Apache Cassandra

Apache Cassandra was initially built by Facebook for their own servers and services, using principles developed by Google and Amazon for BigTable and DynamoDB.

Currently the project was donated to the Apache Foundation who continues to developed it further.

Cassandra is extremely fast, even when dealing with millions of records and operations, a feature that was absolutely crucial to sustaining Facebook's userbase and real-time interaction capabilities.

Getting started with Cassandra should be pretty easy thanks to the very well organized documentation and the plethora of online tutorials.

The database is currently used in many Internet companies like Rackspace, Spotify, Netflix, RockMelt, Embedly, Ericcson, Expedia, FormSpring, Adobe, Twitter, Cisco, Digg, Mollom, Digital River, Page2RSS, Real, Reddit, eBuddy, EngineYard, Symantec, Soundcloud, Accenture, Rhapsody, Gowalla, HP, IBM, Mahalo, Walmart, OpenX and others. As you can see, this is a Who's Who list when it comes to Silicon Valley companies.

What is new in this release:

• EACH_QUORUM is now a supported consistency level for read requests.
• Support for IN restrictions on any partition key component or clustering key as well as support for EQ and IN multicolumn restrictions has been added to UPDATE and DELETE statement.
• Support for single-column and multi-colum slice restrictions (>, >=, <= and <) has been added to DELETE statements
• nodetool rebuild_index accepts the index argument without the redundant table name
• Materialized Views, which allow for server-side denormalization, is now available. Materialized views provide an alternative to secondary indexes for non-primary key queries, and perform much better for indexing high cardinality columns.
• Hinted handoff has been completely rewritten. Hints are now stored in flat files, with less overhead for storage and more efficient dispatch.
• Option to not purge unrepaired tombstones. To avoid users having data resurrected if repair has not been run within gc_grace_seconds, an option has been added to only allow tombstones from repaired sstables to be purged. To enable, set the compaction option 'only_purge_repaired_tombstones':true but keep in mind that if you do not run repair for a long time, you will keep all tombstones around which can cause other problems.
• Enabled warning on GC taking longer than 1000ms.

What is new in version 3.4.0 / 2.2.5:

• EACH_QUORUM is now a supported consistency level for read requests.
• Support for IN restrictions on any partition key component or clustering key as well as support for EQ and IN multicolumn restrictions has been added to UPDATE and DELETE statement.
• Support for single-column and multi-colum slice restrictions (>, >=, <= and <) has been added to DELETE statements
• nodetool rebuild_index accepts the index argument without the redundant table name
• Materialized Views, which allow for server-side denormalization, is now available. Materialized views provide an alternative to secondary indexes for non-primary key queries, and perform much better for indexing high cardinality columns.
• Hinted handoff has been completely rewritten. Hints are now stored in flat files, with less overhead for storage and more efficient dispatch.
• Option to not purge unrepaired tombstones. To avoid users having data resurrected if repair has not been run within gc_grace_seconds, an option has been added to only allow tombstones from repaired sstables to be purged. To enable, set the compaction option 'only_purge_repaired_tombstones':true but keep in mind that if you do not run repair for a long time, you will keep all tombstones around which can cause other problems.
• Enabled warning on GC taking longer than 1000ms.

What is new in version 3.3.0 / 2.2.5:

• EACH_QUORUM is now a supported consistency level for read requests.
• Support for IN restrictions on any partition key component or clustering key as well as support for EQ and IN multicolumn restrictions has been added to UPDATE and DELETE statement.
• Support for single-column and multi-colum slice restrictions (>, >=, <= and <) has been added to DELETE statements
• nodetool rebuild_index accepts the index argument without the redundant table name
• Materialized Views, which allow for server-side denormalization, is now available. Materialized views provide an alternative to secondary indexes for non-primary key queries, and perform much better for indexing high cardinality columns.
• Hinted handoff has been completely rewritten. Hints are now stored in flat files, with less overhead for storage and more efficient dispatch.
• Option to not purge unrepaired tombstones. To avoid users having data resurrected if repair has not been run within gc_grace_seconds, an option has been added to only allow tombstones from repaired sstables to be purged. To enable, set the compaction option 'only_purge_repaired_tombstones':true but keep in mind that if you do not run repair for a long time, you will keep all tombstones around which can cause other problems.
• Enabled warning on GC taking longer than 1000ms.

What is new in version 3.2.1 / 2.2.4:

• EACH_QUORUM is now a supported consistency level for read requests.
• Support for IN restrictions on any partition key component or clustering key as well as support for EQ and IN multicolumn restrictions has been added to UPDATE and DELETE statement.
• Support for single-column and multi-colum slice restrictions (>, >=, <= and <) has been added to DELETE statements
• nodetool rebuild_index accepts the index argument without the redundant table name
• Materialized Views, which allow for server-side denormalization, is now available. Materialized views provide an alternative to secondary indexes for non-primary key queries, and perform much better for indexing high cardinality columns.
• Hinted handoff has been completely rewritten. Hints are now stored in flat files, with less overhead for storage and more efficient dispatch.
• Option to not purge unrepaired tombstones. To avoid users having data resurrected if repair has not been run within gc_grace_seconds, an option has been added to only allow tombstones from repaired sstables to be purged. To enable, set the compaction option 'only_purge_repaired_tombstones':true but keep in mind that if you do not run repair for a long time, you will keep all tombstones around which can cause other problems.
• Enabled warning on GC taking longer than 1000ms.

What is new in version 3.0.0 / 2.2.3:

• EACH_QUORUM is now a supported consistency level for read requests.
• Support for IN restrictions on any partition key component or clustering key as well as support for EQ and IN multicolumn restrictions has been added to UPDATE and DELETE statement.
• Support for single-column and multi-colum slice restrictions (>, >=, <= and <) has been added to DELETE statements
• nodetool rebuild_index accepts the index argument without the redundant table name
• Materialized Views, which allow for server-side denormalization, is now available. Materialized views provide an alternative to secondary indexes for non-primary key queries, and perform much better for indexing high cardinality columns.
• Hinted handoff has been completely rewritten. Hints are now stored in flat files, with less overhead for storage and more efficient dispatch.
• Option to not purge unrepaired tombstones. To avoid users having data resurrected if repair has not been run within gc_grace_seconds, an option has been added to only allow tombstones from repaired sstables to be purged. To enable, set the compaction option 'only_purge_repaired_tombstones':true but keep in mind that if you do not run repair for a long time, you will keep all tombstones around which can cause other problems.
• Enabled warning on GC taking longer than 1000ms.

What is new in version 2.2.1 / 1.2.19 / 3.0.0-rc1:

• Removed "ALTER TYPE <name> RENAME TO <name>" from tab-completion
• Fixed IllegalStateException in anticompaction
• cqlsh: DESCRIBE support for frozen UDTs, tuples
• Avoid exposing internal classes over JMX
• Added null check for keys when freezing collection
• Improved stress workload realism

What is new in version 2.1.6 / 1.2.19 / 2.2.0-beta1:

• Removed "ALTER TYPE <name> RENAME TO <name>" from tab-completion
• Fixed IllegalStateException in anticompaction
• cqlsh: DESCRIBE support for frozen UDTs, tuples
• Avoid exposing internal classes over JMX
• Added null check for keys when freezing collection
• Improved stress workload realism

What is new in version 2.1.5 / 1.2.19 / 2.2.0-beta1:

• Removed "ALTER TYPE <name> RENAME TO <name>" from tab-completion
• Fixed IllegalStateException in anticompaction
• cqlsh: DESCRIBE support for frozen UDTs, tuples
• Avoid exposing internal classes over JMX
• Added null check for keys when freezing collection
• Improved stress workload realism

What is new in version 2.1.5 / 1.2.19:

• Removed "ALTER TYPE <name> RENAME TO <name>" from tab-completion
• Fixed IllegalStateException in anticompaction
• cqlsh: DESCRIBE support for frozen UDTs, tuples
• Avoid exposing internal classes over JMX
• Added null check for keys when freezing collection
• Improved stress workload realism

What is new in version 2.1.3 / 1.2.19:

• Removed "ALTER TYPE <name> RENAME TO <name>" from tab-completion
• Fixed IllegalStateException in anticompaction
• cqlsh: DESCRIBE support for frozen UDTs, tuples
• Avoid exposing internal classes over JMX
• Added null check for keys when freezing collection
• Improved stress workload realism

What is new in version 2.0.2:

• Fixes for speculative retry
• Improved memory usage of metadata min/max column names
• Fixed thrift validation refusing row markers on CQL3 tables
• Fixed insertion of collections with CAS
• Correctly send metadata on SELECT COUNT
• Track clients' remote addresses in ClientState
• Create snapshot dir if it does not exist when migrating leveled manifest
• Make sequential nodetool repair the default
• Added more hooks for compaction strategy implementations
• Fixed potential NPE on composite 2ndary indexes
• Updated FailureDetector to use nanontime
• Fixed FileCacheService regressions
• Never return WriteTimeout for CL.ANY
• Fixed race conditions in bulk loader
• Added configurable metrics reporting
• Drop queries exceeding a configurable number of tombstones
• Track and persist sstable read activity

What is new in version 2.0.0:

• Fix thrift validation when inserting into CQL3 tables
• Fix periodic memtable flushing behavior with clean memtables
• Fix dateOf() function for pre-2.0 timestamp columns
• Fix SSTable unintentionally loads BF when opened for batch
• Add stream session progress to JMX
• Fix NPE during CAS operation
• Fix getBloomFilterDiskSpaceUsed for AlwaysPresentFilter
• Don't announce schema version until we've loaded the changes locally
• Fix to support off heap bloom filters size greater than 2 GB
• Properly handle parsing huge map and set literals

What is new in version 1.2.8:

• Fixed reading DeletionTime from 1.1-format stables
• cqlsh: add collections support to COPY
• Retry important messages for any IOException
• Allow empty IN relations in SELECT/UPDATE/DELETE statements
• cqlsh: fix crashing on Windows due to libedit detection

What is new in version 1.2.0:

• Disallow counters in collections.
• cqlsh: added unit tests.
• Fixed default bloom_filter_fp_chance for LeveledCompactionStrategy.

What is new in version 1.2.0 beta3:

• Removed all remnants of removed nodes.
• Added aut-reloading of the log4j file in debian package.
• Fixed estimated row cache entry size.
• reset getRangeSlice filter after finishing a row for get_paged_slice
• Expunges row cache post-truncate.
• Allow static CF definition with compact storage.
• Fixed endless loop/compaction of schema_* CFs due to broken timestamps.
• Fixed 'wrong class type' assertion in CounterColumn.

What is new in version 1.1.6:

• Wait for writes on synchronous read digest mismatch.
• Fix commitlog replay for nanotime-infected sstables.
• Preflight check ttl for maximum of 20 years.
• (Pig) fix widerow input with single column rows.
• Fix HH to compact with correct gcBefore, which avoids wiping out undelivered hints.
• LCS will merge up to 32 L0 sstables as intended.
• NTS will default unconfigured DC replicas to zero.
• Use default consistency level in counter validation if none is explicitly provide.
• Improve IAuthority interface by introducing fine-grained access permissions and grant/revoke commands.
• Fix assumption error in CLI when updating/describing keyspace.
• Adds offline sstablescrub to debian packaging.

What is new in version 1.1.4:

• Fixed offline scrub to catch >= out of order rows.
• Fixed cassandra-env.sh on RHEL and other non-dash-based systems.

What is new in version 1.1.2:

• Use CF comparator to sort indexed columns in SecondaryIndexManager.
• Added strategy_options to the KSMetaData.toString() output.
• Fixed range queries containing unqueried results.
• Allow updating column_alias types.
• Fixed deletion bug.
• Fixed computation of overlapping sstable for leveled compaction.
• Improved scrub and allow to run it offline.
• Fixed assertionError in StorageService.bulkLoad.

What is new in version 1.1.1:

• Fixed split generation regression.
• Ignore min/max compactions settings in LCS, while preserving behavior that min=max=0 disables autocompaction.
• Log number of rows read from saved cache.
• Calculate exact size required for cleanup operations.
• Avoid blocking additional writes during flush when the commitlog gets behind temporarily.
• Enable caching on index CFs based on data CF cache setting.
• Warn on invalid replication strategy creation options.
• Removed [Freeable]Memory finalizers.
• Include tombstone size in ColumnFamily.size, which can prevent OOM during sudden mass delete operations by yielding a nonzero liveRatio.
• Open 1 sstableScanner per level for leveled compaction.

What is new in version 1.1.0:

• Average a reduced liveRatio estimate with the previous one.
• Allow KS and CF names up to 48 characters.
• Fixed stress build.
• Added time remaining estimate to nodetool compactionstats.

Requirements:

• Java 1.7 or higher
• Python 2.7 or higher