Monitoring the event log can quickly become straining for both the computer as well as the administrator as the event log grows and grows. To make this simpler for both the administrator and the computer NSClient++ 0.4.0 introduced real-time event log monitoring. This means we no longer scan the event log instead we simply scan events as they come in. The benefit, in addition to lowering the resources required, is that we can also get notified instantly when an error occurs instead of every 5 minutes or however often we check the log. Another addition is a simple client o generate event log message to help administrators debug event log filters. This is a quick introduction to event log monitoring and real-time event log monitoring showing how to set up real-time event log monitoring both for active and passive use via NSCA and NRPE.
So to start this off first we need to understand how the event log works and what we can do with it. The first thing to notice is that the event log is a bit “odd”. The core event log message is an EVENTLOGRECORD which is we simplify a bit has the following fields:
The time at which this entry was submitted.
The time at which this entry was received by the service to be written to the log.
The type of message is one of the following:
- Failure Audit
- Success Audit
A list of strings for rendering the message.
A token with which the developer can provide filtering and such.
Security identifier of the “active user” when the event was logged.
Now event id seems to me like something which is not really used as intended or was badly designed. The general idea is that each message (read text) has a unique identifier provided by the implementing application. Thus far it is even pretty smart as it means that the event log does not contain any actual messages only an id and some metadata. The oddities if you ask me is that they decided to use the event-id for a lot of other things as well. The field contained in this number (and since it is a number many of these are just bits) are:
- Severity. The severity is defined as follows:
- 00 – Success
- 01 – Informational
- 10 – Warning
- 11 – Error
- Customer bit. This bit is defined as follows:
- 0 – System code
- 1 – Customer code
Facility code (I think of this as a component within the application).
Number identifying the message
The first thing to notice is that if you look at modern windows the event viewer will not show you any of this information unless you “click advanced” and then only as a number next to the name Qualifiers.
The second thing is that some of these fields are a bit odd. Severity sounds a lot like “type” or level as it is called now. Facility sounds a lot like source. None of these fields will help provide any information for the end user and is mainly intended for the developer to help them organize message files or so it seems to me.
Back to NSClient++
So what can we do in NSClient++?
Well, first off we have two commands:
There is also one command line command available used to generate event log messages (mainly useful when debugging filters).
So why two commands?
Well, the first command is the “old normal” check command when every time you run the check the entire event log is scanned which means if you have a few hundred thousand items it will quickly use up a lot of resources.
To combat this in 0.4.0 I introduced something called “real-time” event log monitoring. The idea is that we listen for events and as they happen we scan them and match them against the filters. The up side of this is obviously a much reduced CPU footprint. as we instead of every 5 minutes have to scan hundreds of thousands of entry’s we only have to scan the hand full your server generates. The other benefit is that we can now get real-time errors if we use passive checking. So you no longer have to wait “for 5 minutes” instead once the error is triggered you get the red light in your monitoring tool almost instantly.
But lets start with understanding filters.
Filters was introduced in 0.3.8 of NSClient++ and consists of something similar to SQL statements (or WQL statements for that matter) where you can filter information you want. As of 0.4.0 we have the following keywords on which we can filter:
The event id
The event category
Qualifier part of event id
Facility part of event id
Customer bit part of event id
The entire event id including any qualifier/facility/customer/* parts
This is obviously faster than separating them into multiple statements.
- file / log
The renderd message
The string data (used to render the message)
This is much much faster and yields the same result as message.
- type (level)
So essentially we can filter on anything except sid which for some strange reason I missed . Writing filters is pretty straight forward as well and if you know some basic SQL you should not have too much trouble. This article is not about writing event log queries so I will only give you an example here (let me know if you want me to write an article about writing filters):
generated gt -2d AND severity NOT IN ('success', 'informational')
Real time event log
Real time event log monitoring is almost simpler to set up then using server-side checks. The drawback though is that it requires up-front and client-side configuration which is obviously not for everyone.
[/modules] CheckEventLog=1 [/settings/eventlog/real-time] enabled=true filter=id = 1000 and category = 0
The above configuration will do two things, first enable CheckEventLog module and then enable real-time checking. We also configure a random filter I was using when testing something. So configuring this is pretty straight forward and simple. Apart from this there are a hand full of options you can tweak as well. The only problem here is what happens when you hit a message?
Well, since we have not configured a destination nothing will happen so thus far this is a bit useless. Unfortunately this blog post is not about configuring NSCA so the next section will only skim through it a bit. But first lets look at the other options we can tweak:
Spawns a background thread which detects issues and reports them back instantly.
The destination for intercepted messages
- maximum age
How long before reporting “ok” (if this is set to off no ok will be reported only errors)
The initial age to scan when starting NSClient++
The initial age to scan when starting NSClient++
Coma separated list of logs to check
Log missed records (useful to detect issues with filters) not useful in production as it is a bit of a resource hog.
- enable active
This will store all matches so you can use real-time filters from active monitoring (use CheckEventlogCache).
- ok message
This is the message sent periodically when no error is discovered.
The alias to use for this event (in NSCA this is the service name).
If we are planning to use this with NSCA (and we are) there are a few things to extra to configure. First we have a concept called service_name which we need to define. A service name is the way Nagios (or Icinga) associates the result with a check. So if you have multiple checks you need to be able to configure multiple service names for different filters. The way this is handled is using aliases. Aliases can be configured in two ways:
- Locally by settings the key for the filter query under […/filters].
- Globally in the […/real-time] section
All filters without alias will use this alias.
So if you in Nagios have two different service_checks (eventlog_1 and eventlog_2) you need to set this using the […/filters] section like so:
[/settings/eventlog/real-time/filters] eventlog_1=id = 1000 and category = 1 eventlog_1=id = 1000 and category = 0
The other thing we need to configure is the NSCA client itself which for simple scenarios is pretty straight forward to configure as well:
[/modules] ; ... NSCAClient = 1 [/settings/NSCA/client/targets/default] address=nsca://127.0.0.1:5667 encryption=aes256 password=YL04nBb14stIgCjZxcudGtMqz4E6NN3W
Finally we need to tell CheckEventlog to send messages to NSCA which is done by specifying the destination:
[/settings/eventlog/real-time] ; ... destination=NSCA
Now, how can we test this (apart from sitting around waiting for something to happen?
Well, NSClient++ provides a rather nifty (and dangerous) command which allow you to inject messages into the event log.
nscp eventlog CheckEventLog Command line syntax: Allowed options: -h [ --help ] Show help screen -s [ --source ] arg (=Application Error) source to use -t [ --type ] arg Event type -l [ --level ] arg Event level (type) -f [ --facility ] arg Facility/Qualifier -q [ --qualifier ] arg Facility/Qualifier --severity arg Event severity -c [ --category ] arg Event category --customer arg Customer bit 0,1 -a [ --arguments ] arg Message arguments (strings) --eventlog-arguments arg Message arguments (strings) --event-arguments arg Message arguments (strings) -i [ --id ] arg Event ID
In our case since we filter on event id 1000 we can use the following command to insert an application error.
nscp eventlog --exec insert-eventlog --source "Application Error" --id 1000 --level error --category 0
Which will trigger the following message to be sent to NSCA: “Felet uppstod i programmet med namn: %1, version %2, tidsstämpel 0x%3…” all the %1, %2 represent insert points which is where your event log message strings would normally end up. To add this we can add a series of –eventlog-argument options to insert some strings into these markers like so:
nscp eventlog --exec insert-eventlog --source "Application Error" --id 1000 --level error --category 0 --eventlog-argument a --eventlog-argument b ...
Active monitoring and real time
So maybe you are thinking, darn I just configure everything to use active monitoring via NRPE, cant I get any real-time goodness? Well fortunately the answer is: Yes you can!
UPDATE The details in this section is deprecated as in 0.4.1 the check_eventlog_cache command was replaced by the more generic check_cache command in the SimpleCache module. It works the same though so the concepts are still valid. For details about the SimpleCache module please refer to the following blog posts:
If you recall we had two commands: check_eventlog and check_eventlog_cache and the latter does just that. This is done by configuring active monitoring to store all matches for you and then you can check the results cache using the check_eventlog_cache command.
To enable this we need to add one more option to the configuration file.
[/settings/pytest_eventlog/real-time] ;... enable active=true
And then we need to run the following command:
check_eventlog_cache warn=gt:0 crit=gt:0
Which will give you a critical message when you have more then 0 items in the cache. Be advised though that when you check the cache the cache is emptied meaning you will only get this warning “the first time” if you only have a single problem. So be sure you configure your Nagios server accordingly. Currently this is a bit simple and not very powerful so while you can use it it might not be what you are looking for and if that is the case please let me know so I can improve it in the future.
So I think with the introduction of filters in 0.3.8 and real-time filters in 0.4.0 as well as the ability to inject errors into the event log makes CheckEventlog a both powerful and simple tool for checking for errors in your event log. That’s pretty much it for now, please let me know what you want me to write about next!
UPDATE: Added information about check_eventlog_cache being replaced by the more generic check_cache.