Install Zabbix: Binaries (RPM or DEB) vs Code Source

[lmontalvanr]

Hi everyone.

I have a question about installation for zabbix.
For a production environment, is it advisable to use the repositories or is it better to compile through source code?
Some say that there is better performance when compiling because it takes better advantage of all the server resources and others that the performance is the same (binary / code source).
How true is this?
What is the best practice and recommendations for installing zabbix? (source or binarie)

[tim.mooney]

I have a question about installation for zabbix.
For a production environment, is it advisable to use the repositories or is it better to compile through source code?
Some say that there is better performance when compiling because it takes better advantage of all the server resources and others that the performance is the same (binary / code source).

In my opinion, unless you need to change something about how the official packages are constructed (for example, if your workplace policy doesn't allow non-distro packages to use /etc for config files -- that's just an example and not a great policy, but you get the point), most sites should use the official packages. The official packages come with "maximal features" compiled in.

If you compile Zabbix yourself (hint: you could compile it yourself and still generate packages, possibly even using the official build recipes), it might be possible to use platform-specific compilation options that squeeze out a few % performance improvement, but on the Zabbix server at least, it's rare that the server processes are the bottleneck. Typically, sites have problems because of bottlenecks or I/O issues with their database backend, not with the server itself. In the case of a very large environment where your backend database is already well-tuned, adding one or more Zabbix proxies is going to do far more to improve performance than getting a 5% or even 10% performance improvement by recompiling the server software. So, although self-compiling the server might get you a small performance improvement, when addressing a performance problem (if you have one), you always want to start with the largest bottleneck.

On the agents, it's unlikely that any one agent is going to be running code that is so performance intensive that recompiling it would help out. Except in the case of some UserParameter items, most items the agent collects are already quite efficient. I don't think you would see much of a benefit trying to maximize the performance of the agent.

One place where you do want to compile your own agent (again, you can still package it, even if you've compiled it yourself) is for OS platforms and versions where there aren't official packages available from zabbix.com.