Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
tree: 31e669ab42
Fetching contributors…

Cannot retrieve contributors at this time

1068 lines (732 sloc) 62.258 kb

Cassandra Query Language (CQL) v3.0.2


CQL Syntax


This document describes the Cassandra Query Language (CQL) version 3. CQL v3 is not backward compatible with CQL v2 and differs from it in numerous ways. Note that this document describes the last version of the languages. However, the changes section provides the diff between the different versions of CQL v3.

CQL v3 offers a model very close to SQL in the sense that data is put in tables containing rows of columns. For that reason, when used in this document, these terms (tables, rows and columns) have the same definition than they have in SQL. But please note that as such, they do not refer to the concept of rows and columns found in the internal implementation of Cassandra and in the thrift and CQL v2 API.


To aid in specifying the CQL syntax, we will use the following conventions in this document:

  • Language rules will be given in a BNF -like notation:
<start> ::= TERMINAL <non-terminal1> <non-terminal1>
  • Nonterminal symbols will have <angle brackets>.
  • As additional shortcut notations to BNF, we’ll use traditional regular expression’s symbols (?, + and *) to signify that a given symbol is optional and/or can be repeated. We’ll also allow parentheses to group symbols and the [<characters>] notation to represent any one of <characters>.
  • The grammar is provided for documentation purposes and leave some minor details out. For instance, the last column definition in a CREATE TABLE statement is optional but supported if present even though the provided grammar in this document suggest it is not supported.
  • Sample code will be provided in a code block:
SELECT sample_usage FROM cql;
  • References to keywords or pieces of CQL code in running text will be shown in a fixed-width font.

Identifiers and keywords

The CQL language uses identifiers (or names_) to identify tables, columns and other objects. An identifier is a token matching the regular expression @[a-zA-Z][a-zA-Z0-9@*.

A number of such identifiers, like SELECT or WITH, are keywords. They have a fixed meaning for the language and most are reserved. The list of those keywords can be found in Appendix A.

Identifiers and (unquoted) keywords are case insensitive. Thus SELECT is the same than select or sElEcT, and myId is the same than myid or MYID for instance. A convention often used (in particular by the samples of this documentation) is to use upper case for keywords and lower case for other identifiers.

There is a second kind of identifiers called quoted identifiers defined by enclosing an arbitrary sequence of characters in double-quotes("). Quoted identifiers are never keywords. Thus "select" is not a reserved keyword and can be used to refer to a column, while select would raise a parse error. Also, contrarily to unquoted identifiers and keywords, quoted identifiers are case sensitive ("My Quoted Id" is different from "my quoted id"). A fully lowercase quoted identifier that matches [a-zA-Z][a-zA-Z0-9_]* is equivalent to the unquoted identifier obtained by removing the double-quote (so "myid" is equivalent to myid and to myId but different from "myId"). Inside a quoted identifier, the double-quote character can be repeated to escape it, so "foo "" bar" is a valid identifier.


CQL defines the following kind of constants: strings, integers, floats, booleans, uuids and blobs:

  • A string constant is an arbitrary sequence of characters characters enclosed by single-quote('). One can include a single-quote in a string by repeating it, e.g. 'It''s raining today'. Those are not to be confused with quoted identifiers that use double-quotes.
  • An integer constant is defined by '-'?[0-9]+.
  • A float constant is defined by '-'?[0-9]+('.'[0-9]*)?([eE][+-]?[0-9+])?.
  • A boolean constant is either true or false up to case-insensitivity (i.e. True is a valid boolean constant).
  • A UUID constant is defined by hex{8}-hex{4}-hex{4}-hex{4}-hex{12} where hex is an hexadecimal character, e.g. [0-9a-fA-F] and {4} is the number of such characters.
  • A blob constant is an hexadecimal number defined by 0[xX](hex)+ where hex is an hexadecimal character, e.g. [0-9a-fA-F].

For how these constants are typed, see the data types section.


A comment in CQL is a line beginning by either double dashes (--) or double slash (//).

-- This is a comment
// This is a comment too


CQL consists of statements. As in SQL, these statements can be divided in 3 categories:

  • Data definition statements, that allow to set and change the way data is stored.
  • Data manipulation statements, that allow to change data
  • Queries, to look up data

All statements end with a semicolon (;) but that semicolon can be omitted when dealing with a single statement. The supported statements are described in the following sections. When describing the grammar of said statements, we will reuse the non-terminal symbols defined below:

<identifier> ::= any quoted or unquoted identifier, excluding reserved keywords
 <tablename> ::= (<identifier> '.')? <identifier>

    <string> ::= a string constant
   <integer> ::= an integer constant
     <float> ::= a float constant
    <number> ::= <integer> | <float>
      <uuid> ::= a uuid constant
   <boolean> ::= a boolean constant
       <hex> ::= a blob constant

  <constant> ::= <string>
               | <number>
               | <uuid>
               | <boolean>
               | <hex>
  <variable> ::= '?'
      <term> ::= <constant>
               | <collection-literal>
               | <variable>
               | <function> '(' (<term> (',' <term>)*)? ')'

  <collection-literal> ::= <map-literal>
                         | <set-literal>
                         | <list-literal>
         <map-literal> ::= '{' ( <term> ':' <term> ( ',' <term> ':' <term> )* )? '}'
         <set-literal> ::= '{' ( <term> ( ',' <term> )* )? '}'
        <list-literal> ::= '[' ( <term> ( ',' <term> )* )? ']'

    <function> ::= <ident>

  <properties> ::= <property> (AND <property>)*
    <property> ::= <identifier> '=' ( <identifier> | <constant> | <map-literal> )
Please note that not every possible productions of the grammar above will be valid in practice. Most notably, @<variable>@ and nested @<collection-literal>@ are currently not allowed inside @<collection-literal>@.

The question mark (?) of <variable> is a bind variables for prepared statements.

The <properties> production is use by statement that create and alter keyspaces and tables. Each <property> is either a simple one, in which case it just has a value, or a map one, in which case it’s value is a map grouping sub-options. The following will refer to one or the other as the kind (simple or map) of the property.

A <tablename> will be used to identify a table. This is an identifier representing the table name that can be preceded by a keyspace name. The keyspace name, if provided, allow to identify a table in another keyspace than the currently active one (the currently active keyspace is set through the USE statement).

For supported <function>, see the section on functions.

Prepared Statement

CQL supports prepared statements. Prepared statement is an optimization that allows to parse a query only once but execute it multiple times with different concrete values.

In a statement, each time a column value is expected (in the data manipulation and query statements), a bind variable marker (denoted by a ? symbol) can be used instead. A statement with bind variables must then be prepared. Once it has been prepared, it can executed by providing concrete values for the bind variables (values for bind variables must be provided in the order the bind variables are defined in the query string). The exact procedure to prepare a statement and execute a prepared statement depends on the CQL driver used and is beyond the scope of this document.

Data Definition



<create-keyspace-stmt> ::= CREATE KEYSPACE <identifier> WITH <properties>

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

           WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1' : 1, 'DC2' : 3}
            AND durable_writes = false;
The @CREATE KEYSPACE@ statement creates a new top-level _keyspace_. A keyspace is a namespace that defines a replication strategy and some options for a set of tables. Valid keyspaces names are identifiers composed exclusively of alphanumerical characters and whose length is lesser or equal to 32. Note that as identifiers, keyspace names are case insensitive: use a quoted identifier for case sensitive keyspace names.

The supported @<properties>@ for @CREATE KEYSPACE@ are:

|_. name          |_. kind   |_. mandatory |_. default |_. description|
|@replication@    | _map_    | yes         |           | The replication strategy and options to use for the keyspace. |
|@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. It must at least contains the @'class'@ sub-option which defines the replication strategy class to use. The rest of the sub-options depends on that replication strategy class. By default, Cassandra support the following @'class'@:
* @'SimpleStrategy'@: A simple strategy that defines a simple replication factor for the whole cluster. The only sub-options supported is @'replication_factor'@ to define that replication factor and is mandatory.
* @'NetworkTopologyStrategy'@: A 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 each time the key is the name of a datacenter and the value the replication factor for that data-center.
* @'OldNetworkTopologyStrategy'@: A legacy replication strategy. You should avoid this strategy for new keyspaces and prefer @'NetworkTopologyStrategy'@.



<use-stmt> ::= USE <identifier>


USE myApp;

The USE statement takes an existing keyspace name as argument and set it as the per-connection current working keyspace. All subsequent keyspace-specific actions will be performed in the context of the selected keyspace, unless otherwise specified, until another USE statement is issued or the connection terminates.



<create-keyspace-stmt> ::= ALTER KEYSPACE <identifier> WITH <properties>

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

The ALTER KEYSPACE statement alter the properties of an existing keyspace. The supported <properties> are the same that for the CREATE TABLE statement.



 <drop-keyspace-stmt> ::= DROP KEYSPACE <identifier>



A DROP KEYSPACE statement results in the immediate, irreversible removal of an existing keyspace, including all column families in it, and all data contained in those column families.



<create-table-stmt> ::= CREATE (TABLE | COLUMNFAMILY) <tablename>
                          '(' <definition> ( ',' <definition> )* ')'
                          ( WITH <option> ( AND <option>)* )?

<column-definition> ::= <identifier> <type> ( PRIMARY KEY )?
                      | PRIMARY KEY '(' <partition-key> ( ',' <identifier> )* ')'

<partition-key> ::= <partition-key>
                  | '(' <partition-key> ( ',' <identifier> )* ')'

<partition-key> ::= <identifier>
                  | '(' <identifier> (',' <identifier> )* ')'

<option> ::= <property>
           | COMPACT STORAGE

CREATE TABLE monkeySpecies (
    species text PRIMARY KEY,
    common_name text,
    population varint,
    average_size int
) WITH comment='Important biological records'
   AND read_repair_chance = 1.0;

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' };
The @CREATE TABLE@ statement creates a new table. Each such table is a set of _rows_ (usually representing related entities) for which it defines a number of properties. A table is defined by a "name":#createTableName, it defines the <a href="#createTableColumn"><it>columns</it></a> composing rows of the table and have a number of "options":#createTableOptions. Note that the @CREATE COLUMNFAMILY@ syntax is supported as an alias for @CREATE TABLE@ (for historical reasons).


Valid table names are the same than valid keyspace names (up to 32 characters long alphanumerical identifiers). If the table name is provided alone, the table is created within the current keyspace (see USE), but if it is prefixed by an existing keyspace name (see <tablename> grammar), it is created in the specified keyspace (but does not change the current keyspace).


A CREATE TABLE statement defines the columns that rows of the table can have. A column is defined by its name (an identifier) and its type (see the data types section for more details on allowed types and their properties).

Within a table, a row is uniquely identified by its PRIMARY KEY (or more simply the key), and hence all table definitions must define a PRIMARY KEY (and only one). A PRIMARY KEY is composed of one or more of the columns defined in the table. If the PRIMARY KEY is only one column, this can be specified directly after the column definition. Otherwise, it must be specified by following PRIMARY KEY by the comma-separated list of column names composing the key within parenthesis. Note that:

    k int PRIMARY KEY,
    other text

is equivalent to

    k int,
    other text,

Moreover, a table must define at least one column that is not part of the PRIMARY KEY as a row exists in Cassandra only if it contains at least one value for one such column.

Partition key and clustering

In CQL, the order in which columns are defined for the PRIMARY KEY matters. The first column of the key is called the partition key. It has the property that all the rows sharing the same partition key (even across table in fact) are stored on the same physical node. Also, insertion/update/deletion on rows sharing the same partition key for a given table are performed atomically and in isolation. Note that it is possible to have a composite partition key, i.e. a partition key formed of multiple columns, using an extra set of parentheses to define which columns forms the partition key.

The remaining columns of the PRIMARY KEY definition, if any, are called clustering keys. On a given physical node, rows for a given partition key are stored in the order induced by the clustering keys, making the retrieval of rows in that clustering order particularly efficient (see SELECT).


The CREATE TABLE statement supports a number of options that controls the configuration of a new table. These options can be specified after the WITH keyword.

The first of these option is COMPACT STORAGE. This option is meanly targeted towards backward compatibility with some table definition created before CQL3. But it also provides a slightly more compact layout of data on disk, though at the price of flexibility and extensibility, and for that reason is not recommended unless for the backward compatibility reason. The restriction for table with COMPACT STORAGE is that they support one and only one column outside of the ones part of the PRIMARY KEY. It also follows that columns cannot be added nor removed after creation. A table with COMPACT STORAGE must also define at least one clustering key.

Another option is CLUSTERING ORDER. It allows to define the ordering of rows on disk. It takes the list of the clustering key names with, for each of them, the on-disk order (Ascending or descending). Note that this option affects what ORDER BY are allowed during SELECT.

Table creation supports the following other <property>:

option kind default description
comment simple none A free-form, human-readable comment.
read_repair_chance simple 0.1 The probability with which to query extra nodes (e.g. more nodes than required by the consistency level) for the purpose of read repairs.
dclocal_read_repair_chance simple 0 The probability with which to query extra nodes (e.g. more nodes than required by the consistency level) belonging to the same data center than the read coordinator for the purpose of read repairs.
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)
compaction map see below The compaction otpions to use, see below.
compression map see below Compression options, see below.
replicate_on_write simple true Whether to replicate data on write. This can only be set to false for tables with counters values. Disabling this is dangerous and can result in random lose of counters, don’t disable unless you are sure to know what you are doing
caching simple keys_only Whether to cache keys (“key cache”) and/or rows (“row cache”) for this table. Valid values are: all, keys_only, rows_only and none.

compaction options

The compaction property must at least define the 'class' sub-option, that defines the compaction strategy class to use. The default supported class are 'SizeTieredCompactionStrategy' and 'LeveledCompactionStrategy'. Custom strategy can be provided by specifying the full class name as a string constant. The rest of the sub-options depends on the chosen class. The sub-options supported by the default classes are:

option supported compaction strategy default description
tombstone_threshold all 0.2 A ratio such that if a sstable has more than this ratio of gcable tombstones over all contained columns, the sstable will be compacted (with no other sstables) for the purpose of purging those tombstones.
tombstone_compaction_interval all 1 day The mininum time to wait after an sstable creation time before considering it for “tombstone compaction”, where “tombstone compaction” is the compaction triggered if the sstable has more gcable tombstones than tombstone_threshold.
min_sstable_size SizeTieredCompactionStrategy 50MB The size tiered strategy groups SSTables to compact in buckets. A bucket groups SSTables that differs from less than 50% in size. However, for small sizes, this would result in a bucketing that is too fine grained. min_sstable_size defines a size threshold (in bytes) below which all SSTables belong to one unique bucket
min_threshold SizeTieredCompactionStrategy 4 Minimum number of SSTables needed to start a minor compaction.
max_threshold SizeTieredCompactionStrategy 32 Maximum number of SSTables processed by one minor compaction.
bucket_low SizeTieredCompactionStrategy 0.5 Size tiered consider sstables to be within the same bucket if their size is within [average_size * bucket_low, average_size * bucket_high ] (i.e the default groups sstable whose sizes diverges by at most 50%)
bucket_high SizeTieredCompactionStrategy 1.5 Size tiered consider sstables to be within the same bucket if their size is within [average_size * bucket_low, average_size * bucket_high ] (i.e the default groups sstable whose sizes diverges by at most 50%).
sstable_size_in_mb LeveledCompactionStrategy 5MB The target size (in MB) for sstables in the leveled strategy. Note that while sstable sizes should stay less or equal to sstable_size_in_mb, it is possible to exceptionally have a larger sstable as during compaction, data for a given partition key are never split into 2 sstables

For the compression property, the following default sub-options are available:

option default description
sstable_compression SnappyCompressor The compression algorithm to use. Default compressor are: SnappyCompressor and DeflateCompressor. Use an empty string ('') to disable compression. Custom compressor can be provided by specifying the full class name as a string constant.
chunk_length_kb 64KB 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
crc_check_chance 1.0 When compression is enabled, each compressed block includes a checksum of that block for the purpose of detecting disk bitrot and avoiding the propagation of corruption to other replica. This option defines the probability with which those checksums are checked during read. By default they are always checked. Set to 0 to disable checksum checking and to 0.5 for instance to check them every other read

Other considerations:

  • When insertingupdating a given row, not all columns needs to be defined (except for those part of the key), and missing columns occupy no space on disk. Furthermore, adding new columns (see ALTER TABLE) is a constant time operation. There is thus no need to try to anticipate future usage (or to cry when you haven’t) when creating a table.



<alter-table-stmt> ::= ALTER (TABLE | COLUMNFAMILY) <tablename> <instruction>

<instruction> ::= ALTER <identifier> TYPE <type>
                | ADD   <identifier> <type>
                | WITH  <option> ( AND <option> )*

ALTER TABLE addamsFamily
ALTER lastKnownLocation TYPE uuid;

ALTER TABLE addamsFamily
ADD gravesite varchar;

ALTER TABLE addamsFamily
WITH comment = 'A most excellent and useful column family'
 AND read_repair_chance = 0.2;
The @ALTER@ statement is used to manipulate table definitions. It allows to add new columns, drop existing ones, change the type of existing columns, or update the table options. As for table creation, @ALTER COLUMNFAMILY@ is allowed as an alias for @ALTER TABLE@.

The @<tablename>@ is the table name optionally preceded by the keyspace name.  The @<instruction>@ defines the alteration to perform:
* @ALTER@: Update the type of a given defined column. Note that the type of the "clustering keys":#createTablepartitionClustering cannot be modified as it induces the on-disk ordering of rows. Columns on which a "secondary index":#createIndexStmt is defined have the same restriction. Other columns are free from those restrictions (no validation of existing data is performed), but it is usually a bad idea to change the type to a non-compatible one, unless no data have been inserted for that column yet, as this could confuse CQL drivers/tools.
* @ADD@: Adds a new column to the table. The @<identifier>@ for the new column must not conflict with an existing column. Moreover, columns cannot be added to tables defined with the @COMPACT STORAGE@ option.
* @WITH@: Allows to update the options of the table. The "supported @<option>@":#createTableOptions (and syntax) are the same as for the @CREATE TABLE@ statement except that @COMPACT STORAGE@ is not supported. 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 is no yet supported but is on "the roadmap": In the meantime, a declared but unused column has no impact on performance nor uses any storage.



<drop-table-stmt> ::= DROP TABLE <tablename>


DROP TABLE worldSeriesAttendees;

The DROP TABLE statement results in the immediate, irreversible removal of a table, including all data contained in it. As for table creation, DROP COLUMNFAMILY is allowed as an alias for DROP TABLE.



<truncate-stmt> ::= TRUNCATE <tablename>


TRUNCATE superImportantData;

The TRUNCATE statement permanently removes all data from a table.



<create-index-stmt> ::= CREATE INDEX <identifier>? ON <tablename> '(' <identifier> ')'


CREATE INDEX userIndex ON NerdMovies (user);
CREATE INDEX ON Mutants (abilityId);

The CREATE INDEX statement is used to create a new (automatic) secondary index for a given (existing) column in a given table. A name for the index itself can be specified before the ON keyword, if desired. If data already exists for the column, it will be indexed during the execution of this statement. After the index is created, new data for the column is indexed automatically at insertion time.



 <drop-index-stmt> ::= DROP INDEX <identifier>


DROP INDEX userIndex;

The DROP INDEX statement is used to drop an existing secondary index. The argument of the statement is the index name.

Data Manipulation



<insertStatement> ::= INSERT INTO <tablename>
                             '(' <identifier> ( ',' <identifier> )* ')'
                      VALUES '(' <term-or-literal> ( ',' <term-or-literal> )* ')'
                      ( USING <option> ( AND <option> )* )?

<term-or-literal> ::= <term>
                    | <collection-literal>

<option> ::= TIMESTAMP <integer>
           | TTL <integer>

INSERT INTO NerdMovies (movie, director, main_actor, year)
                VALUES ('Serenity', 'Joss Whedon', 'Nathan Fillion', 2005)
USING TTL 86400;

The INSERT statement writes one or more columns for a given row in a table. Note that since a row is identified by its PRIMARY KEY, the columns that compose it must be specified. Also, since a row only exists when it contains one value for a column not part of the PRIMARY KEY, one such value must be specified too.

Note that unlike in SQL, INSERT does not check the prior existence of the row: the row is created if none existed before, and updated otherwise. Furthermore, there is no mean to know which of creation or update happened. In fact, the semantic of INSERT and UPDATE are identical.

All updates for an INSERT are applied atomically and in isolation.

Please refer to the UPDATE section for information on the <option> available and to the collections section for use of <collection-literal>. Also note that INSERT does not support counters, while UPDATE does.



<update-stmt> ::= UPDATE <tablename>
                  ( USING <option> ( AND <option> )* )?
                  SET <assignment> ( ',' <assignment> )*
                  WHERE <where-clause>

<assignment> ::= <identifier> '=' <term>
               | <identifier> '=' <identifier> ('+' | '-') (<int-term> | <set-literal> | <list-literal>)
               | <identifier> '=' <identifier> '+' <map-literal>
               | <identifier> '[' <term> ']' '=' <term>

<where-clause> ::= <identifier> '=' <term>
                 | <identifier> IN '(' <term> ( ',' <term> )* ')'

<option> ::= TIMESTAMP <integer>
           | TTL <integer>

SET director = 'Joss Whedon',
    main_actor = 'Nathan Fillion',
    year = 2005
WHERE movie = 'Serenity';

UPDATE UserActions SET total = total + 2 WHERE user = B70DE1D0-9908-4AE3-BE34-5573E5B09F14 AND action = 'click';
The @UPDATE@ statement writes one or more columns for a given row in a table. The @<where-clause>@ is used to select the row to update and must include all columns composing the @PRIMARY KEY@. Other columns values are specified through @<assignment>@ after the @SET@ keyword.

Note that unlike in SQL, @UPDATE@ does not check the prior existence of the row: the row is created if none existed before, and updated otherwise. Furthermore, there is no mean to know which of creation or update happened. In fact, the semantic of @INSERT@ and @UPDATE@ are identical.

In an @UPDATE@ statement, all updates within the same partition key are applied atomically and in isolation.

The @c = c + 3@ form of @<assignment>@ is used to increment/decrement counters. The identifier after the '=' sign *must* be the same than the one before the '=' sign (Only increment/decrement is supported on counters, not the assignment of a specific value).

The @id = id + <collection-literal>@ and @id[value1] = value2@ forms of @<assignment>@ are for collections. Please refer to the "relevant section":#collections for more details.


The UPDATE and INSERT statements allows to specify the following options for the insertion:

  • TIMESTAMP: sets the timestamp for the operation. If not specified, the current time of the insertion (in microseconds) is used. This is usually a suitable default.
  • TTL: allows to specify an optional Time To Live (in seconds) for the inserted values. If set, the inserted values are automatically removed from the database after the specified time. Note that the TTL concerns the inserted values, not the column themselves. This means that any subsequent update of the column will also reset the TTL (to whatever TTL is specified in that update). By default, values never expire.



<delete-stmt> ::= DELETE ( <selection> ( ',' <selection> )* )?
                  FROM <tablename>
                  ( USING TIMESTAMP <integer>)?
                  WHERE <where-clause>

<selection> ::= <identifier> ( '[' <term> ']' )?

<where-clause> ::= <identifier> '=' <term>
                 | <identifier> IN '(' <term> ( ',' <term> )* ')'

DELETE FROM NerdMovies USING TIMESTAMP 1240003134 WHERE movie = 'Serenity';

DELETE phone FROM Users WHERE userid IN (C73DE1D3-AF08-40F3-B124-3FF3E5109F22, B70DE1D0-9908-4AE3-BE34-5573E5B09F14);
The @DELETE@ statement deletes columns and rows. If column names are provided directly after the @DELETE@ keyword, only those columns are deleted from the row indicated by the @<where-clause>@ (the @id[value]@ syntax in @<selection>@ is for collection, please refer to the "collection section":#collections for more details).  Otherwise whole rows are removed. The @<where-clause>@ allows to specify the key for the row(s) to delete.

@DELETE@ supports the @TIMESTAMP@ options with the same semantic that in the "@UPDATE@":#updateStmt statement.

In a @DELETE@ statement, all deletions within the same partition key are applied atomically and in isolation.



<batch-stmt> ::= BEGIN BATCH
                 ( USING <option> ( AND <option> )* )?
                    <modification-stmt> ( ';' <modification-stmt> )*
                 APPLY BATCH

<modification-stmt> ::= <insert-stmt>
                      | <update-stmt>
                      | <delete-stmt>

<option> ::= TIMESTAMP <integer>

  INSERT INTO users (userid, password, name) VALUES ('user2', 'ch@ngem3b', 'second user');
  UPDATE users SET password = 'ps22dhds' WHERE userid = 'user3';
  INSERT INTO users (userid, password) VALUES ('user4', 'ch@ngem3c');
  DELETE name FROM users WHERE userid = 'user1';

The BATCH statement group multiple modification statements (insertions/updates and deletions) into a single statement. It mainly serves two purposes:

  1. it saves network round-trips between the client and the server (and sometimes between the server coordinator and the replicas) when batching multiple updates.
  2. all updates in a BATCH belonging to a given partition key are performed atomically and in isolation
    Note however that the BATCH statement only allows UPDATE, INSERT and DELETE statements and is not a full analogue for SQL transactions.


BATCH supports both the TIMESTAMP option, with similar semantic to the one described in the UPDATE statement (the timestamp applies to all the statement inside the batch). However, if used, TIMESTAMP must not be used in the statements within the batch.




<select-stmt> ::= SELECT <select-clause>
                  FROM <tablename>
                  ( WHERE <where-clause> )?
                  ( ORDER BY <order-by> )?
                  ( LIMIT <integer> )?
                  ( ALLOW FILTERING )?

<select-clause> ::= <selection-list>
                  | COUNT '(' ( '*' | '1' ) ')'

<selection-list> ::= <selector> ( ',' <selector> )*
                   | '*'

<selector> ::= <identifier>
             | WRITETIME '(' <identifier> ')'
             | TTL '(' <identifier> ')'
             | <function> '(' (<selector> (',' <selector>)*)? ')'

<where-clause> ::= <relation> ( "AND" <relation> )*

<relation> ::= <identifier> ("=" | "<" | ">" | "<=" | ">=") <term>
             | <identifier> IN '(' <term> ( ',' <term>)* ')'
             | TOKEN '(' <identifier> ')' ("=" | "<" | ">" | "<=" | ">=") (<term> | TOKEN '( <term> ')' )

<order-by> ::= <ordering> ( ',' <odering> )*
<ordering> ::= <identifer> ( ASC | DESC )?

SELECT name, occupation FROM users WHERE userid IN (199, 200, 207);

SELECT time, value
FROM events
WHERE event_type = 'myEvent'
  AND time > 2011-02-03
  AND time <= 2012-01-01

The @SELECT@ statements reads one or more columns for one or more rows in a table. It returns a result-set of rows, where each row contains the collection of columns corresponding to the query.


The <select-clause> determines which columns needs to be queried and returned in the result-set. It consists of either the comma-separated list of or the wildcard character (*) to select all the columns defined for the table.

A <selector> is either a column name to retrieve, or a <function> of one or multiple column names. The functions allows are the same that for <term> and are describe in the function section. In addition to these generic functions, the WRITETIME (resp. TTL) function allows to select the timestamp of when the column was inserted (resp. the time to live (in seconds) for the column (or null if the column has no expiration set)).

The COUNT keyword can be used with parenthesis enclosing *. If so, the query will return a single result: the number of rows matching the query. Note that COUNT(1) is supported as an alias.


The <where-clause> specifies which rows must be queried. It is composed of relations on the columns that are part of the PRIMARY KEY and/or have a secondary index defined on them.

Not all relations are allowed in a query. For instance, non-equal relations (where IN is considered as an equal relation) on a partition key is only supported if the partitioner for the keyspace is an ordered one. Moreover, for a given partition key, the clustering keys induce an ordering of rows and relations on them is restricted to the relations that allow to select a contiguous (for the ordering) set of rows. For instance, given

    userid text,
    blog_title text,
    posted_at timestamp,
    entry_title text,
    content text,
    category int,
    PRIMARY KEY (userid, blog_title, posted_at)

The following query is allowed:

SELECT entry_title, content FROM posts WHERE userid='john doe' AND blog_title='John's Blog' AND posted_at >= 2012-01-01 AND posted_at < 2012-01-31

But the following one is not, as it does not select a contiguous set of rows (and we suppose no secondary indexes are set):

// Needs a blog_title to be set to select ranges of posted_at
SELECT entry_title, content FROM posts WHERE userid='john doe' AND posted_at >= 2012-01-01 AND posted_at < 2012-01-31

When specifying relations, the TOKEN function can be used on the PARTITION KEY column to query. In that case, rows will be selected based on the token of their PARTITION_KEY rather than on the value (note that the token of a key depends on the partitioner in use, and that in particular the RandomPartitioner won’t yeld a meaningful order). Example:

SELECT * FROM posts WHERE token(userid) > token('tom') AND token(userid) < token('bob')


The ORDER BY option allows to select the order of the returned results. It takes as argument a list of column names along with the order for the column (ASC for ascendant and DESC for descendant, omitting the order being equivalent to ASC). Currently the possible orderings are limited (which depends on the table CLUSTERING ORDER):

  • if the table has been defined without any specific CLUSTERING ORDER, then then allowed orderings are the order induced by the clustering key and the reverse of that one.
  • otherwise, the orderings allowed are the order of the CLUSTERING ORDER option and the reversed one.


The LIMIT option to a SELECT statement limits the number of rows returned by a query.


By default, CQL only allows select queries that don’t involve “filtering” server side, i.e. queries where we know that all (live) record read will be returned (maybe partly) in the result set. The reasoning is that those “non filtering” queries have predictable performance in the sense that they will execute in a time that is proportional to the amount of data returned by the query (which can be controlled through LIMIT).

The ALLOW FILTERING option allows to explicitely allow (some) queries that require filtering. Please note that a query using ALLOW FILTERING may thus have unpredictable performance (for the definition above), i.e. even a query that selects a handful of records may exhibit performance that depends on the total amount of data stored in the cluster.

For instance, considering the following table holding user profiles with their year of birth (with a secondary index on it) and country of residence:

    username text PRIMARY KEY,
    firstname text,
    lastname text,
    birth_year int,
    country text

CREATE INDEX ON users(birth_year);

Then the following queries are valid:

SELECT * FROM users;
SELECT firstname, lastname FROM users WHERE birth_year = 1981;

because in both case, Cassandra guarantees that these queries performance will be proportional to the amount of data returned. In particular, if no users are born in 1981, then the second query performance will not depend of the number of user profile stored in the database (not directly at least: due to 2ndary index implementation consideration, this query may still depend on the number of node in the cluster, which indirectly depends on the amount of data stored. Nevertheless, the number of nodes will always be multiple number of magnitude lower than the number of user profile stored). Of course, both query may return very large result set in practice, but the amount of data returned can always be controlled by adding a LIMIT.

However, the following query will be rejected:

SELECT firstname, lastname FROM users WHERE birth_year = 1981 AND country = 'FR';

because Cassandra cannot guarantee that it won’t have to scan large amount of data even if the result to those query is small. Typically, it will scan all the index entries for users born in 1981 even if only a handful are actually from France. However, if you “know what you are doing”, you can force the execution of this query by using ALLOW FILTERING and so the following query is valid:

SELECT firstname, lastname FROM users WHERE birth_year = 1981 AND country = 'FR' ALLOW FILTERING;

Data Types

CQL supports a rich set of data types for columns defined in a table, including collection types. On top of those native and collection types, users can also provide custom types (through a JAVA class extending AbstractType loadable by Cassandra). The syntax of types is thus:

<type> ::= <native-type>
         | <collection-type>
         | <string>       // Used for custom types. The fully-qualified name of a JAVA class

<native-type> ::= ascii
                | bigint
                | blob
                | boolean
                | counter
                | decimal
                | double
                | float
                | inet
                | int
                | text
                | timestamp
                | timeuuid
                | uuid
                | varchar
                | varint

<collection-type> ::= list '<' <native-type> '>'
                    | set  '<' <native-type> '>'
                    | map  '<' <native-type> ',' <native-type> '>'
p. Note that the native types are keywords and as such are case-insensitive. They are however not reserved ones.

The following table gives additional informations on the native data types, and on which kind of constants each type supports:

type constants supported description
ascii strings ASCII character string
bigint integers 64-bit signed long
blob blobs Arbitrary bytes (no validation)
boolean booleans true or false
counter integers Counter column (64-bit signed value). See Counters for details
decimal integers, floats Variable-precision decimal
double integers 64-bit IEEE-754 floating point
float integers, floats 32-bit IEEE-754 floating point
inet strings An IP address. It can be either 4 bytes long (IPv4) or 16 bytes long (IPv6). There is no inet constant, IP address should be inputed as strings
int integers 32-bit signed int
text strings UTF8 encoded string
timestamp integers, strings A timestamp. Strings constant are allow to input timestamps as dates, see Working with dates below for more information.
timeuuid uuids Type 1 UUID. This is generally used as a “conflict-free” timestamp. Also see the functions on Timeuuid
uuid uuids Type 1 or type 4 UUID
varchar strings UTF8 encoded string
varint integers Arbitrary-precision integer

For more information on how to use the collection types, see the Working with collections section below.

Working with dates

Values of the timestamp type are encoded as 64-bit signed integers representing a number of milliseconds since the standard base time known as “the epoch”: January 1 1970 at 00:00:00 GMT.

Timestamp can be input in CQL as simple long integers, giving the number of milliseconds since the epoch, as defined above.

They can also be input as string literals in any of the following ISO 8601 formats, each representing the time and date Mar 2, 2011, at 04:05:00 AM, GMT.:

  • 2011-02-03 04:05+0000
  • 2011-02-03 04:05:00+0000
  • 2011-02-03T04:05+0000
  • 2011-02-03T04:05:00+0000

The +0000 above is an RFC 822 4-digit time zone specification; +0000 refers to GMT. US Pacific Standard Time is -0800. The time zone may be omitted if desired— the date will be interpreted as being in the time zone under which the coordinating Cassandra node is configured.

  • 2011-02-03 04:05
  • 2011-02-03 04:05:00
  • 2011-02-03T04:05
  • 2011-02-03T04:05:00

There are clear difficulties inherent in relying on the time zone configuration being as expected, though, so it is recommended that the time zone always be specified for timestamps when feasible.

The time of day may also be omitted, if the date is the only piece that matters:

  • 2011-02-03
  • 2011-02-03+0000

In that case, the time of day will default to 00:00:00, in the specified or default time zone.


The counter type is used to define counter columns. A counter column is a column whose value is a 64-bit signed integer and on which 2 operations are supported: incrementation and decrementation (see UPDATE for syntax). Note the value of a counter cannot be set. A counter doesn’t exist until first incremented/decremented, and the first incrementation/decrementation is made as if the previous value was 0. Deletion of counter columns is supported but have some limitations (see the Cassandra Wiki for more information).

The use of the counter type is limited in the following way:

  • It cannot be used for column that is part of the PRIMARY KEY of a table.
  • A table that contains a counter can only contain counters. In other words, either all the columns of a table outside the PRIMARY KEY have the counter type, or none of them have it.

Working with collections


A map is a typed set of key-value pairs, where keys are unique. Furthermore, note that the map are internally sorted by their keys and will thus always be returned in that order. To create a column of type map, use the map keyword suffixed with comma-separated key and value types, enclosed in angle brackets. For example:

    id text PRIMARY KEY,
    given text,
    surname text,
    favs map<text, text>   // A map of text keys, and text values

Writing map data is accomplished with a JSON-inspired syntax. To write a record using INSERT, specify the entire map as a JSON-style associative array. Note: This form will always replace the entire map.

// Inserting (or Updating)
INSERT INTO users (id, given, surname, favs)
           VALUES ('jsmith', 'John', 'Smith', { 'fruit' : 'apple', 'band' : 'Beatles' })

Adding or updating key-values of a (potentially) existing map can be accomplished either by subscripting the map column in an UPDATE statement or by adding a new map literal:

// Updating (or inserting)
UPDATE users SET favs['author'] = 'Ed Poe' WHERE id = 'jsmith'
UPDATE users SET favs = favs +  { 'movie' : 'Cassablanca' } WHERE id = 'jsmith'

Note that TTLs are allowed for both INSERT and UPDATE, but in both case the TTL set only apply to the newly inserted/updated values. In other words,

// Updating (or inserting)
UPDATE users USING TTL 10 SET favs['color'] = 'green' WHERE id = 'jsmith'

will only apply the TTL to the { 'color' : 'green' } record, the rest of the map remaining unaffected.

Deleting a map record is done with:

DELETE favs['author'] FROM plays WHERE id = 'jsmith'


A set is a typed collection of unique values. Sets are ordered by their values. To create a column of type set, use the set keyword suffixed with the value type enclosed in angle brackets. For example:

    name text PRIMARY KEY,
    owner text,
    date timestamp,
    tags set<text>

Writing a set is accomplished by comma separating the set values, and enclosing them in curly braces. Note: An INSERT will always replace the entire set.

INSERT INTO images (name, owner, date, tags)
            VALUES ('cat.jpg', 'jsmith', 'now', { 'kitten', 'cat', 'pet' });

Adding and removing values of a set can be accomplished with an UPDATE by adding/removing new set values to an existing set column.

UPDATE images SET tags = tags + { 'cute', 'cuddly' } WHERE name = 'cat.jpg';
UPDATE images SET tags = tags - { 'lame' } WHERE name = 'cat.jpg';

As with maps, TTLs if used only apply to the newly inserted/updated values.


A list is a typed collection of non-unique values where elements are ordered by there position in the list. To create a column of type list, use the list keyword suffixed with the value type enclosed in angle brackets. For example:

    id text PRIMARY KEY,
    game text,
    players int,
    scores list<int>

Do note that as explained below, lists have some limitations and performance considerations to take into account, and it is advised to prefer sets over lists when this is possible.

Writing list data is accomplished with a JSON-style syntax. To write a record using INSERT, specify the entire list as a JSON array. Note: An INSERT will always replace the entire list.

INSERT INTO plays (id, game, players, scores)
           VALUES ('123-afde', 'quake', 3, [17, 4, 2]);

Adding (appending or prepending) values to a list can be accomplished by adding a new JSON-style array to an existing list column.

UPDATE plays SET players = 5, scores = scores + [ 14, 21 ] WHERE id = '123-afde';
UPDATE plays SET players = 5, scores = [ 12 ] + scores WHERE id = '123-afde';

It should be noted that append and prepend are not idempotent operations. This means that if during an append or a prepend the operation timeout, it is not always safe to retry the operation (as this could result in the record appended or prepended twice).

Lists also provides the following operation: setting an element by its position in the list, removing an element by its position in the list and remove all the occurrence of a given value in the list. However, and contrarily to all the other collection operations, these three operations induce an internal read before the update, and will thus typically have slower performance characteristics. Those operations have the following syntax:

UPDATE plays SET scores[1] = 7 WHERE id = '123-afde';                // sets the 2nd element of scores to 7 (raises an error is scores has less than 2 elements)
DELETE scores[1] FROM plays WHERE id = '123-afde';                   // deletes the 2nd element of scores (raises an error is scores has less than 2 elements)
UPDATE plays SET scores = scores - [ 12, 21 ] WHERE id = '123-afde'; // removes all occurences of 12 and 21 from scores

As with maps, TTLs if used only apply to the newly inserted/updated values.


CQL3 supports a few functions (more to come). Currently, it only support functions on values (functions that transform one or more column values into a new value) and in particular aggregation functions are not supported. The functions supported are described below:


The token function allows to compute the token for a given partition key. The exact signature of the token function depends on the table concerned and of the partitioner used by the cluster.

The type of the arguments of the token depend on the type of the partition key columns. The return type depend on the partitioner in use:

  • For Murmur3Partitioner, the return type is bigint.
  • For RandomPartitioner, the return type is varint.
  • For ByteOrderedPartitioner, the return type is blob.

For instance, in a cluster using the default Murmur3Partitioner, if a table is defined by

    userid text PRIMARY KEY,
    username text,

then the token function will take a single argument of type text (in that case, the partition key is userid (there is no clustering key so the partition key is the same than the primary key)), and the return type will be bigint.

Timeuuid functions


The now function takes no arguments and generates a new unique timeuuid (at the time where the statement using it is executed). Note that this method is useful for insertion but is largely non-sensical in WHERE clauses. For instance, a query of the form

SELECT * FROM myTable WHERE t = now()

will never return any result by design, since the value returned by now() is guaranteed to be unique.

minTimeuuid and maxTimeuuid

The minTimeuuid (resp. maxTimeuuid) function takes a timestamp value t (which can be either a timestamp or a date string) and return a fake timeuuid corresponding to the smallest (resp. biggest) possible timeuuid having for timestamp t. So for instance:

SELECT * FROM myTable WHERE t > maxTimeuuid('2013-01-01 00:05+0000') AND t < minTimeuuid('2013-02-02 10:00+0000')

will select all rows where the timeuuid column t is strictly older than ‘2013-01-01 00:05+0000’ but stricly younger than ‘2013-02-02 10:00+0000’. Please note that t >= maxTimeuuid('2013-01-01 00:05+0000') would still not select a timeuuid generated exactly at ‘2013-01-01 00:05+0000’ and is essentially equivalent to t > maxTimeuuid('2013-01-01 00:05+0000').

Warning: We called the values generated by minTimeuuid and maxTimeuuid fake UUID because they do no respect the Time-Based UUID generation process specified by the RFC 4122. In particular, the value returned by these 2 methods will not be unique. This means you should only use those methods for querying (as in the example above). Inserting the result of those methods is almost certainly a bad idea.

dateOf and unixTimestampOf

The dateOf and unixTimestampOf functions take a timeuuid argument and extract the embeded timestamp. However, while the dateof function return it with the timestamp type (that most client, including cqlsh, interpret as a date), the unixTimestampOf function returns it as a bigint raw value.

Blob conversion functions

A number of functions are provided to “convert” the native types into binary data (blob). For every <native-type> type supported by CQL3 (a notable exceptions is blob, for obvious reasons), the function typeAsBlob takes a argument of type type and return it as a blob. Conversely, the function blobAsType takes a 64-bit blob argument and convert it to a bigint value. And so for instance, bigintAsBlob(3) is 0x0000000000000003 and blobAsBigint(0x0000000000000003) is 3.

Appendix A: CQL Keywords

CQL distinguishes between reserved and non-reserved keywords. Reserved keywords cannot be used as identifier, they are truly reserved for the language (but one can enclose a reserved keyword by double-quotes to use it as an identifier). Non-reserved keywords however only have a specific meaning in certain context but can used as identifer otherwise. The only raison d’être of these non-reserved keywords is convenience: some keyword are non-reserved when it was always easy for the parser to decide whether they were used as keywords or not.

Keyword Reserved?
ADD yes
ALL no
AND yes
ANY yes
ASC yes
BY yes
DESC yes
DROP yes
FROM yes
IN yes
INT no
INTO yes
KEY no
OF yes
ON yes
ONE yes
SET yes
TTL no
TWO yes
USE yes
WITH yes


The following describes the addition/changes brought for each version of CQL.


  • Type validation for the constants has been fixed. For instance, the implementation used to allow '2' as a valid value for an int column (interpreting it has the equivalent of 2), or 42 as a valid blob value (in which case 42 was interpreted as an hexadecimal representation of the blob). This is no longer the case, type validation of constants is now more strict. See the data types section for details on which constant is allowed for which type.
  • The type validation fixed of the previous point has lead to the introduction of blobs constants to allow inputing blobs. Do note that while inputing blobs as strings constant is still supported by this version (to allow smoother transition to blob constant), it is now deprecated (in particular the data types section does not list strings constants as valid blobs) and will be removed by a future version. If you were using strings as blobs, you should thus update your client code asap to switch blob constants.
  • A number of functions to convert native types to blobs have also been introduced. Furthermore the token function is now also allowed in select clauses. See the section on functions for details.


  • Date strings (and timestamps) are no longer accepted as valid timeuuid values. Doing so was a bug in the sense that date string are not valid timeuuid, and it was thus resulting in confusing behaviors. However, the following new methods have been added to help working with timeuuid: now, minTimeuuid, maxTimeuuid , dateOf and unixTimestampOf. See the section dedicated to these methods for more detail.
  • “Float constants”#constants now support the exponent notation. In other words, 4.2E10 is now a valid floating point value.


Versioning of the CQL language adheres to the Semantic Versioning guidelines. Versions take the form X.Y.Z where X, Y, and Z are integer values representing major, minor, and patch level respectively. There is no correlation between Cassandra release versions and the CQL language version.

version description
Major The major version must be bumped when backward incompatible changes are introduced. This should rarely occur.
Minor Minor version increments occur when new, but backward compatible, functionality is introduced.
Patch The patch version is incremented when bugs are fixed.
Jump to Line
Something went wrong with that request. Please try again.