The CL-postgres module implements a rather low-level interface for communicating with a PostgreSQL database server. It is part of the Postmodern library, but can be used separately.
Objects of this type represent database connections.
open-database (database user password host &optional (port 5432) (use-ssl :no))
Create and open a connection for the specified
server, database, and user.
use-ssl may be
:try means 'if the server supports it'. When it is
:no, you must have the CL+SSL
package loaded to initiate the connection.
On SBCL and Clozure CL, the value
:unix may be passed for
host, in order
to connect using a Unix domain socket instead of a TCP socket.
function close-database (database-connection)
Close a database connection. It is advisable to call this on connections when you are done with them. Otherwise the open socket will stick around until it is garbage collected, and no one will tell the database server that we are done with it.
function reopen-database (database-connection)
Re-establish a database connection for a previously closed connection object. (Calling this on a connection that is still open is harmless.)
Test whether a database connection is still open.
This method provides access to a hash table that is associated with the current database connection. When the connection is closed and re-opened this hash table is reset. The most obvious use for this is for storing information about the prepared statements that have been parsed for this connection.
This method returns a mapping (string to string) containing all the configuration parameters for the connection.
On SBCL, when using the
as host argument when creating a connection, this variable
determines the directory in which CL-Postgres will look for the
When using SSL (see
open-database), these can
be used to provide client key and certificate files. They can be
NIL, for no file, or a pathname.
function wait-for-notification (database-connection)
This function blocks until a notification is
received on the connection. The PostgreSQL
command must be used to enable listening for notifications.
exec-query (database-connection query &optional (row-reader '
Sends the given query to the given connection, and interprets the results (if there are any) with the given row-reader. If the database returns information about the amount of rows affected, this is returned as a second value.
function prepare-query (database-connection name query)
Parse and plan the given query, and store it under the given name. Note that prepared statements are per-connection, so they can only be executed through the same connection that prepared them.
exec-prepared (database-connection name parameters &optional (row-reader '
Execute the prepared statement by the given name.
Parameters should be given as a list. Each value in this list
should be of a type that
to-sql-string has been
specialised on. (Byte arrays will be passed in their binary form,
without being put through
to-sql-string.) The result
of the executing the statement, if any, is interpreted by the
given row reader, and returned. Again,
the number or affected rows is optionally returned as a second
function unprepare-query (database-connection name)
Close the prepared statement by the given name. This will free resources and allow the name to be associated with a new prepared query.
→ (values string needs-escaping)
Convert a Lisp value to its textual unescaped SQL representation. Returns a second value indicating whether this value should be escaped if it is to be put directly into a query.
You can define
for your own datatypes if you want to be able to pass them to
exec-prepared. When a
NIL second value is returned, this may be
T to indicate that the first value should simply be
escaped as a string, or a second string providing a type prefix
for the value. (This is used by S-SQL.)
When a rational number is passed into a query (as
it can not be expressed within 38 decimal digits (for example
1/3), it will be truncated, and lose some precision.
Set this variable to
nil to suppress that behaviour
and raise an error instead.
When debugging, it can be helpful to inspect the
queries that are being sent to the database. Set this variable to
an output stream value (
example) to have CL-postgres log every query it makes.
When profiling or debugging,
*query-log* may not
give enough information, or reparsing its output may not be
feasible. This variable may be set to a designator of function
taking two arguments. This function will be then called after
every query, and receive query string and internal time units
(CL:GET-INTERNAL-REAL-TIME)) spent in query
as its arguments.
Default value of this variable
'LOG-QUERY, which takes
processing. If you provide custom query callback and wish to
functionality, you will have to
LOG-QUERY from your
function log-query (query internal-time)
This function is default value
and logs queries
*QUERY-LOG* if it is
CL-postgres knows how to convert commonly used PostgreSQL data types to Lisp values. This table shows the mapping:
|bytea||(vector (unsigned-byte 8))|
The mapping from PostgreSQL types (identified by OID numbers) to the functions that interpret them is kept in so-called SQL readtables. All types for which no reader is defined will be returned as string values containing their PostgreSQL representation.
This variable is used to choose the current readtable. For simple use, you will not have to touch this, but it is possible that code within a Lisp image requires different readers in different situations, in which case you can create separate read tables.
Copies a given readtable.
Returns the default readtable, containing only the readers defined by CL-postgres itself.
function set-sql-reader (oid function &key table binary-p)
Define a new reader for a given type.
table defaults to
reader function should take a single argument, a string, and
transform that into some kind of equivalent Lisp value. When
binary-p is true, the reader function is supposed to
directly read the binary representation of the value. In most
cases this is not recommended, but if you want to use it: provide
a function that takes a binary input stream and an integer (the
size of the value, in bytes), and reads the value from that
stream. Note that reading less or more bytes than the given size
will horribly break your connection.
function set-sql-datetime-readers (&key date timestamp timestamp-with-timezone time interval table)
Since there is no widely recognised standard way
of representing dates and times in Common Lisp, and reading these
from string representation is clunky and slow, this function
provides a way to easily plug in binary readers for the
interval types. It should be given functions with the
daysis the amount of days since January 1st, 2000.
:timestamp, but for values of the 'timestamp with time zone' type (which PostgreSQL internally stores exactly the same as regular timestamps).
:interval (months days useconds)
Row readers are a way to read and group the results of queries. Roughly, they are functions that perform the iteration over the rows and cells in the result, and do something with the returned values.
row-reader ((fields) &body body)
Creates a row-reader, using the given name for the
variable. Inside the body this variable refers to a vector of
field descriptions. On top of that, two local functions are bound,
next-field. The first will
start reading the next row in the result, and returns a boolean
indicating whether there is another row. The second will read and
return one field, and should be passed the corresponding field
description from the fields argument as a parameter.
A row reader should take care to iterate over all
the rows in a result, and within each row iterate over all the
fields. This means it should contain an outer loop that calls
next-row, and every time
T it should iterate over the fields vector
next-field for every field.
The definition of
give you an idea what a row reader looks like:
(row-reader (fields) (loop :while (next-row) :collect (loop :for field :across fields :collect (next-field field))))
Obviously, row readers should not do things with the database connection like, say, close it or start a new query, since it still reading out the results from the current query.
macro def-row-reader (name (fields) &body body)
defun-like variant of
row-reader: creates a row
reader and gives it a top-level function name.
This can be used to get information about the fields read by a row reader. Given a field description, it returns the name the database associated with this column.
This extracts the PostgreSQL OID associated with this column. You can, if you really want to, query the pg_types table to find out more about the types denoted by OIDs.
list-row-reader (socket fields)
A row reader that builds a list of lists from the query results.
alist-row-reader (socket fields)
A row reader that returns a list of alists, which associate column names with values.
function ignore-row-reader (socket fields)
A row reader that completely ignores the result of a query.
When loading large amounts of data into PostgreSQL, it can be done significantly faster using the bulk copying feature. The drawback to this approach is that you don't find out about data integrity errors until the entire batch is completed but sometimes the speed is worth it
function open-db-writer (db table &optional columns)
Opens a table stream into which rows can be
written one at a time
db is either
a connection object or a list
of arguments that could be passed to
table is the name of an existing table into which this writer
will write rows. If you don't have data for all columns, use
columns to indicate those that you do.
function close-db-writer (writer &key abort)
Closes a bulk writer opened
open-db-writer. Will close the associated database
connection when it was created for this copier,
abort is true.
function db-write-row (writer row-data)
row-data into the table and columns
referenced by the writer.
row-data is a list of Lisp objects,
one for each column included when opening the writer. Arrays (the elements
of which must all be the same type) will be serialized into their Postgres
representation before being written into the DB.
Opening or querying a database may raise errors. CL-postgres will wrap the errors that the server returns in a lisp condition, and raise conditions of the same type when it detects some problem itself. Socket errors are let through as they are.
The type of database-related conditions. For
errors that you may want to catch by type, the
cl-postgres-error package defines a bucket of
subtypes used for specific errors. See the
cl-postgres/package.lisp file for a list.
A short message associated with this error.
A longer description of the problem, or
NIL if none is available.
The error code PostgreSQL associated with this error, if any. See the PostgreSQL manual for their meaning.
The query that led to this error, or
NIL if no query was involved.
The condition that caused this error, or
NIL when it was not caused by another condition.
For integrity-violation errors, returns the name
of the constraint that was violated (or
nil if no
constraint was found.)
database-error. An error
of this type (or one of its subclasses) is signaled when a query
is attempted with a connection object that is no longer connected,
or a database connection becomes invalid during a query. Always
:reconnect restart, which will cause the
library to make an attempt to restore the connection and re-try
The following shows an example use of this feature, a way to ensure that the first connection error causes a reconnect attempt, while others pass through as normal. A variation on this theme could continue trying to reconnect, with successively longer pauses.
(defun call-with-single-reconnect (fun) (let ((reconnected nil)) (handler-bind ((database-connection-error (lambda (err) (when (not reconnected) (setf reconnected t) (invoke-restart :reconnect))))) (funcall fun))))
The condition that is signalled when a
notification message is received from the PostgreSQL server.
This is a
WARNING condition which is caught by
WAIT-FOR-NOTIFICATION function that implements
synchronous waiting for notifications.
The channel string of this notification.
The payload of this notification.
The process ID of the process that sent the notification.