Skip to content

jameslikeslinux/duplog

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Duplog

Duplog will deduplicate messages from multiple similar streams. This can be used to take syslog messages from multiple redundant rsyslog servers, strip out duplicates between the streams, and produce a complete record of log messages for an application like Splunk.

Building

You need Apache Ant and a Java JDK installed. Then run:

% ant

to fetch all of the project's dependencies and compile the source. An executable jar file can be built with:

% ant jar

Running

  • On each rsyslog server, create a file such as:

    % cat /usr/local/libexec/rsyslog/send-to-rabbitmq 
    #!/bin/sh
    exec /usr/bin/java -jar /path/to/duplog.jar inject
    

    Then define a configuration in rsyslog such as:

    $ModLoad omprog
    $ActionOMProgBinary /usr/local/libexec/rsyslog/send-to-rabbitmq
    *.* :omprog:
    

    Finally, make sure RabbitMQ is running locally on the default port.

  • On the destination server, where deduplicated log messages are required, simply run:

    % java -jar /path/to/duplog.jar extract [-o OUTPUT_FILE] [-r REDIS_SERVER] syslog_server [syslog_server ...]
    

    where syslog_server is the hostname of a syslog server running RabbitMQ as above. A Redis server must be available to perform deduplication. It should be running on the default port with the following parameters set in /etc/redis/redis.conf:

    maxmemory <bytes>    # each unique message will consume about 100 bytes; configure based on messaging rate and available memory
    maxmemory-policy allkeys-lru
    

Benchmarking

To get a rough idea of how Duplog performs, you can pipe generated messages through the system.

  • On one or more syslog servers (as defined above), run:

    % java -cp /path/to/duplog.jar edu.umd.it.duplog.benchmark.Producer <token> | java -jar /path/to/duplog.jar inject
    

    where token is a short string that is the same on each message producer, but different for each run. You should see an updating message like:

    Messages produced: A last second / B per second average
    
  • On one or more deduplicating servers (as defined above), run:

    % java -jar /path/to/duplog.jar extract [-r REDIS_SERVER] syslog_server [syslog_server ...] -o - | java -cp /path/to/duplog.jar edu.umd.it.duplog.benchmark.Consumer
    

    You should see an updating message like:

    Messages consumed: X last second / Y per second average
    

About

Syslog Deduplicator

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages