaller table called ofstored multiple consists chunks regions. smaller called ofstored multiple chunks regions. smaller called chunks regions. called Given the Given underlying that Given underlying data that is the Given underlying data that in is the Given underlying data that in is the Given underlying data that in is the underlying data in is data in is stored in HDFS , which HDFS is available , which HDFS is to available all , which clients HDFS is to as available all , which clients HDFS is to as available all , which clients HDF is to a a ers can theoretically ers a can theoretically ers and can write theoretically ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing part isread also of cessing the a part workloads, isread also of cessing the a part workloads, TaskTrackers, is also of cessing the a part workloads, TaskTrackers, is DataNodes, also of cessing the a part workloads, TaskTrackers, is DataNodes, and also of the athe part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB a single namespace, a single namespace, all aRegionServers single namespace, all aRegionServers single have namespace, all access aRegionServers single have to the namespace, all access same aRegionServers single have to persisted the namespace, all access same RegionServers files have to persisted the in all access the same RegionServers file files have to persisted the in access the same file files have to persisted in access the same fil fil t You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. inthe this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. store/access store/access data on HDFS store/access data . The on master HDFS store/access data . process The on master HDFS store/access does data . process The on the master HDFS distribution store/access does data . process The on the master HDFS distribution of does data regions . process The on the master HDFS distribution among of does regions . process The the master distribution among of does regions process the distrib amon of does reg system and system can therefore and system can host therefore and any system can region host therefore and any (figure system can region host therefore 3.8). and any (figure By system can region physically host therefore 3.8). and any (figure By can collocating region physically host therefore 3.8). any (figure By collocating region Dataphysically host 3.8). any (figure By collocating region Dataphysically 3.8). (figure By collo Data phy 3 HBase deployments, HB ase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the if MapReduce the framework isnt workload atdeployed all the if MapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i RegionServers, RegionServers, and each RegionServers, RegionServer and each RegionServers, RegionServer and typically each RegionServers, RegionServer hosts and typically multiple each RegionServers, RegionServer hosts and typically regions. multiple each RegionServer hosts and typically regions. multiple each RegionServer hosts typically regions. multiple hosts typically regions. multiple ho Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro Given that the Given underlying that the Given underlying data that is stored the Given underlying data that in is stored the Given underlying data that in is stored the Given underlying data that in is stored the underlying data in is stored data in is stored in HDFS , which HDFS is available , which HDFS is to available all , which clients HDFS is to as available all , which clients HDFS is to as available all , which clients HDF is to a a ers can theoretically ers RegionServer can theoretically ers and can write theoretically ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and the DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing a part isread also of cessing the a part workloads, isread also of cessing the a part workloads, TaskTrackers, is also of cessing the a part workloads, TaskTrackers, is DataNodes, also of cessing the a part workloads, TaskTrackers, is DataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re a single namespace, a single namespace, all aRegionServers single namespace, all aRegionServers single have namespace, all access aRegionServers single have to the namespace, all access same aRegionServers single have to persisted the namespace, all access same RegionServers files have to persisted the in all access the same RegionServers file files have to persisted the in access the same file files have to persisted in access the same fil fil t You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. inthe this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. system and system can therefore and system can host therefore and any system can region host therefore and any (figure system can region host therefore 3.8). and any (figure By system can region physically host therefore 3.8). and any (figure By can collocating region physically host therefore 3.8). any (figure By collocating region Dataphysically host 3.8). any (figure By collocating region Dataphysically 3.8). (figure By collo Data phy 3 HBase deployments, HB ase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the if MapReduce the framework isnt workload atdeployed all the if MapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro ers can theoretically ers can theoretically read ers and can write theoretically read ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and the DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing a DataNode part is also of cessing the a DataNode part workloads, is also of cessing the a DataNode part workloads, TaskTrackers, isRegionServer also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. in this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. HBase deployments, HBase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the ifMapReduce the framework isnt workload atdeployed all the ifMapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro Figure 3.7 HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on cessing is also cessing a DataNode part is also of cessing the a DataNode part workloads, is also of cessing the a DataNode part workloads, TaskTrackers, isRegionServer also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re Servers canServers run together. canServers run together. canServers run together. canServers run together. canServers run together. can run together.
C1 tree C0 tree Disk C1 tree Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Figure 3.7

DataNode
Disk

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on
Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Memory

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

RegionServer DataNode
Disk

Memory

RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re
Disk Memory Disk Memory Disk Memory Disk Memory

C1 tree

Licensed to Licensed Nick Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Nick <ndimiduk@gmail.com Dimiduk <ndimiduk@
C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Memory C0 tree

Disk C1 tree

Figure 3.7

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on
Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory

Disk C1 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Licensed to Licensed Nick Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Nick <ndimiduk@gmail.com Dimiduk <ndimiduk@
Figure 3.7
Attribution-ShareAlike 3.0 Unported License. Page 11

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on Licensed under a Creative Commons

So what is HBase anyway?

C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree

Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro Figure 3.6 that AFigure table consists 3.6 AFigure table ofthe multiple consists 3.6 A smaller Figure table ofstored multiple consists 3.6 chunks A smaller Figure table called ofstored multiple consists 3.6 chunks regions. A smaller Figure table called ofstored multiple consists 3.6 chunks regions. A smaller table called ofstored multiple consists chunks regions. smaller called ofstored multiple chunks regions. smaller called chunks regions. called Given the Given underlying that Given underlying data that is the Given underlying data that in is the Given underlying data that in is the Given underlying data that in is the underlying data in is data in is stored in HDFS , which HDFS is available , which HDFS is to available all , which clients HDFS is to as available all , which clients HDFS is to as available all , which clients HDF is to a a ers can theoretically ers a can theoretically ers and can write theoretically ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing part isread also of cessing the a part workloads, isread also of cessing the a part workloads, TaskTrackers, is also of cessing the a part workloads, TaskTrackers, is DataNodes, also of cessing the a part workloads, TaskTrackers, is DataNodes, and also of the athe part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB a single namespace, a single namespace, all aRegionServers single namespace, all aRegionServers single have namespace, all access aRegionServers single have to the namespace, all access same aRegionServers single have to persisted the namespace, all access same RegionServers files have to persisted the in all access the same RegionServers file files have to persisted the in access the same file files have to persisted in access the same fil fil t You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. inthe this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. store/access store/access data on HDFS store/access data . The on master HDFS store/access data . process The on master HDFS store/access does data . process The on the master HDFS distribution store/access does data . process The on the master HDFS distribution of does data regions . process The on the master HDFS distribution among of does regions . process The the master distribution among of does regions process the distrib amon of does reg system and system can therefore and system can host therefore and any system can region host therefore and any (figure system can region host therefore 3.8). and any (figure By system can region physically host therefore 3.8). and any (figure By can collocating region physically host therefore 3.8). any (figure By collocating region Dataphysically host 3.8). any (figure By collocating region Dataphysically 3.8). (figure By collo Data phy 3 HBase deployments, HB ase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the if MapReduce the framework isnt workload atdeployed all the if MapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i RegionServers, RegionServers, and each RegionServers, RegionServer and each RegionServers, RegionServer and typically each RegionServers, RegionServer hosts and typically multiple each RegionServers, RegionServer hosts and typically regions. multiple each RegionServer hosts and typically regions. multiple each RegionServer hosts typically regions. multiple hosts typically regions. multiple ho Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro Given that the Given underlying that the Given underlying data that is stored the Given underlying data that in is stored the Given underlying data that in is stored the Given underlying data that in is stored the underlying data in is stored data in is stored in HDFS , which HDFS is available , which HDFS is to available all , which clients HDFS is to as available all , which clients HDFS is to as available all , which clients HDF is to a a ers can theoretically ers RegionServer can theoretically ers and can write theoretically ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and the DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing a part isread also of cessing the a part workloads, isread also of cessing the a part workloads, TaskTrackers, is also of cessing the a part workloads, TaskTrackers, is DataNodes, also of cessing the a part workloads, TaskTrackers, is DataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re a single namespace, a single namespace, all aRegionServers single namespace, all aRegionServers single have namespace, all access aRegionServers single have to the namespace, all access same aRegionServers single have to persisted the namespace, all access same RegionServers files have to persisted the in all access the same RegionServers file files have to persisted the in access the same file files have to persisted in access the same fil fil t You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. inthe this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. system and system can therefore and system can host therefore and any system can region host therefore and any (figure system can region host therefore 3.8). and any (figure By system can region physically host therefore 3.8). and any (figure By can collocating region physically host therefore 3.8). any (figure By collocating region Dataphysically host 3.8). any (figure By collocating region Dataphysically 3.8). (figure By collo Data phy 3 HBase deployments, HB ase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the if MapReduce the framework isnt workload atdeployed all the if MapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you the and can data Nodes RegionServers, use locality you the and can data property; Nodes RegionServers, use locality you the and can data that property; RegionServers, use is, locality you the RegionServcan data that property; use is, locality you the RegionServcan data that property; use is, locality the RegionServ data that prope is, loc primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro ers can theoretically ers can theoretically read ers and can write theoretically read ers to and can the write theoretically local read ers to DataNode and can the write theoretically local read ers to as DataNode and can the write theoretically primary local read to as DataNode and the DataNode. write primary local read to as DataNode and the DataNode. write primary local to as DataNode the DataNode. primary local as Dat th cessing is also cessing a DataNode part is also of cessing the a DataNode part workloads, is also of cessing the a DataNode part workloads, TaskTrackers, isRegionServer also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re 1368394583 7 You can may wonder You may where wonder You the may TaskTrackers where wonder You the may TaskTrackers where are wonder You in the may this TaskTrackers where are scheme wonder You in the may this of TaskTrackers where are scheme things. wonder in the this In of TaskTrackers where some are scheme things. in the this In of TaskTrackers some are scheme things. in this In of som are sche thin Servers Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. "bar" 1368394261 "hello" HBase deployments, HBase deployments, the HB MapReduce ase deployments, the HB MapReduce ase framework deployments, the HB MapReduce ase framework isnt deployments, deployed the HB MapReduce ase framework isnt at deployments, deployed all the ifMapReduce the framework isnt workload atdeployed all the ifMapReduce the framework isnt isworkload atdeployed all if the framework isnt isworkload atdeployed all if the isn i primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, primarily and In other writes. reads random deployments, primarily and In where other writes. reads the random deployments, MapReduce and In where other writes. reads the deployments, MapReduce and In where proother writes. the deployments, MapReduce In where proother the deploy Map whe pro cf1 1368394583 22 Figure 3.7 HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on cessing is also cessing a DataNode part is also of cessing the a DataNode part workloads, is also of cessing the a DataNode part workloads, TaskTrackers, isRegionServer also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, also of cessing the a DataNode part workloads, TaskTrackers, isDataNodes, and also of the aRegionServer part workloads, TaskTrackers, DataNodes, and of the workloads, TaskTrackers, DataNodes, and TaskTracke DataNod and HB ase RegionHB ase RegionHB ase Region HB DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re "foo" 1368394925 13.6 a Servers canServers run together. canServers run together. canServers run together. canServers run together. canServers run together. can run together.
C1 tree C0 tree Disk C1 tree Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

Disk

Memory

1368393847
C1 tree Disk C1 tree

"world"

C1 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

C0 tree

C1 tree

C0 tree

C1 tree

C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Figure 3.7

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on "2011-07-04" 1368396302 "fourth of processes July" DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServ Re cf2 1.0001 1368387684 "almost the loneliest number"
Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree C1 tree C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Licensed to Licensed Nick Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Nick <ndimiduk@gmail.com Dimiduk <ndimiduk@
Memory C0 tree

Figure 3.7

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on
Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk C1 tree Memory C0 tree Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory Disk Memory

Disk C1 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Cache

Licensed to Licensed Nick Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Nick <ndimiduk@gmail.com Dimiduk <ndimiduk@
Figure 3.7
Attribution-ShareAlike 3.0 Unported License. Page 12

HBase Figure RegionServer 3.7 HBase Figure and RegionServer 3.7 HDFS HBase Figure DataNode and RegionServer 3.7 HDFS processes HBase Figure DataNode and RegionServer 3.7 are HDFS processes typically HBase Figure DataNode and RegionServer 3.7 collocated are HDFS processes typically HBase DataNode on and RegionServer the collocated are HDFS same processes typically DataNode host. on and the collocated are HDFS same processes typically DataNode host. on the collocated are same processe typically host on Licensed under a Creative Commons

High-level Architecture

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 13

Logical Data Model

Table A
rowkey column family column qualier timestamp 1368394583 1368394261 1368394583 "foo" 1368394925 1368393847 "2011-07-04" cf2 1.0001 b cf2 "thumb" 1368387684 1368387247 "almost the loneliest number" [3.6 kb png data] 1368396302 value 7 "hello" 22 13.6 "world" "fourth of July"

"bar" cf1 a

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 14

Logical Data Model

Table A
rowkey column family column qualier timestamp 1368394583 1368394261 1368394583 "foo" 1368394925 1368393847 "2011-07-04" cf2 1.0001 b cf2 "thumb" 1368387684 1368387247 "almost the loneliest number" [3.6 kb png data] 1368396302

Column Families
value 7 "hello" 22 13.6 "world" "fourth of July"

Rows
"bar" cf1 a

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 15

User API
! {rowkey => {family => {qualifier => {version => value}}}}
! Think: nested OrderedDictionary (C#), TreeMap (Java)

! Basic data operations: GET, PUT, DELETE ! SCAN over range of key-values
! benefit of the sorted rowkey business ! this is how you implement any kind of "complex query *

! GET, SCAN support Filters

! Push application logic to RegionServers

! INCREMENT, APPEND, CheckAnd{Put,Delete}

! Server-side, atomic data operations, can be contentious!
* This is also a foundational component in what we refer to as schema design in this schemaless database.
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 16

Logical Architecture
Table A
a b c d e f g h i j k l m n o p

Region Server 7
Table A, Region 1 Table A, Region 2 Table G, Region 1070 Table L, Region 25

Region 1

Region Server 86
Table A, Region 3 Table C, Region 30 Table F, Region 160 Table F, Region 776

Region 2

Region 3

Region Server 367

Table A, Region 4 Table C, Region 17 Table E, Region 52 Table P, Region 1116

Region 4

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 17

Servers can Servers run together. can Servers run together. can Servers run together. can Servers run together. can Servers run together. can run together. ers namespace, can theoretically read and write to the local DataNode as the primary DataNode. a single a single namespace, all RegionServers all RegionServers have access have to the access same to persisted the same files persisted in the files file in the file ers can theoretically read and write to the local DataNode as the primary DataNod Nodes and Nodes RegionServers, and Nodes RegionServers, you and can Nodes RegionServers, use you and the can RegionServers, data use you locality the can data property; use you locality the can data that prop use lo i system and can therefore host You may wonder where the TaskTrackers are in this scheme of things. In some system and system can therefore and can therefore host any region host any (figure region 3.8). (figure By physically 3.8). By physically collocating collocating DataDataYou may wonder where the TaskTrackers are in this scheme of things. In so ers can theoretically ers can theoretically read ers can and theoretically write read ers can to and the theoretically write read local to and DataNode the write read local to and DataNode as the the write local primar to Da as th Nodes and RegionServers, you HB ase deployments, the MapReduce framework isnt deployed at all if the workload is Nodes and Nodes RegionServers, and you canmay use you the can data use locality the data property; locality property; that is, RegionServthat is, HB aseRegionServers, deployments, the MapReduce framework isnt deployed at allRegionServif the workload You wonder You may where wonder You the may where TaskTrackers wonder You the may where TaskTrackers wonder are the in where this TaskTrackers are scheme the in this TaskTr of are sch th ers can theoretically read and DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer RegionServer DataNode RegionServer DataNode Regio Dat R primarily random reads and writes. In other deployments, where the MapReduce proers can theoretically ers can theoretically read and write read and toHB the write local tothe DataNode the local DataNode asdeployments, the primary as the DataNode. primary DataNode. primarily random reads and writes. In other where the MapReduce p HB ase deployments, ase deployments, HB ase MapReduce deployments, the HB ase MapReduce framework deployments, the MapReduce framework isnt deployed MapReduce framework isnt at deploye all if fra is t REST/Thrift JavaApp JavaApp JavaApp JavaApp HBase Shellthe You may wonder where th Gateway cessing also a part of the workloads, TaskTrackers, DataNodes, and HB ase RegionYou may is wonder You may where wonder the TaskTrackers the are in this are scheme in this of scheme things. of In things. some In some cessing is also a where part of theTaskTrackers workloads, TaskTrackers, DataNodes, and HB ase Regi primarily random primarily reads random primarily and writes. reads random primarily and In other writes. reads random deployments, and In other writes. reads deployments, and Inwhere other writes. the deplo InM wh o HB ase deployments, the MapR Servers run the together. HBase deployments, HBcan aseServers deployments, MapReduce the MapReduce framework isnt isnt deployed at all if the at all if of the workload is workloads, is can run together. cessing is also cessing a part is framework also cessing of the a deployed part workloads, is ... also cessing of the a part workloads, is TaskTrackers, also of the a workload part workloads, TaskTrackers, DataNodes, the TaskTracke DataN and HBase HBase HBase HBase primarily HBase random reads and wr HBase primarily primarily random reads random and reads writes. and In writes. other deployments, In Servers other deployments, where the where MapReduce the MapReduce proClient Clienttogether. Client Client Client Servers can Servers run can run together. can Servers run together. can run together. Client cessing is also a part of prothe wo cessing is cessing also a part is also of the a part workloads, of the workloads, TaskTrackers, TaskTrackers, DataNodes, DataNodes, and and HBase RegionHBase RegionServers can run together. DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer DataNode RegionServer Servers can Servers run together. can run together.

Physical Architecture

Figure 3.7
DataNode

HBase FigureRegionServer 3.7 HBase FigureRegionServer 3.7 and HDFS HBase Figure DataNode RegionServer 3.7 and HDFS HBase Figure processes DataNode RegionServer 3.7 and HDFS HBase are Figure processes typically DataNode RegionServer 3.7 and HDFS collocated HBase are processes typically DataNode RegionServer and on HDFS collocated are the processes typically same DataNode and host. on HDF col are thp
Region Zoo HBase Server RegionServer DataNode RegionServer DataNode Keeper Master

DataNode

RegionServer DataNode RegionServer DataNode DataNode RegionServer DataNode RegionServer DataNode RegionServer RegionServer DataNode DataNode RegionSe DataNod DataNode RegionServer Region Region Region Server DataNode Server RegionServer RegionServer DataNode RegionServer DataNode RegionServer Server
... Name Node

Zoo Keeper

Zoo Keeper

Data Node

Data Node

Data Node

Data Node

Figure 3.7 HBase and HDFS DataNode processes are typically collocated oncollocated the same host. FigureRegionServer 3.7 HBase RegionServer and HDFS DataNode processes are typically on the same h
Attribution-ShareAlike 3.0 Unported License.

Licensed to Licensed Nick Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gmail.com> Dimiduk to Licensed Nick <ndimiduk@gm Dimiduk to Nick <ndim Dim

Figure 3.7 HBase Figure RegionServer 3.7 HBase Figure RegionServer 3.7 and HDFS HBase Figure DataNode RegionServer 3.7 and HDFS HBase processes DataNode RegionServer and HDFS are processes typically DataNode and HDFS collocated are process typical DataN Licensed under a Creative Commons Page 18

Anatomy of a RegionServer

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 19

So what is HBase anyway?

C1 tree C0 tree

DataNode

RegionServer

Disk C1 tree

Memory C0 tree

Disk C1 tree

Memory C0 tree

Cache

Disk C1 tree

Memory C0 tree

Disk

Memory

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 20

Storage Machinery
RegionServer (HBase) BlockCache HLog (WAL) HRegion HStore StoreFile StoreFile MemStore ... ... HFile HFile ... ... HStore HRegion HStore HStore

DataNode (Hadoop DFS)

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 21

Storage Machinery
RegionServer (HBase)

Cache
BlockCache HLog (WAL) HRegion HStore StoreFile StoreFile MemStore HStore
C1

HRegion HStore
C1

HStore
C1

C1
HFile HFile

...

C0

... ...
C0 C0

...
C0

DataNode (Hadoop DFS)

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 22

Storage Machinery: Write Path

1
RegionServer (HBase)

BlockCache HLog (WAL)

HRegion HStore StoreFile StoreFile MemStore HStore

HRegion HStore HStore

3
... HFile HFile

... ... ...

DataNode (Hadoop DFS)

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 23

Storage Machinery: Read Path

1
RegionServer (HBase)

3
HLog (WAL) HRegion HStore StoreFile

BlockCache

4
HStore

HRegion HStore HStore

2 3

StoreFile

MemStore

2
... ... ...

... HFile HFile

DataNode (Hadoop DFS)

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 24

By Example

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 25

Database Dichotomy
Latency
Lossy WAL Durability Compression Smart Rowkey Design Smaller Block Size Smart Rowkey Design Compression Row+Column Bloomlter Larger BlockCache

Write

Read
Compression Smart Rowkey Design Less Frequent Compactions Larger Increased Block Size Scanner Caching

Larger Flush Size

Larger Region Size

More Regions Bulkload Smart Rowkey HFiles Design Compression

Throughput
Page 26

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Web-scale Database
App server Counters Sessions User proles Social Media Application Data App server
Write Read Latency

App server

App server

App server

Throughput

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 27

BigIndex
Search Search App server Search Search App server "BigIndex" Document store Search App server

App server

Latency

App server

Write

Read

Throughput

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 28

Materialized View
App server App server

App server

App server

App server

Latency

Write

Read

Throughput

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 29

ETL Assist

Latency

Write

Read

Throughput

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 30

Lambda Architecture

App server

App server

App server

App server

App server

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 31

Resources

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 32

Join the Community!

! hbase.apache.org
! blogs.apache.org/hbase/ ! hbase.apache.org/mail-lists.html

! IRC: irc.freenode.net#hbase ! JIRA: issues.apache.org/jira/browse/HBASE ! Source

! git clone git://git.apache.org/hbase.git ! svn checkout http://svn.apache.org/repos/asf/hbase/trunk hbase

! In person
! HBaseCon, hbasecon.com ! Hadoop Summit, hadoopsummit.org ! Local meetup near you
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 33

HBase 0.96.x
! Stability ! Horizontal Scalability (1000s of machines) ! Speed of Recovery (MTTR) ! Operations ! Future-proofing RPC (protobuf) ! Improved Multi-tenancy ! Hadoop1, Hadoop2 ! Plenty more enhancements, new features:
! https://blogs.apache.org/hbase/entry/hbase_0_96_0_released

! *Backward incompatible*

Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

Page 34

Thanks!
Nick Dimiduk
github.com/ndimiduk
Nick Dimiduk Amandeep Khurana
FOREWORD BY Michael Stack

@xefyr n10k.com

MANNING

hbaseinaction.com
Licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Page 35