Our (contextual) approach to monitoring...

Let me preface this with a disclaimer: I am not the creator/inventor of any of these ideas, this is just a combination of Google and the context of our environment.  For all I know people have been doing this for ages...it's new to me!

Our context
Our 'prod' environment is in a managed hosting context for various reasons (compliance, past decisions, etc).  This means that from a 'traditional' monitoring point of view we are covered.  Anything beyond that we're on our own.  Also, we are 95% Windows.

What is monitored today: Host/svc, host/svc, ports alive, HTTP GET = XYZ, is CPU crazy, is RAM maxed, etc...
Picking up what's left: Logs! Centrally aggregate all logs and query that pile of data to get alerts.  i.e. the services/applications we run (written in-house) dump stuff to the event log, among other places.

What are our goals?

  1. Get visibility into what our svs/apps are doing - know (be alerted) that svcX is dumping errors right now, not a week from now when something's gone wrong and someone finally looks at the event log
  2. Reduce the time required to troubleshoot things like firewall/load balancer logs (right now a lot of grep)
  3. Know what's going on inside the svcs/apps so we have insight into svc/app health, what our customers are doing/not doing, etc
  4. Give EVERYONE in the company this kind of visibility through 'public' dashboards

Note that all of these goals line up with 'better the customer experience' & 'improving our flow'.

What do we have to work with?

  • Windows, SQL, IIS, Windows services, BizTalk
  • Linux - a few Apache proxies, other Linux applications
  • Load balancers & firewalls
  • RSyslog
  • OMD/Check_MK (Nagios)

The tools
So what we are going to try and build is this:

  • NXlog (running on each server) dumps event logs to central syslog
  • Firewalls/load balancers/etc already dump to syslog
  • Central syslog forwards to ELK stack (Elasticsearch, Logstash, Kibana) and does 'permanent' archiving
  • Failure tolerant & performant ELK stack stores X amt of logs
  • Elasticsearch queries -> Nagios for alerting (pagerduty gets thrown around a lot, have to investigate)
How does this help?
Well first off, we have pretty much nothing in this department today, so anything is better than nothing.  There are obvious caveats to that rule, but really, 'overdue' is a word that would not be totally out of place for us.

Because we've relied on our hosting provider for so much, it's made us somewhat complacent, always good to be on your toes for this kind of thing.

One last point...we are pretty budget-constrained, so 'real' insight tools are out of reach.  We have the competance/expertise to pull this off, even if nobody here has actually done this before.

Is this the right way to go about it?
From what I have read...it's A way to go about it.  Is it THE way?  I don't think there is a rubber stamp answer to this, but the ELK stack is rapidly becoming a go-to tool.  Nagios/Zabbix and the like are probably not going away, and syslog is a standard we can all live with.  NXlog is lightweight, so acceptable.  Windows eventlog is Windows eventlog.

Ideally we'd start looking at WMI queries and whatnot, but for 'out of the box' least path of resistence, this gives us the logging visibility we need.  Alerting on ES queries...others are doing this, so at least we know it's not science fiction.

Future plans
The future goal is to re-code our services/applications to dump metrics to something like Bucky/StatsD, then on to Graphite, but this is a fairly large task.  WMI queries hopefully will be the path we take to get Windows performance stats into Graphite.

If you feel there is a better way, or would like to discuss, comment below.

Comments

Popular posts from this blog

DFSR - eventid 4312 - replication just won't work

Fixing duplicate SPNs (service principal name)

Logstash to Nagios - alerting based on Windows Event ID