PROVIDER Β· Distributed wide-table database for the Hadoop ecosystem
Apache HBase
Apache HBase is the open implementation of the Bigtable paper, adding random access to Hadoop on top of HDFS and sharing data with Spark, Hive, and Phoenix, fitting wide-table and detail lookups in existing big data platforms.
3.8pts
Pros
Cons
Product Information
Detailed Notes
Overview
Apache HBase was open sourced in 2008 and became an Apache top-level project in 2010 as the open implementation of the Bigtable paper, built on HDFS to add random access to the Hadoop ecosystem and among the most common wide-table stores in big data platforms.
For Backend & Database work the value is random access on HDFS, since HDFS alone suits sequential scans while HBase provides millisecond row-key reads and writes through RegionServers and MemStore, and Phoenix adds SQL on top, fitting detail lookups, tag storage, and write-wide log data.
The limit is operational complexity, because it depends on ZooKeeper and HDFS with region splits, compaction, and GC tuning needing experience, lacks secondary indexes without Phoenix, and has no cross-row transactions, so standing up a dedicated HBase stack without an existing Hadoop platform rarely pays off.
Core Strengths
- Millisecond random access on HDFS fills the sequential-scan-only gap in Hadoop
- Tables partition automatically into regions and scale horizontally with data
- Seamless integration with Spark, Hive, and Phoenix shares one copy of data
- Per-column-family tuning suits sparse wide tables
Product Ecosystem
Tables and column families
Data sorts by row key with column families fixed at creation and qualifiers added dynamically, while row key design governs hotspots and distribution.
Components and dependencies
HMaster handles metadata and RegionServers serve reads and writes, depending on ZooKeeper coordination and HDFS storage across a long component chain.
SQL and integration
Phoenix adds secondary indexes and SQL while the Spark connector handles bulk reads, writes, and offline compute.
Drawbacks
- ZooKeeper and HDFS dependencies make deployment and troubleshooting costly
- No native secondary indexes or cross-row transactions without Phoenix or app logic
- Region split and compaction tuning needs senior experience or latency jitters
- Poor value when no Hadoop platform already exists
Use Cases
- Big data estates already on Hadoop (β β β β ) - Shares storage with Hive and Spark
- Random lookup across massive detail (β β β β ) - Millisecond row-key reads
- Sparse wide tables and tags (β β β β ) - Column families grow on demand
- Lightweight projects without Hadoop (β ) - Infrastructure cost too high
Pricing
| Plan | Price | Scope |
|---|---|---|
| Apache HBase | $0 | Open source, self-managed |
| Managed HBase | Per node and storage | Operations-free with backup |
| Cloud-native alternatives | Per node or request | Protocol compatible, managed |
The open source edition carries no licence fee while real cost is cluster size and staff, needing at least five machines across ZooKeeper and HDFS roles with dedicated operations, so managed wide-table services usually cost less without an existing Hadoop platform.
FAQ
-
HBase or Cassandra? HBase when a Hadoop estate exists and data must be shared with Spark and Hive, Cassandra for multi-datacenter resilience and simpler operations.
-
How are row key hotspots avoided? Avoid timestamp or sequential ID prefixes, use hashed or salted prefixes, and keep keys compact since length affects storage.
-
Is Phoenix worth adopting? Yes when SQL and secondary indexes are needed, though index maintenance adds write load, so weigh it for frequently updated tables.