I have a ruby daemon that selects 100 records from database and do a task with it.
To make it faster I usually create 3 instances of the same daemon. And each one selects diferents data by using mysql LIMIT and OFFSET.
The problem is that sometimes a task is performed 2 or 3 times with the same data record.
So I think that trusting only on database LIMIT and OFFSET is not enough ... since 2 or more daemons can actually collects the same data at the same time sometimes.
How can I do it safely? Avoiding 2 instances to select the same data
Rather than rolling your own solution, you might want to look at existing solutions for processing background jobs like Resque (a personal favorite). With Resque, you would queue a job for each of your rows using a trigger that makes sense in your application (it's hard to say without any context) for example a link on your website. At all times you would keep X number of workers running (three in your case) and Resque will do the queue management work for you. Resque uses Redis as a backend, so it supports atomic push/pop out of the gate (no more double-processing).
Resque also comes with a very intuitive and easy to use web interface for monitoring your jobs and workers.