icinga2/doc/8-advanced-topics.md

35 KiB

Advanced Topics

This chapter covers a number of advanced topics. If you're new to Icinga, you can safely skip over things you're not interested in.

Downtimes

Downtimes can be scheduled for planned server maintenance or any other targeted service outage you are aware of in advance.

Downtimes will suppress any notifications, and may trigger other downtimes too. If the downtime was set by accident, or the duration exceeds the maintenance, you can manually cancel the downtime. Planned downtimes will also be taken into account for SLA reporting tools calculating the SLAs based on the state and downtime history.

Multiple downtimes for a single object may overlap. This is useful when you want to extend your maintenance window taking longer than expected. If there are multiple downtimes triggered for one object, the overall downtime depth will be greater than 1.

If the downtime was scheduled after the problem changed to a critical hard state triggering a problem notification, and the service recovers during the downtime window, the recovery notification won't be suppressed.

Fixed and Flexible Downtimes

A fixed downtime will be activated at the defined start time, and removed at the end time. During this time window the service state will change to NOT-OK and then actually trigger the downtime. Notifications are suppressed and the downtime depth is incremented.

Common scenarios are a planned distribution upgrade on your linux servers, or database updates in your warehouse. The customer knows about a fixed downtime window between 23:00 and 24:00. After 24:00 all problems should be alerted again. Solution is simple - schedule a fixed downtime starting at 23:00 and ending at 24:00.

Unlike a fixed downtime, a flexible downtime will be triggered by the state change in the time span defined by start and end time, and then last for the specified duration in minutes.

Imagine the following scenario: Your service is frequently polled by users trying to grab free deleted domains for immediate registration. Between 07:30 and 08:00 the impact will hit for 15 minutes and generate a network outage visible to the monitoring. The service is still alive, but answering too slow to Icinga 2 service checks. For that reason, you may want to schedule a downtime between 07:30 and 08:00 with a duration of 15 minutes. The downtime will then last from its trigger time until the duration is over. After that, the downtime is removed (may happen before or after the actual end time!).

Scheduling a downtime

You can schedule a downtime either by using the Icinga 2 API action schedule-downtime or by sending an external command.

Fixed Downtime

If the host/service changes into a NOT-OK state between the start and end time window, the downtime will be marked as in effect and increases the downtime depth counter.

   |       |         |
start      |        end
       trigger time

Flexible Downtime

A flexible downtime defines a time window where the downtime may be triggered from a host/service NOT-OK state change. It will then last until the specified time duration is reached. That way it can happen that the downtime end time is already gone, but the downtime ends at trigger time + duration.

   |       |         |
start      |        end               actual end time
           |--------------duration--------|
       trigger time

Triggered Downtimes

This is optional when scheduling a downtime. If there is already a downtime scheduled for a future maintenance, the current downtime can be triggered by that downtime. This renders useful if you have scheduled a host downtime and are now scheduling a child host's downtime getting triggered by the parent downtime on NOT-OK state change.

Recurring Downtimes

ScheduledDowntime objects can be used to set up recurring downtimes for services.

Example:

apply ScheduledDowntime "backup-downtime" to Service {
  author = "icingaadmin"
  comment = "Scheduled downtime for backup"

  ranges = {
    monday = "02:00-03:00"
    tuesday = "02:00-03:00"
    wednesday = "02:00-03:00"
    thursday = "02:00-03:00"
    friday = "02:00-03:00"
    saturday = "02:00-03:00"
    sunday = "02:00-03:00"
  }

  assign where "backup" in service.groups
}

Comments

Comments can be added at runtime and are persistent over restarts. You can add useful information for others on repeating incidents (for example "last time syslog at 100% cpu on 17.10.2013 due to stale nfs mount") which is primarily accessible using web interfaces.

You can add a comment either by using the Icinga 2 API action add-comment or by sending an external command.

Acknowledgements

If a problem persists and notifications have been sent, you can acknowledge the problem. That way other users will get a notification that you're aware of the issue and probably are already working on a fix.

Note: Acknowledgements also add a new comment which contains the author and text fields.

You can send an acknowledgement either by using the Icinga 2 API action acknowledge-problem or by sending an external command.

Sticky Acknowledgements

The acknowledgement is removed if a state change occurs or if the host/service recovers (OK/Up state).

If you acknowlege a problem once you've received a Critical notification, the acknowledgement will be removed if there is a state transition to Warning.

OK -> WARNING -> CRITICAL -> WARNING -> OK

If you prefer to keep the acknowledgement until the problem is resolved (OK recovery) you need to enable the sticky parameter.

Expiring Acknowledgements

Once a problem is acknowledged it may disappear from your handled problems dashboard and no-one ever looks at it again since it will suppress notifications too.

This fire-and-forget action is quite common. If you're sure that a current problem should be resolved in the future at a defined time, you can define an expiration time when acknowledging the problem.

Icinga 2 will clear the acknowledgement when expired and start to re-notify, if the problem persists.

Time Periods

Time Periods define time ranges in Icinga where event actions are triggered, for example whether a service check is executed or not within the check_period attribute. Or a notification should be sent to users or not, filtered by the period and notification_period configuration attributes for Notification and User objects.

Note

If you are familiar with Icinga 1.x, these time period definitions are called legacy timeperiods in Icinga 2.

An Icinga 2 legacy timeperiod requires the ITL provided template legacy-timeperiod.

The TimePeriod attribute ranges may contain multiple directives, including weekdays, days of the month, and calendar dates. These types may overlap/override other types in your ranges dictionary.

The descending order of precedence is as follows:

  • Calendar date (2008-01-01)
  • Specific month date (January 1st)
  • Generic month date (Day 15)
  • Offset weekday of specific month (2nd Tuesday in December)
  • Offset weekday (3rd Monday)
  • Normal weekday (Tuesday)

If you don't set any check_period or notification_period attribute on your configuration objects, Icinga 2 assumes 24x7 as time period as shown below.

object TimePeriod "24x7" {
  import "legacy-timeperiod"

  display_name = "Icinga 2 24x7 TimePeriod"
  ranges = {
    "monday"    = "00:00-24:00"
    "tuesday"   = "00:00-24:00"
    "wednesday" = "00:00-24:00"
    "thursday"  = "00:00-24:00"
    "friday"    = "00:00-24:00"
    "saturday"  = "00:00-24:00"
    "sunday"    = "00:00-24:00"
  }
}

If your operation staff should only be notified during workhours, create a new timeperiod named workhours defining a work day from 09:00 to 17:00.

object TimePeriod "workhours" {
  import "legacy-timeperiod"

  display_name = "Icinga 2 8x5 TimePeriod"
  ranges = {
    "monday"    = "09:00-17:00"
    "tuesday"   = "09:00-17:00"
    "wednesday" = "09:00-17:00"
    "thursday"  = "09:00-17:00"
    "friday"    = "09:00-17:00"
  }
}

Furthermore if you wish to specify a notification period across midnight, you can define it the following way:

object Timeperiod "across-midnight" {
  import "legacy-timeperiod"

  display_name = "Nightly Notification"
  ranges = {
    "saturday" = "22:00-24:00"
    "sunday" = "00:00-03:00"
  }
}

Below you can see another example for configuring timeperiods across several days, weeks or months. This can be useful when taking components offline for a distinct period of time.

object Timeperiod "standby" {
  import "legacy-timeperiod"

  display_name = "Standby"
  ranges = {
    "2016-09-30 - 2016-10-30" = "00:00-24:00"
  }
}

Please note that the spaces before and after the dash are mandatory.

Once your time period is configured you can Use the period attribute to assign time periods to Notification and Dependency objects:

object Notification "mail" {
  import "generic-notification"

  host_name = "localhost"

  command = "mail-notification"
  users = [ "icingaadmin" ]
  period = "workhours"
}

Time Periods Inclusion and Exclusion

Sometimes it is necessary to exclude certain time ranges from your default time period definitions, for example, if you don't want to send out any notification during the holiday season, or if you only want to allow small time windows for executed checks.

The TimePeriod object provides the includes and excludes attributes to solve this issue. prefer_includes defines whether included or excluded time periods are preferred.

The following example defines a time period called holidays where notifications should be supressed:

object TimePeriod "holidays" {
  import "legacy-timeperiod"

  ranges = {
    "january 1" = "00:00-24:00"                 //new year's day
    "july 4" = "00:00-24:00"                    //independence day
    "december 25" = "00:00-24:00"               //christmas
    "december 31" = "18:00-24:00"               //new year's eve (6pm+)
    "2017-04-16" = "00:00-24:00"                //easter 2017
    "monday -1 may" = "00:00-24:00"             //memorial day (last monday in may)
    "monday 1 september" = "00:00-24:00"        //labor day (1st monday in september)
    "thursday 4 november" = "00:00-24:00"       //thanksgiving (4th thursday in november)
  }
}

In addition to that the time period weekends defines an additional time window which should be excluded from notifications:

object TimePeriod "weekends-excluded" {
  import "legacy-timeperiod"

  ranges = {
    "saturday"  = "00:00-09:00,18:00-24:00"
    "sunday"    = "00:00-09:00,18:00-24:00"
  }
}

The time period prod-notification defines the default time ranges and adds the excluded time period names as an array.

object TimePeriod "prod-notification" {
  import "legacy-timeperiod"

  excludes = [ "holidays", "weekends-excluded" ]

  ranges = {
    "monday"    = "00:00-24:00"
    "tuesday"   = "00:00-24:00"
    "wednesday" = "00:00-24:00"
    "thursday"  = "00:00-24:00"
    "friday"    = "00:00-24:00"
    "saturday"  = "00:00-24:00"
    "sunday"    = "00:00-24:00"
  }
}

Check Result Freshness

In Icinga 2 active check freshness is enabled by default. It is determined by the check_interval attribute and no incoming check results in that period of time.

threshold = last check execution time + check interval

Passive check freshness is calculated from the check_interval attribute if set.

threshold = last check result time + check interval

If the freshness checks are invalid, a new check is executed defined by the check_command attribute.

Check Flapping

Icinga 2 supports optional detection of hosts and services that are "flapping".

Flapping occurs when a service or host changes state too frequently, resulting in a storm of problem and recovery notifications. Flapping can be the source of configuration problems (i.e. thresholds set too low), troublesome services, or real network problems.

Flapping detection can be enabled or disabled using the enable_flapping attribute. The flapping_threshold attributes allows to specify the percentage of state changes when a host or service is considered to flap.

Note: There are known issues with flapping detection. Please refrain from enabling flapping until #4982 is fixed.

Volatile Services

By default all services remain in a non-volatile state. When a problem occurs, the SOFT state applies and once max_check_attempts attribute is reached with the check counter, a HARD state transition happens. Notifications are only triggered by HARD state changes and are then re-sent defined by the interval attribute.

It may be reasonable to have a volatile service which stays in a HARD state type if the service stays in a NOT-OK state. That way each service recheck will automatically trigger a notification unless the service is acknowledged or in a scheduled downtime.

Monitoring Icinga 2

Why should you do that? Icinga and its components run like any other service application on your server. There are predictable issues such as "disk space is running low" and your monitoring suffers from just that.

You would also like to ensure that features and backends are running and storing required data. Be it the database backend where Icinga Web 2 presents fancy dashboards, forwarded metrics to Graphite or InfluxDB or the entire distributed setup.

This list isn't complete but should help with your own setup. Windows client specific checks are highlighted.

Type Description Plugins and CheckCommands
System Filesystem disk, disk-windows (Windows Client)
System Memory, Swap mem, swap, memory (Windows Client)
System Hardware hpasm, ipmi-sensor
System Virtualization VMware, esxi_hardware
System Processes procs, service-windows (Windows Client)
System System Activity Reports check_sar_perf
System I/O iostat
System Network interfaces nwc_health, interfaces
System Users users, users-windows (Windows Client)
System Logs Forward them to Elastic Stack or Graylog and add your own alerts.
System NTP ntp_time
System Updates apt, yum
Icinga Status & Stats icinga (more below)
Icinga Cluster & Clients health checks
Database MySQL mysql_health
Database PostgreSQL postgres
Database Housekeeping Check the database size and growth and analyse metrics to examine trends.
Database DB IDO ido (more below)
Webserver Apache2, Nginx, etc. http, apache_status, nginx_status
Webserver Certificates http
Webserver Authorization http
Notifications Mail (queue) smtp, mailq
Notifications SMS (GSM modem) check_sms3_status
Notifications Messengers, Cloud services XMPP, Twitter, IRC, Telegram, PagerDuty, VictorOps, etc.
Metrics PNP, RRDTool check_pnp_rrds checks for stale RRD files.
Metrics Graphite graphite
Metrics InfluxDB check_influxdb
Metrics Elastic Stack elasticsearch, Elastic Stack integration
Metrics Graylog Graylog integration

The icinga CheckCommand provides metrics for the runtime stats of Icinga 2. You can forward them to your preferred graphing solution. If you require more metrics you can also query the REST API and write your own custom check plugin. Or you keep using the built-in object accessor functions to calculate stats in-memory.

There is a built-in ido check available for DB IDO MySQL/PostgreSQL which provides additional metrics for the IDO database.

apply Service "ido-mysql" {
  check_command = "ido"

  vars.ido_type = "IdoMysqlConnection"
  vars.ido_name = "ido-mysql" //the name defined in /etc/icinga2/features-enabled/ido-mysql.conf

  assign where match("master*.localdomain", host.name)
}

More specific database queries can be found in the DB IDO chapter.

Distributed setups should include specific health checks. You might also want to add additional checks for SSL certificate expiration.

Advanced Configuration Hints

Advanced Use of Apply Rules

Apply rules can be used to create a rule set which is entirely based on host objects and their attributes. In addition to that apply for and custom attribute override extend the possibilities.

The following example defines a dictionary on the host object which contains configuration attributes for multiple web servers. This then used to add three checks:

  • A ping4 check using the local IP address of the web server.
  • A tcp check querying the TCP port where the HTTP service is running on.
  • If the url key is defined, the third apply for rule will create service objects using the http CheckCommand. In addition to that you can optionally define the ssl attribute which enables HTTPS checks.

Host definition:

object Host "webserver01" {
  import "generic-host"
  address = "192.168.56.200"
  vars.os = "Linux"

  vars.webserver = {
    instance["status"] = {
      address = "192.168.56.201"
      port = "80"
      url = "/status"
    }
    instance["tomcat"] = {
      address = "192.168.56.202"
      port = "8080"
    }
    instance["icingaweb2"] = {
      address = "192.168.56.210"
      port = "443"
      url = "/icingaweb2"
      ssl = true
    }
  }
}

Service apply for definitions:

apply Service "webserver_ping" for (instance => config in host.vars.webserver.instance) {
  display_name = "webserver_" + instance
  check_command = "ping4"

  vars.ping_address = config.address

  assign where host.vars.webserver.instance
}

apply Service "webserver_port" for (instance => config in host.vars.webserver.instance) {
  display_name = "webserver_" + instance + "_" + config.port
  check_command = "tcp"

  vars.tcp_address = config.address
  vars.tcp_port = config.port

  assign where host.vars.webserver.instance
}

apply Service "webserver_url" for (instance => config in host.vars.webserver.instance) {
  display_name = "webserver_" + instance + "_" + config.url
  check_command = "http"

  vars.http_address = config.address
  vars.http_port = config.port
  vars.http_uri = config.url

  if (config.ssl) {
    vars.http_ssl = config.ssl
  }

  assign where config.url != ""
}

The variables defined in the host dictionary are not using the typical custom attribute prefix recommended for CheckCommand parameters. Instead they are re-used for multiple service checks in this example. In addition to defining check parameters this way, you can also enrich the display_name attribute with more details. This will be shown in in Icinga Web 2 for example.

Use Functions in Object Configuration

There is a limited scope where functions can be used as object attributes such as:

The other way around you can create objects dynamically using your own global functions.

Note

Functions called inside command objects share the same global scope as runtime macros. Therefore you can access host custom attributes like host.vars.os, or any other object attribute from inside the function definition used for set_if or command.

Tips when implementing functions:

  • Use log() to dump variables. You can see the output inside the icinga2.log file depending in your log severity
  • Use the icinga2 console to test basic functionality (e.g. iterating over a dictionary)
  • Build them step-by-step. You can always refactor your code later on.

Use Functions in Command Arguments set_if

The set_if attribute inside the command arguments definition in the CheckCommand object definition is primarily used to evaluate whether the command parameter should be set or not.

By default you can evaluate runtime macros for their existence. If the result is not an empty string, the command parameter is passed. This becomes fairly complicated when want to evaluate multiple conditions and attributes.

The following example was found on the community support channels. The user had defined a host dictionary named compellent with the key disks. This was then used inside service apply for rules.

object Host "dict-host" {
  check_command = "check_compellent"
  vars.compellent["disks"] = {
    file = "/var/lib/check_compellent/san_disks.0.json",
    checks = ["disks"]
  }
}

The more significant problem was to only add the command parameter --disk to the plugin call when the dictionary compellent contains the key disks, and omit it if not found.

By defining set_if as abbreviated lambda function and evaluating the host custom attribute compellent containing the disks this problem was solved like this:

object CheckCommand "check_compellent" {
  command   = [ "/usr/bin/check_compellent" ]
  arguments   = {
    "--disks"  = {
      set_if = {{
        var host_vars = host.vars
        log(host_vars)
        var compel = host_vars.compellent
        log(compel)
        compel.contains("disks")
      }}
    }
  }
}

This implementation uses the dictionary type method contains and will fail if host.vars.compellent is not of the type Dictionary. Therefore you can extend the checks using the typeof function.

You can test the types using the icinga2 console:

# icinga2 console
Icinga (version: v2.3.0-193-g3eb55ad)
<1> => srv_vars.compellent["check_a"] = { file="outfile_a.json", checks = [ "disks", "fans" ] }
null
<2> => srv_vars.compellent["check_b"] = { file="outfile_b.json", checks = [ "power", "voltages" ] }
null
<3> => typeof(srv_vars.compellent)
type 'Dictionary'
<4> =>

The more programmatic approach for set_if could look like this:

    "--disks" = {
      set_if = {{
        var srv_vars = service.vars
        if(len(srv_vars) > 0) {
          if (typeof(srv_vars.compellent) == Dictionary) {
            return srv_vars.compellent.contains("disks")
          } else {
            log(LogInformationen, "checkcommand set_if", "custom attribute compellent_checks is not a dictionary, ignoring it.")
            return false
          }
        } else {
          log(LogWarning, "checkcommand set_if", "empty custom attributes")
          return false
        }
      }}
    }

Use Functions as Command Attribute

This comes in handy for NotificationCommands or EventCommands which does not require a returned checkresult including state/output.

The following example was taken from the community support channels. The requirement was to specify a custom attribute inside the notification apply rule and decide which notification script to call based on that.

object User "short-dummy" {
}

object UserGroup "short-dummy-group" {
  assign where user.name == "short-dummy"
}

apply Notification "mail-admins-short" to Host {
   import "mail-host-notification"
   command = "mail-host-notification-test"
   user_groups = [ "short-dummy-group" ]
   vars.short = true
   assign where host.vars.notification.mail
}

The solution is fairly simple: The command attribute is implemented as function returning an array required by the caller Icinga 2. The local variable mailscript sets the default value for the notification scrip location. If the notification custom attribute short is set, it will override the local variable mailscript with a new value. The mailscript variable is then used to compute the final notification command array being returned.

You can omit the log() calls, they only help debugging.

object NotificationCommand "mail-host-notification-test" {
  command = {{
    log("command as function")
    var mailscript = "mail-host-notification-long.sh"
    if (notification.vars.short) {
       mailscript = "mail-host-notification-short.sh"
    }
    log("Running command")
    log(mailscript)

    var cmd = [ SysconfDir + "/icinga2/scripts/" + mailscript ]
    log(LogCritical, "me", cmd)
    return cmd
  }}

  env = {
  }
}

Use Custom Functions as Attribute

To use custom functions as attributes, the function must be defined in a slightly unexpected way. The following example shows how to assign values depending on group membership. All hosts in the slow-lan host group use 300 as value for ping_wrta, all other hosts use 100.

globals.group_specific_value = function(group, group_value, non_group_value) {
    return function() use (group, group_value, non_group_value) {
        if (group in host.groups) {
            return group_value
        } else {
            return non_group_value
        }
    }
}

apply Service "ping4" {
    import "generic-service"
    check_command = "ping4"

    vars.ping_wrta = group_specific_value("slow-lan", 300, 100)
    vars.ping_crta = group_specific_value("slow-lan", 500, 200)

    assign where true
}

Use Functions in Assign Where Expressions

If a simple expression for matching a name or checking if an item exists in an array or dictionary does not fit, you should consider writing your own global functions. You can call them inside assign where and ignore where expressions for apply rules or group assignments just like any other global functions for example match.

The following example requires the host myprinter being added to the host group printers-lexmark but only if the host uses a template matching the name lexmark*.

template Host "lexmark-printer-host" {
  vars.printer_type = "Lexmark"
}

object Host "myprinter" {
  import "generic-host"
  import "lexmark-printer-host"

  address = "192.168.1.1"
}

/* register a global function for the assign where call */
globals.check_host_templates = function(host, search) {
  /* iterate over all host templates and check if the search matches */
  for (tmpl in host.templates) {
    if (match(search, tmpl)) {
      return true
    }
  }

  /* nothing matched */
  return false
}

object HostGroup "printers-lexmark" {
  display_name = "Lexmark Printers"
  /* call the global function and pass the arguments */
  assign where check_host_templates(host, "lexmark*")
}

Take a different more complex example: All hosts with the custom attribute vars_app as nested dictionary should be added to the host group ABAP-app-server. But only if the app_type for all entries is set to ABAP.

It could read as wildcard match for nested dictionaries:

where host.vars.vars_app["*"].app_type == "ABAP"

The solution for this problem is to register a global function which checks the app_type for all hosts with the vars_app dictionary.

object Host "appserver01" {
  check_command = "dummy"
  vars.vars_app["ABC"] = { app_type = "ABAP" }
}
object Host "appserver02" {
  check_command = "dummy"
  vars.vars_app["DEF"] = { app_type = "ABAP" }
}

globals.check_app_type = function(host, type) {
  /* ensure that other hosts without the custom attribute do not match */
  if (typeof(host.vars.vars_app) != Dictionary) {
    return false
  }

  /* iterate over the vars_app dictionary */
  for (key => val in host.vars.vars_app) {
    /* if the value is a dictionary and if contains the app_type being the requested type */
    if (typeof(val) == Dictionary && val.app_type == type) {
      return true
    }
  }

  /* nothing matched */
  return false
}

object HostGroup "ABAP-app-server" {
  assign where check_app_type(host, "ABAP")
}

Access Object Attributes at Runtime

The Object Accessor Functions can be used to retrieve references to other objects by name.

This allows you to access configuration and runtime object attributes. A detailed list can be found here.

Simple cluster example for accessing two host object states and calculating a virtual cluster state and output:

object Host "cluster-host-01" {
  check_command = "dummy"
  vars.dummy_state = 2
  vars.dummy_text = "This host is down."
}

object Host "cluster-host-02" {
  check_command = "dummy"
  vars.dummy_state = 0
  vars.dummy_text = "This host is up."
}

object Host "cluster" {
  check_command = "dummy"
  vars.cluster_nodes = [ "cluster-host-01", "cluster-host-02" ]

  vars.dummy_state = {{
    var up_count = 0
    var down_count = 0
    var cluster_nodes = macro("$cluster_nodes$")

    for (node in cluster_nodes) {
      if (get_host(node).state > 0) {
        down_count += 1
      } else {
        up_count += 1
      }
    }

    if (up_count >= down_count) {
      return 0 //same up as down -> UP
    } else {
      return 2 //something is broken
    }
  }}

  vars.dummy_text = {{
    var output = "Cluster hosts:\n"
    var cluster_nodes = macro("$cluster_nodes$")

    for (node in cluster_nodes) {
      output += node + ": " + get_host(node).last_check_result.output + "\n"
    }

    return output
  }}
}

The following example sets time dependent thresholds for the load check based on the current time of the day compared to the defined time period.

object TimePeriod "backup" {
  import "legacy-timeperiod"

  ranges = {
    monday = "02:00-03:00"
    tuesday = "02:00-03:00"
    wednesday = "02:00-03:00"
    thursday = "02:00-03:00"
    friday = "02:00-03:00"
    saturday = "02:00-03:00"
    sunday = "02:00-03:00"
  }
}

object Host "webserver-with-backup" {
  check_command = "hostalive"
  address = "127.0.0.1"
}

object Service "webserver-backup-load" {
  check_command = "load"
  host_name = "webserver-with-backup"

  vars.load_wload1 = {{
    if (get_time_period("backup").is_inside) {
      return 20
    } else {
      return 5
    }
  }}
  vars.load_cload1 = {{
    if (get_time_period("backup").is_inside) {
      return 40
    } else {
      return 10
    }
  }}
}