The retcl module is an event-driven, object-oriented, Redis
client library for the Tcl programming language. The
library exposes a single retcl
class, conveniently packaged as a sourceable
Tcl module. Instances of
this class represent connections to a Redis server and are
used to send requests in the form of native Redis commands
and retrieve responses.
Other than a few book-keeping methods, retcl
instances transparently
handle Redis commands as first-class methods.
As an example, r SET K Hello
can be used to set the value of the key K
to the string Hello. This is achieved by proxying all unknown methods to the
Redis server by concatenating all arguments, effectively
making retcl
instances completely decoupled from any version of Redis. This
has several advantages:
-
A
retcl
instance does not need to know about the semantics of a particular Redis command. This includes syntax checks, context verification and arguments validation, which are offloaded to the Redis server. As a consequence, the code base remains clean and small. -
New commands introduced by a server upgrade are immediately available to a live application.
package require retcl
retcl create r
r SET key val
r -sync GET key ;# val
set r [retcl new ?host port? ?option ...?]
set r [retcl new ?-noconnect? ?option ...?]
retcl create r ?host port? ?option ...?
retcl create r ?-noconnect? ?option ...?
Create an instance r of retcl
. If no host
or port
is specified, the
client automatically connects to localhost on port 6379. If -noconnect
is
specified, the client is created in disconnected mode.
Each additional option
is a simple string or a list representing the
invocation of a configuration method. Here are some examples.
retcl create r localhost 6973 -async {+tls -cafile /path/to/ca.crt}
retcl create r -keepCache
retcl create r -noconnect -keepCache +async
$r connect ?host? ?port?
$r disconnect
$r connected
The connect
method can be used to connect to a different host and port. It is
an error to call this method on an already connected client. The disconnect
method can be called no matter the connection status; it disconnects the client
from the current host, if any. The connected
method can be used to query the
current connection status. It returns a true result if the client is connected
and a false result otherwise.
set rid1 [$r SET key val] ;# rid stands for result id
$r result $rid1 ;# OK
set rid2 [$r GET key]
$r result $rid2 ;# val
$r resultType $rid1 ;# SimpleString
$r resultType $rid2 ;# BulkString
$r -sync GET key ;# val
proc mycb {id type body} {
puts " id: $id"
puts "type: $type"
puts "body: $body"
}
$r -cb mycb GET key ;# returns immediately and arrange for mycb to be invoked
;# with {rds:1 BulkString val} when the result arrives
As shown in the examples above, the interaction with Redis is
very straightforwards. Any methods not directly understood by the retcl
class
are forwarded to the Redis server, along with any additional
arguments provided. The result is a small string representing a result id. Each
call to Redis produces a new result id, which can then
be queried to inspect its status, type, and value.
By using the -sync
switch, it is possible to have Redis
commands block and only return as soon as the result is available. In this
case, the return value is the value returned by Redis.
By using the -cb
switch, it is possible to arrange for a callback procedure
to be called whenever the result is ready. In this case, the command returns
immediately.
By default, retcl
objects operate in asynchronous mode: they return
immediately and produce a result id (rid) that can be inspected later on. The
methods -async
, +async
and ?async
can be used to disable, enbale, and
query this setting. When the asynchronous behaviour is off, methods wait and
return the values returned by Redis instead of a result id.
A cache of all results is kept by default. This allows to query previously
returned results. The -keepcache
, +keepcache
, and ?keepcache
methods can
be used to disable, enable, and query this setting. When the results cache is
disabled, results are removed from the cache as soon as they are retrieved by
the client.
A custom error handler can be setup with the errorHandler
method. The
argument is a command prefix that gets expanded and additioned with an error
message string. Passing an empty command prefix resets the error handler to the
default error
proc.
A pipeline can be built with the pipeline
method. The argument is a script
which gets evaluated in the context of the caller. Commands to the
Redis server are held for the duration of the script and
released as a bulk when the script ends.
Publish / subscribe callbacks for specific items can be specified with the
callback
method. The item
argument is a pattern or channel as in PSUBSCRIBE
and SUBSCRIBE. The callback
argument is a command prefix. Whenever a message
arrives on the specific channel, the command prefix is called by appending the
type of the message, the pattern that was subscribed to, the actual channel,
and the payload.
The connection to the Redis server is unencrypted by default.
If the TclTLS extension is available,
the +tls
method can be used to enable TLS. The method takes an optional list
of arguments that are passed as-is to the [tls::socket]
command.
In the default configuration, Redis requires a valid client
certificate on connection, which requires specifying a few parameters, e.g.,
r +tls -cafile /etc/redis/ca.crt -certfile /etc/redis/redis.crt -keyfile /etc/redis/redis.key
.
The -tls
and ?tls
methods can be used to disable and query TLS mode.
The ping-pong example can be run
in TLS mode via tclsh ping-pong.tcl --tls
.
As per the Redis Serialization Protocol (RESP), commands are sent over the wire as arrays of bulk strings. Bulk strings contain binary data, so all strings need to be converted to byte streams with values in the range 0-255. Unicode strings need to be encoded in utf-8. See the utf8-1.1 test for an example and ticket 70c08b5b5dfor for details.