kollekt-erlang

Yet another implementation of the kollekt service (origin by Paul Asmuth).

Motivation

Collect statistical data without handling them in the corresponding app itself.

Using UDP for a fire-and-forget mechanism (if we loose some packets we don't care).

Using Erlang to reduce the packet drops as much as possible (because Erlang has message inboxes for every process).

Building a distributed collector app, so the overlying production app servers do not need to communicate to one single collect service. And the distributed system will give fault-tolerance by sharing collected data across the node cluster.

Bucket data

Can be send as following:

mybucket;mybucketdata

or as concatenated list:

bucket1;valueA|||bucket2;valueB|||...

Delimiters can be changed in src/kollekt.hrl.

Take care, which delimiters you want to use, they must not occur in the bucket or value data!

Usage

Prerequisites

Install Erlang (I build with R15B).

Compile

./compile

Run

Demo service:

./run

Test emitter:

./run_emitter

in development, everything can (and will) change, stay tuned …

Status

• OTP'fied where possible and reasonable
• statistical output on console
• app handles incoming buckets, collects data and let buckets die when needed

TODOs

• when buckets die they should export their bucket data to a storage (files, DB, …)
• exports rules (time based, size based, …)
• distributable (running on different nodes)
• using Mnesia as a first level storage (a dedicated process can then export to further targets based on rules)