Data Definition

CQL stores data in tables, whose schema defines the layout of said data in the table, and those tables are grouped in keyspaces. A keyspace defines a number of options that applies to all the tables it contains, most prominently of which is the replication strategy used by the keyspace. It is generally encouraged to use one keyspace by application, and thus many cluster may define only one keyspace.

This section describes the statements used to create, modify, and remove those keyspace and tables.

Common definitions

The names of the keyspaces and tables are defined by the following grammar:

keyspace_name ::=  name
table_name    ::=  [ keyspace_name '.' ] name
name          ::=  unquoted_name | quoted_name
unquoted_name ::=  re('[a-zA-Z_0-9]{1, 48}')
quoted_name   ::=  '"' unquoted_name '"'

Both keyspace and table name should be comprised of only alphanumeric characters, cannot be empty and are limited in size to 48 characters (that limit exists mostly to avoid filenames (which may include the keyspace and table name) to go over the limits of certain file systems). By default, keyspace and table names are case insensitive (myTable is equivalent to mytable) but case sensitivity can be forced by using double-quotes ("myTable" is different from mytable).

Further, a table is always part of a keyspace and a table name can be provided fully-qualified by the keyspace it is part of. If is is not fully-qualified, the table is assumed to be in the current keyspace (see USE statement).

Further, the valid names for columns is simply defined as:

column_name ::=  identifier

We also define the notion of statement options for use in the following section:

options ::=  option ( AND option )*
option  ::=  identifier '=' ( identifier | constant | map_literal )


A keyspace is created using a CREATE KEYSPACE statement:

create_keyspace_statement ::=  CREATE KEYSPACE [ IF NOT EXISTS ] keyspace_name WITH options

For instance:

    WITH replication = {'class': 'SimpleStrategy', 'replication_factor' : 3};

    WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1' : 1, 'DC2' : 3}
    AND durable_writes = false;

Attempting to create a keyspace that already exists will return an error unless the IF NOT EXISTS option is used. If it is used, the statement will be a no-op if the keyspace already exists.

The supported options are:

name kind mandatory default description
replication map yes   The replication strategy and options to use for the keyspace (see details below).
durable_writes simple no true Whether to use the commit log for updates on this keyspace (disable this option at your own risk!).

The replication property is mandatory and must at least contains the 'class' sub-option which defines the replication strategy class to use. The rest of the sub-options depends on what replication strategy is used. By default, Cassandra support the following 'class':


A simple strategy that defines a replication factor for data to be spread across the entire cluster. This is generally not a wise choice for production because it does not respect datacenter layouts and can lead to wildly varying query latency. For a production ready strategy, see NetworkTopologyStrategy. SimpleStrategy supports a single mandatory argument:

sub-option type since description
'replication_factor' int all The number of replicas to store per range


A production ready replication strategy that allows to set the replication factor independently for each data-center. The rest of the sub-options are key-value pairs where a key is a data-center name and its value is the associated replication factor. Options:

sub-option type since description
'<datacenter>' int all The number of replicas to store per range in the provided datacenter.
'replication_factor' int 4.0 The number of replicas to use as a default per datacenter if not specifically provided. Note that this always defers to existing definitions or explicit datacenter settings. For example, to have three replicas per datacenter, supply this with a value of 3.

Note that when ALTER ing keyspaces and supplying replication_factor, auto-expansion will only add new datacenters for safety, it will not alter existing datacenters or remove any even if they are no longer in the cluster. If you want to remove datacenters while still supplying replication_factor, explicitly zero out the datacenter you want to have zero replicas.

An example of auto-expanding datacenters with two datacenters: DC1 and DC2:

    WITH replication = {'class': 'NetworkTopologyStrategy', 'replication_factor' : 3}

    CREATE KEYSPACE excalibur WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1': '3', 'DC2': '3'} AND durable_writes = true;

An example of auto-expanding and overriding a datacenter:

    WITH replication = {'class': 'NetworkTopologyStrategy', 'replication_factor' : 3, 'DC2': 2}

    CREATE KEYSPACE excalibur WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1': '3', 'DC2': '2'} AND durable_writes = true;

An example that excludes a datacenter while using replication_factor:

    WITH replication = {'class': 'NetworkTopologyStrategy', 'replication_factor' : 3, 'DC2': 0} ;

    CREATE KEYSPACE excalibur WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1': '3'} AND durable_writes = true;

If transient replication has been enabled, transient replicas can be configured for both SimpleStrategy and NetworkTopologyStrategy by defining replication factors in the format '<total_replicas>/<transient_replicas>'

For instance, this keyspace will have 3 replicas in DC1, 1 of which is transient, and 5 replicas in DC2, 2 of which are transient:

CREATE KEYSPACE some_keysopace
           WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1' : '3/1'', 'DC2' : '5/2'};


The USE statement allows to change the current keyspace (for the connection on which it is executed). A number of objects in CQL are bound to a keyspace (tables, user-defined types, functions, …) and the current keyspace is the default keyspace used when those objects are referred without a fully-qualified name (that is, without being prefixed a keyspace name). A USE statement simply takes the keyspace to use as current as argument:

use_statement ::=  USE keyspace_name


An ALTER KEYSPACE statement allows to modify the options of a keyspace:

alter_keyspace_statement ::=  ALTER KEYSPACE keyspace_name WITH options

For instance:

    WITH replication = {'class': 'SimpleStrategy', 'replication_factor' : 4};

The supported options are the same than for creating a keyspace.


Dropping a keyspace can be done using the DROP KEYSPACE statement:

drop_keyspace_statement ::=  DROP KEYSPACE [ IF EXISTS ] keyspace_name

For instance:


Dropping a keyspace results in the immediate, irreversible removal of that keyspace, including all the tables, UTD and functions in it, and all the data contained in those tables.

If the keyspace does not exists, the statement will return an error, unless IF EXISTS is used in which case the operation is a no-op.


Creating a new table uses the CREATE TABLE statement:

create_table_statement ::=  CREATE TABLE [ IF NOT EXISTS ] table_name
                                ( ',' column_definition )*
                                [ ',' PRIMARY KEY '(' primary_key ')' ]
                            ')' [ WITH table_options ]
column_definition      ::=  column_name cql_type [ STATIC ] [ PRIMARY KEY]
primary_key            ::=  partition_key [ ',' clustering_columns ]
partition_key          ::=  column_name
                            | '(' column_name ( ',' column_name )* ')'
clustering_columns     ::=  column_name ( ',' column_name )*
table_options          ::=  COMPACT STORAGE [ AND table_options ]
                            | CLUSTERING ORDER BY '(' clustering_order ')' [ AND table_options ]
                            | options
clustering_order       ::=  column_name (ASC | DESC) ( ',' column_name (ASC | DESC) )*

For instance:

CREATE TABLE monkeySpecies (
    species text PRIMARY KEY,
    common_name text,
    population varint,
    average_size int
) WITH comment='Important biological records';

CREATE TABLE timeline (
    userid uuid,
    posted_month int,
    posted_time uuid,
    body text,
    posted_by text,
    PRIMARY KEY (userid, posted_month, posted_time)
) WITH compaction = { 'class' : 'LeveledCompactionStrategy' };

    machine inet,
    cpu int,
    mtime timeuuid,
    load float,
    PRIMARY KEY ((machine, cpu), mtime)

A CQL table has a name and is composed of a set of rows. Creating a table amounts to defining which columns the rows will be composed, which of those columns compose the primary key, as well as optional options for the table.

Attempting to create an already existing table will return an error unless the IF NOT EXISTS directive is used. If it is used, the statement will be a no-op if the table already exists.

Every rows in a CQL table has a set of predefined columns defined at the time of the table creation (or added later using an alter statement).

A column_definition is primarily comprised of the name of the column defined and it’s type, which restrict which values are accepted for that column. Additionally, a column definition can have the following modifiers:

it declares the column as being a static column.
it declares the column as being the sole component of the primary key of the table.

Some columns can be declared as STATIC in a table definition. A column that is static will be “shared” by all the rows belonging to the same partition (having the same partition key). For instance:

    pk int,
    t int,
    v text,
    s text static,
    PRIMARY KEY (pk, t)

INSERT INTO t (pk, t, v, s) VALUES (0, 0, 'val0', 'static0');
INSERT INTO t (pk, t, v, s) VALUES (0, 1, 'val1', 'static1');

   pk | t | v      | s
   0  | 0 | 'val0' | 'static1'
   0  | 1 | 'val1' | 'static1'

As can be seen, the s value is the same (static1) for both of the row in the partition (the partition key in that example being pk, both rows are in that same partition): the 2nd insertion has overridden the value for s.

The use of static columns as the following restrictions:

  • tables with the COMPACT STORAGE option (see below) cannot use them.
  • a table without clustering columns cannot have static columns (in a table without clustering columns, every partition has only one row, and so every column is inherently static).
  • only non PRIMARY KEY columns can be static.

Within a table, a row is uniquely identified by its PRIMARY KEY, and hence all table must define a PRIMARY KEY (and only one). A PRIMARY KEY definition is composed of one or more of the columns defined in the table. Syntactically, the primary key is defined the keywords PRIMARY KEY followed by comma-separated list of the column names composing it within parenthesis, but if the primary key has only one column, one can alternatively follow that column definition by the PRIMARY KEY keywords. The order of the columns in the primary key definition matter.

A CQL primary key is composed of 2 parts:

  • the partition key part. It is the first component of the primary key definition. It can be a single column or, using additional parenthesis, can be multiple columns. A table always have at least a partition key, the smallest possible table definition is:

  • the clustering columns. Those are the columns after the first component of the primary key definition, and the order of those columns define the clustering order.

Some example of primary key definition are:

  • PRIMARY KEY (a): a is the partition key and there is no clustering columns.
  • PRIMARY KEY (a, b, c) : a is the partition key and b and c are the clustering columns.
  • PRIMARY KEY ((a, b), c) : a and b compose the partition key (this is often called a composite partition key) and c is the clustering column.

Within a table, CQL defines the notion of a partition. A partition is simply the set of rows that share the same value for their partition key. Note that if the partition key is composed of multiple columns, then rows belong to the same partition only they have the same values for all those partition key column. So for instance, given the following table definition and content:

    a int,
    b int,
    c int,
    d int,
    PRIMARY KEY ((a, b), c, d)

   a | b | c | d
   0 | 0 | 0 | 0    // row 1
   0 | 0 | 1 | 1    // row 2
   0 | 1 | 2 | 2    // row 3
   0 | 1 | 3 | 3    // row 4
   1 | 1 | 4 | 4    // row 5

row 1 and row 2 are in the same partition, row 3 and row 4 are also in the same partition (but a different one) and row 5 is in yet another partition.

Note that a table always has a partition key, and that if the table has no clustering columns, then every partition of that table is only comprised of a single row (since the primary key uniquely identifies rows and the primary key is equal to the partition key if there is no clustering columns).

The most important property of partition is that all the rows belonging to the same partition are guarantee to be stored on the same set of replica nodes. In other words, the partition key of a table defines which of the rows will be localized together in the Cluster, and it is thus important to choose your partition key wisely so that rows that needs to be fetch together are in the same partition (so that querying those rows together require contacting a minimum of nodes).

Please note however that there is a flip-side to this guarantee: as all rows sharing a partition key are guaranteed to be stored on the same set of replica node, a partition key that groups too much data can create a hotspot.

Another useful property of a partition is that when writing data, all the updates belonging to a single partition are done atomically and in isolation, which is not the case across partitions.

The proper choice of the partition key and clustering columns for a table is probably one of the most important aspect of data modeling in Cassandra, and it largely impact which queries can be performed, and how efficiently they are.

The clustering columns of a table defines the clustering order for the partition of that table. For a given partition, all the rows are physically ordered inside Cassandra by that clustering order. For instance, given:

    a int,
    b int,
    c int,
    PRIMARY KEY (a, b, c)

   a | b | c
   0 | 0 | 4     // row 1
   0 | 1 | 9     // row 2
   0 | 2 | 2     // row 3
   0 | 3 | 3     // row 4

then the rows (which all belong to the same partition) are all stored internally in the order of the values of their b column (the order they are displayed above). So where the partition key of the table allows to group rows on the same replica set, the clustering columns controls how those rows are stored on the replica. That sorting allows the retrieval of a range of rows within a partition (for instance, in the example above, SELECT * FROM t WHERE a = 0 AND b > 1 and b <= 3) to be very efficient.

A CQL table has a number of options that can be set at creation (and, for most of them, altered later). These options are specified after the WITH keyword.

Amongst those options, two important ones cannot be changed after creation and influence which queries can be done against the table: the COMPACT STORAGE option and the CLUSTERING ORDER option. Those, as well as the other options of a table are described in the following sections.


Since Cassandra 3.0, compact tables have the exact same layout internally than non compact ones (for the same schema obviously), and declaring a table compact only creates artificial limitations on the table definition and usage. It only exists for historical reason and is preserved for backward compatibility And as COMPACT STORAGE cannot, as of Cassandra 4.0-alpha5, be removed, it is strongly discouraged to create new table with the COMPACT STORAGE option.

A compact table is one defined with the COMPACT STORAGE option. This option is only maintained for backward compatibility for definitions created before CQL version 3 and shouldn’t be used for new tables. Declaring a table with this option creates limitations for the table which are largely arbitrary (and exists for historical reasons). Amongst those limitation:

  • a compact table cannot use collections nor static columns.
  • if a compact table has at least one clustering column, then it must have exactly one column outside of the primary key ones. This imply you cannot add or remove columns after creation in particular.
  • a compact table is limited in the indexes it can create, and no materialized view can be created on it.

The clustering order of a table is defined by the clustering columns of that table. By default, that ordering is based on natural order of those clustering order, but the CLUSTERING ORDER allows to change that clustering order to use the reverse natural order for some (potentially all) of the columns.

The CLUSTERING ORDER option takes the comma-separated list of the clustering column, each with a ASC (for ascendant, e.g. the natural order) or DESC (for descendant, e.g. the reverse natural order). Note in particular that the default (if the CLUSTERING ORDER option is not used) is strictly equivalent to using the option with all clustering columns using the ASC modifier.

Note that this option is basically a hint for the storage engine to change the order in which it stores the row but it has 3 visible consequences:

# it limits which ORDER BY clause are allowed for selects on that table. You can only
order results by the clustering order or the reverse clustering order. Meaning that if a table has 2 clustering column a and b and you defined WITH CLUSTERING ORDER (a DESC, b ASC), then in queries you will be allowed to use ORDER BY (a DESC, b ASC) and (reverse clustering order) ORDER BY (a ASC, b DESC) but not ORDER BY (a ASC, b ASC) (nor ORDER BY (a DESC, b DESC)).
# it also change the default order of results when queried (if no ORDER BY is provided). Results are always returned
in clustering order (within a partition).
# it has a small performance impact on some queries as queries in reverse clustering order are slower than the one in
forward clustering order. In practice, this means that if you plan on querying mostly in the reverse natural order of your columns (which is common with time series for instance where you often want data from the newest to the oldest), it is an optimization to declare a descending clustering order.


review (misses cdc if nothing else) and link to proper categories when appropriate (compaction for instance)

A table supports the following options:

option kind default description
comment speculative_retry simple simple none 99PERCENTILE A free-form, human-readable comment. Speculative retry options.
cdc boolean false Create a Change Data Capture (CDC) log on the table.
additional_write_policy simple 99PERCENTILE Speculative retry options.
gc_grace_seconds simple 864000 Time to wait before garbage collecting tombstones (deletion markers).
bloom_filter_fp_chance simple 0.00075 The target probability of false positive of the sstable bloom filters. Said bloom filters will be sized to provide the provided probability (thus lowering this value impact the size of bloom filters in-memory and on-disk)
default_time_to_live simple 0 The default expiration time (“TTL”) in seconds for a table.
compaction map see below Compaction options.
compression map see below Compression options.
caching map see below Caching options.
memtable_flush_period_in_ms simple 0 Time (in ms) before Cassandra flushes memtables to disk.
read_repair simple BLOCKING Sets read repair behavior (see below)

By default, Cassandra read coordinators only query as many replicas as necessary to satisfy consistency levels: one for consistency level ONE, a quorum for QUORUM, and so on. speculative_retry determines when coordinators may query additional replicas, which is useful when replicas are slow or unresponsive. Speculative retries are used to reduce the latency. The speculative_retry option may be used to configure rapid read protection with which a coordinator sends more requests than needed to satisfy the Consistency level.

Pre-4.0 speculative Retry Policy takes a single string as a parameter, this can be NONE, ALWAYS, 99PERCENTILE (PERCENTILE), 50MS (CUSTOM).

Examples of setting speculative retry are:

ALTER TABLE users WITH speculative_retry = '10ms';


ALTER TABLE users WITH speculative_retry = '99PERCENTILE';

The problem with these settings is when a single host goes into an unavailable state this drags up the percentiles. This means if we are set to use p99 alone, we might not speculate when we intended to to because the value at the specified percentile has gone so high. As a fix 4.0 adds support for hybrid MIN(), MAX() speculative retry policies (CASSANDRA-14293). This means if the normal p99 for the table is <50ms, we will still speculate at this value and not drag the tail latencies up… but if the p99th goes above what we know we should never exceed we use that instead.

In 4.0 the values (case-insensitive) discussed in the following table are supported:

As of version 4.0 speculative retry allows more friendly params (CASSANDRA-13876). The speculative_retry is more flexible with case. As an example a value does not have to be NONE, and the following are supported alternatives.

alter table users WITH speculative_retry = 'none';
alter table users WITH speculative_retry = 'None';

The text component is case insensitive and for nPERCENTILE version 4.0 allows nP, for instance 99p. In a hybrid value for speculative retry, one of the two values must be a fixed millisecond value and the other a percentile value.

Some examples:


Two values of the same kind cannot be specified such as min(90percentile,99percentile) as it wouldn’t be a hybrid value. This setting does not affect reads with consistency level ALL because they already query all replicas.

Note that frequently reading from additional replicas can hurt cluster performance. When in doubt, keep the default 99PERCENTILE.

additional_write_policy specifies the threshold at which a cheap quorum write will be upgraded to include transient replicas.

The compaction options must at least define the 'class' sub-option, that defines the compaction strategy class to use. The supported class are 'SizeTieredCompactionStrategy' (STCS), 'LeveledCompactionStrategy' (LCS) and 'TimeWindowCompactionStrategy' (TWCS) (the 'DateTieredCompactionStrategy' is also supported but is deprecated and 'TimeWindowCompactionStrategy' should be preferred instead). The default is 'SizeTieredCompactionStrategy'. Custom strategy can be provided by specifying the full class name as a string constant.

All default strategies support a number of common options, as well as options specific to the strategy chosen (see the section corresponding to your strategy for details: STCS, LCS and TWCS).

The compression options define if and how the sstables of the table are compressed. Compression is configured on a per-table basis as an optional argument to CREATE TABLE or ALTER TABLE. The following sub-options are available:

Option Default Description
class LZ4Compressor The compression algorithm to use. Default compressor are: LZ4Compressor, SnappyCompressor, DeflateCompressor and ZstdCompressor. Use 'enabled' : false to disable compression. Custom compressor can be provided by specifying the full class name as a “string constant”:#constants.
enabled true Enable/disable sstable compression. If the enabled option is set to false no other options must be specified.
chunk_length_in_kb 64

On disk SSTables are compressed by block (to allow random reads). This defines the size (in KB) of said block. Bigger values may improve the compression rate, but increases the minimum size of data to be read from disk for a read. The default value is an optimal value for compressing tables. Chunk length must be a power of 2 because so is assumed so when computing the chunk number from an uncompressed file offset. Block size may be adjusted based on read/write access patterns such as:

  • How much data is typically requested at once
  • Average size of rows in the table
crc_check_chance 1.0 Determines how likely Cassandra is to verify the checksum on each compression chunk during reads.
compression_level 3 Compression level. It is only applicable for ZstdCompressor and accepts values between -131072 and 22.

For instance, to create a table with LZ4Compressor and a chunk_lenth_in_kb of 4KB:

   id int,
   key text,
   value text,
   PRIMARY KEY (key, value)
) with compression = {'class': 'LZ4Compressor', 'chunk_length_in_kb': 4};

Caching optimizes the use of cache memory of a table. The cached data is weighed by size and access frequency. The caching options allows to configure both the key cache and the row cache for the table. The following sub-options are available:

Option Default Description
keys ALL Whether to cache keys (“key cache”) for this table. Valid values are: ALL and NONE.
rows_per_partition NONE The amount of rows to cache per partition (“row cache”). If an integer n is specified, the first n queried rows of a partition will be cached. Other possible options are ALL, to cache all rows of a queried partition, or NONE to disable row caching.

For instance, to create a table with both a key cache and 10 rows per partition:

id int,
key text,
value text,
PRIMARY KEY (key, value)
) WITH caching = {'keys': 'ALL', 'rows_per_partition': 10};

The read_repair options configures the read repair behavior to allow tuning for various performance and consistency behaviors. Two consistency properties are affected by read repair behavior.

  • Monotonic Quorum Reads: Provided by BLOCKING. Monotonic quorum reads prevents reads from appearing to go back in time in some circumstances. When monotonic quorum reads are not provided and a write fails to reach a quorum of replicas, it may be visible in one read, and then disappear in a subsequent read.
  • Write Atomicity: Provided by NONE. Write atomicity prevents reads from returning partially applied writes. Cassandra attempts to provide partition level write atomicity, but since only the data covered by a SELECT statement is repaired by a read repair, read repair can break write atomicity when data is read at a more granular level than it is written. For example read repair can break write atomicity if you write multiple rows to a clustered partition in a batch, but then select a single row by specifying the clustering column in a SELECT statement.

The available read repair settings are:

The default setting. When read_repair is set to BLOCKING, and a read repair is triggered, the read will block on writes sent to other replicas until the CL is reached by the writes. Provides monotonic quorum reads, but not partition level write atomicity

When read_repair is set to NONE, the coordinator will reconcile any differences between replicas, but will not attempt to repair them. Provides partition level write atomicity, but not monotonic quorum reads.

  • Adding new columns (see ALTER TABLE below) is a constant time operation. There is thus no need to try to anticipate future usage when creating a table.


Altering an existing table uses the ALTER TABLE statement:

alter_table_statement   ::=  ALTER TABLE table_name alter_table_instruction
alter_table_instruction ::=  ADD column_name cql_type ( ',' column_name cql_type )*
                             | DROP column_name ( column_name )*
                             | WITH options

For instance:

ALTER TABLE addamsFamily ADD gravesite varchar;

ALTER TABLE addamsFamily
       WITH comment = 'A most excellent and useful table';

The ALTER TABLE statement can:

  • Add new column(s) to the table (through the ADD instruction). Note that the primary key of a table cannot be changed and thus newly added column will, by extension, never be part of the primary key. Also note that compact tables have restrictions regarding column addition. Note that this is constant (in the amount of data the cluster contains) time operation.
  • Remove column(s) from the table. This drops both the column and all its content, but note that while the column becomes immediately unavailable, its content is only removed lazily during compaction. Please also see the warnings below. Due to lazy removal, the altering itself is a constant (in the amount of data removed or contained in the cluster) time operation.
  • Change some of the table options (through the WITH instruction). The supported options are the same that when creating a table (outside of COMPACT STORAGE and CLUSTERING ORDER that cannot be changed after creation). Note that setting any compaction sub-options has the effect of erasing all previous compaction options, so you need to re-specify all the sub-options if you want to keep them. The same note applies to the set of compression sub-options.


Dropping a column assumes that the timestamps used for the value of this column are “real” timestamp in microseconds. Using “real” timestamps in microseconds is the default is and is strongly recommended but as Cassandra allows the client to provide any timestamp on any table it is theoretically possible to use another convention. Please be aware that if you do so, dropping a column will not work correctly.


Once a column is dropped, it is allowed to re-add a column with the same name than the dropped one unless the type of the dropped column was a (non-frozen) column (due to an internal technical limitation).


Dropping a table uses the DROP TABLE statement:

drop_table_statement ::=  DROP TABLE [ IF EXISTS ] table_name

Dropping a table results in the immediate, irreversible removal of the table, including all data it contains.

If the table does not exist, the statement will return an error, unless IF EXISTS is used in which case the operation is a no-op.


A table can be truncated using the TRUNCATE statement:

truncate_statement ::=  TRUNCATE [ TABLE ] table_name

Note that TRUNCATE TABLE foo is allowed for consistency with other DDL statements but tables are the only object that can be truncated currently and so the TABLE keyword can be omitted.

Truncating a table permanently removes all existing data from the table, but without removing the table itself.