japester: (Default)
I've been digging into performance trends and metrics over the last few months and trying to find a sane and scalable app for deploying across Unix space. I'm really saddened by the complete and utter lack of useful open source software in this space. There are some of the semi commercial apps which are getting there, but are somewhat difficult to get running on Solaris. The GNU/libre ones are either complete and utter tripe, or suffer serious architectural faults. (And yes, I do suffer from the 'If it doesn't work as described by the documentation, I'll throw it away. Documentation is not hard people)

  • Hyperic.
    bloated and awkward interface. graphs? um .... they're there, somewhere. Damned if I could work out how to get it do them easily
    Summary: overweight monitoring app that doesn't do performance metrics
  • munin
    Just works! I am very sad to say that it gathers data by pulling it from hosts, that's a method that just does not scale.
    Summary: Very easy to use. works well in small/medium environments. Will not scale.
  • collectd
    collects data well, but there is no frontend worth using. They are all, and I mean all, alpha quality. Heymon appears to have come closest to being able to 'actually work,' but it's pretty ugly still.
    Summary: Reasonably easy to get items monitored and configured. Build a working and stable front end and it'd be just right.
  • hobbit/xymon
    Also, just works. It has some serious architecture flaws though, like trying to navigate just which file to edit to get it to monitor something, and problems with how it generates rrds.
    Summary: flawed architecture

I'll be looking at others in the near future. if you have any good/bright recommendations, I'm all ears. I'm not fussed about alerts/alarms, just performance/utilisation metrics.

Oh wait, the other rant.
Why the f*ck is it so damn hard to build anything on Solaris? I am obviously spoilt by having had the dream life living with Debian (and Ubuntu), where if i want to test some random app, all I have to do is 'apt-get install widget' and voila! It's there. and it runs. If it's not quite what i need, apt-get remove widget, and it's gone. Try that on Solaris? OpenCSW come kind of close, but does not have anywhere near the breadth of app choice that Debian has. Oh, and people don't build stuff on Solaris, so it's fairly common to find that it just won't work.

To get over the 'problems getting stuff built on Solaris', I've spun up a couple of VirtualBox Ubuntu servers and have been developing on them, then porting stuff back to Solaris. That's kept the frustration levels reasonably low.
japester: (Default)
I've been spending a fair chunk of the last month getting my brain wrapped around Reductive Lab's Puppet product. It's a configuration management framework aimed at SysAdmins who spend more than 5 minutes a week managing config files.

It has taken me quite some time to actually get started on it, as the learning curve is quite steep. The online documentation is enough to get you started, and then missed significant steps on getting from basic configuration to a usable environment. So, it's a typical OpenSource product then. I'm going to write up my experiences with getting started sometime RSN.

It's good. It works. It's frickin' powerful. So much so, that one of my colleagues is too scared to touch it, he believes that it's too easy to wreck your entire environment in quick easy steps. Well, that's half the fun, really :)

I have sucked in all our core infrastructure to the puppet brethren. Over the coming days, or weeks, I will roll the rest of the application servers in as well.

The only point of contention and definite confusion will be to know what is managed and what is not. I can see that being painful for the NOC monkeys who come in in a month (or three) to manage this environment. There are 90 of the little Slowaris beasts here though, so having some form of centralised management is required. Otherwise entropy will develop at a rate too quick to manage.
Just adding in hosts keeps me on my toes as each one has so far, at least one thing different from the SOE that it should be conforming to!


japester: (Default)

August 2013

18 192021222324


RSS Atom

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags