TCP input events are expensive
Description
Gliffy Diagrams
Activity
Aaron Mildenstein February 6, 2015 at 9:39 PM
Since it's been 11 months with no response I am closing this with a status of incomplete.
With the new plugin nature of Logstash 1.5, if you have a custom implementation you need, you can write it yourself and keep it internal, or publish and share—your choice completely. Feel free to open a new issue at https://github.com/elasticsearch/logstash if you have any other issues.
Former user March 5, 2014 at 4:01 PM
OK. I'll take it on. I'm slow to push things back, as I wanted to stress test it on our 80 node hadoop cluster first.
But I was preparing to get the pull request out this week to clear my org ticket for this.
Jason Kendall March 5, 2014 at 3:57 PM
Sounds like a plan - just trying to triage what I can, so having this setup would be great. Thanks.
Former user March 5, 2014 at 3:54 PM
I have a fix a this point for 1.3.2 that I need to port to the head and will issue a pull request for.
I'll formally assign it to myself, if there are no objections.
Jason Kendall March 5, 2014 at 3:50 PM
Can we close this? Looks like a Will not fix.
We are writing own small and customized log shippers in various languages.
The problem is that we have to open a new TCP socket, write some log data and close it afterwards in order to ship a new event to Logstash every time!
This is a highly expensive task, both on client and server side.
It would be cool if there is a configurable option for specifying a special character for separating log events and keep the socket open during the logging session.
//edit
I was not able to find any documentation, specifying the necessary requirements for sending an event via TCP which led to annoying try and error...