/predis

A redis-server written in Perl.

Primary LanguagePerl

predis

A Redis server implemented in 100% pure Perl which proxies commands to local storage.

This server will listen upon a network, and accept/process Redis commands. As such it can be useful for hacking around interesting problems.

Because the storage is persistent database-values can be stored between different invocations.

Currently we have several back-ends implemented:

  • An SQLite database-based back-end.
    • This is the most complete, and contains functionality not present in the others.
  • A Berkeley database back-end.
  • An LMDB database back-end.

Limitations

The implementation includes the string-primitives for all back-ends, other structures are ignored in the name of simplicity.

For each of the backends the following primitives are implemented:

  • get( key )
  • strlen( key )
  • set( key, value )
  • incr( key )
  • incrby( key, amount )
  • decr( key )
  • decrby( key, amount )
  • del( key )

Some of the set-primitives are supported ONLY with the SQLite backend:

  • sadd(key,val)
  • scard(key)
  • sdel(key,val)
  • smembers(key)
  • srandommember(key)

Testing

The code can be tested, as expected, with the redis-cli script.

First start the server:

$ perl predis

Then experiment

$ redis-cli set foo "$(cat /etc/passwd )"
OK
$ redis-cli get foo | head -n 5
root:x:0:0:root:/root:/bin/bash
daemon:x:1:1:daemon:/usr/sbin:/bin/sh
..
..
..

There are some simple tests included beneath t/ which can be executed as expected:

$ prove --shuffle t/

Benchmarks

There is a trivial benchmark script included, which will run a number of operations for 5 seconds at a time, and count the number that completed.

This is not intended to be an exhaustive benchmark, merely something to give an order of magnitude comparison between native redis and predis.

Upon my test system the following results were achieved:

Redis:

incr_decr  30895 requests in 5 seconds
get_set  33267 requests in 5 seconds
set_del  38706 requests in 5 seconds

Predis using SQLite:

incr_decr 11064 requests in 5 seconds
get_set   13100 requests in 5 seconds
set_del   11769 requests in 5 seconds

Predis using BDB (launched via ./predis --backend=BDB)

incr_decr 14926 requests in 5 seconds
get_set   15896 requests in 5 seconds
set_del   16282 requests in 5 seconds

Predis using LMDB (launched via ./predis --backend=LMDB)

incr_decr 14425 requests in 5 seconds
get_set   14827 requests in 5 seconds
set_del   14841 requests in 5 seconds

This suggests that predis is 3-4 times slower than Redis using SQlite, and half as fast using the Berkeley database back-end, but obviously benchmarks are unrealistic and will vary depending on your system and use-case.

Extending

To make this more more compatible with Redis we need to implement the various missing primitives:

  • The hash-related functionality.
  • The missing set-primitives need to be added.

The coding isn't hard, I just don't need it yet - so pull requests are most welcome.

The biggest reason for avoiding these right now is that the SQLite backend is more complete, and implementing sets & hashes with only key-value storage on the back-end is hard.

Steve