icinga2/doc/15-troubleshooting.md

31 KiB

Icinga 2 Troubleshooting

Required Information

Please ensure to provide any detail which may help reproduce and understand your issue. Whether you ask on the community channels or you create an issue at GitHub, make sure that others can follow your explanations. If necessary, draw a picture and attach it for better illustration. This is especially helpful if you are troubleshooting a distributed setup.

We've come around many community questions and compiled this list. Add your own findings and details please.

  • Describe the expected behavior in your own words.
  • Describe the actual behavior in one or two sentences.
  • Ensure to provide general information such as:
    • How was Icinga 2 installed (and which repository in case) and which distribution are you using
    • icinga2 --version
    • icinga2 feature list
    • icinga2 daemon -C
    • Icinga Web 2 version (screenshot from System - About)
    • Icinga Web 2 modules e.g. the Icinga Director (optional)
  • Configuration insights:
    • Provide complete configuration snippets explaining your problem in detail
    • Your icinga2.conf file
    • If you run multiple Icinga 2 instances, the zones.conf file (or icinga2 object list --type Endpoint and icinga2 object list --type Zone) from all affected nodes.
  • Logs
    • Relevant output from your main and debug log in /var/log/icinga2. Please add step-by-step explanations with timestamps if required.
    • The newest Icinga 2 crash log if relevant, located in /var/log/icinga2/crash
  • Additional details
    • If the check command failed, what's the output of your manual plugin tests?
    • In case of debugging Icinga 2, the full back traces and outputs

Analyze your Environment

There are many components involved on a server running Icinga 2. When you analyze a problem, keep in mind that basic system administration knowledge is also key to identify bottlenecks and issues.

Tip

Monitor Icinga 2 and use the hints for further analysis.

  • Analyze the system's performance and dentify bottlenecks and issues.
  • Collect details about all applications (e.g. Icinga 2, MySQL, Apache, Graphite, Elastic, etc.).
  • If data is exchanged via network (e.g. central MySQL cluster) ensure to monitor the bandwidth capabilities too.
  • Add graphs and screenshots to your issue description

Install tools which help you to do so. Opinions differ, let us know if you have any additions here!

Analyse your Linux/Unix Environment

htop is a better replacement for top and helps to analyze processes interactively.

yum install htop
apt-get install htop

If you are for example experiencing performance issues, open htop and take a screenshot. Add it to your question and/or bug report.

Analyse disk I/O performance in Grafana, take a screenshot and obfuscate any sensitive details. Attach it when posting a question to the community channels.

The sysstat package provides a number of tools to analyze the performance on Linux. On FreeBSD you could use systat for example.

yum install sysstat
apt-get install sysstat

Example for vmstat (summary of memory, processes, etc.):

// summary
vmstat -s
// print timestamps, format in MB, stats every 1 second, 5 times
vmstat -t -S M 1 5

Example for iostat:

watch -n 1 iostat

Example for sar:

sar //cpu
sar -r //ram
sar -q //load avg
sar -b //I/O

sysstat also provides the iostat binary. On FreeBSD you could use systat for example.

If you are missing checks and metrics found in your analysis, add them to your monitoring!

Analyze your Windows Environment

A good tip for Windows are the tools found inside the Sysinternals Suite.

You can also start perfmon and analyze specific performance counters. Keep notes which could be important for your monitoring, and add service checks later on.

Enable Debug Output

Enable Debug Output on Linux/Unix

Enable the debuglog feature:

# icinga2 feature enable debuglog
# service icinga2 restart

The debug log file can be found in /var/log/icinga2/debug.log.

Alternatively you may run Icinga 2 in the foreground with debugging enabled. Specify the console log severity as an additional parameter argument to -x.

# /usr/sbin/icinga2 daemon -x notice

The log severity can be one of critical, warning, information, notice and debug.

Enable Debug Output on Windows

Open a command prompt with administrative privileges and enable the debug log feature.

C:> icinga2.exe enable feature debuglog

Ensure that the Icinga 2 service already writes the main log into C:\ProgramData\icinga2\var\log\icinga2. Restart the Icinga 2 service and open the newly created debug.log file.

C:> net stop icinga2
C:> net start icinga2

List Configuration Objects

The icinga2 object list CLI command can be used to list all configuration objects and their attributes. The tool also shows where each of the attributes was modified.

Tip

Use the Icinga 2 API to access config objects at runtime directly.

That way you can also identify which objects have been created from your apply rules.

# icinga2 object list

Object 'localhost!ssh' of type 'Service':
  * __name = 'localhost!ssh'
  * check_command = 'ssh'
    % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 5:3-5:23
  * check_interval = 60
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 24:3-24:21
  * host_name = 'localhost'
    % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 4:3-4:25
  * max_check_attempts = 3
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 23:3-23:24
  * name = 'ssh'
  * retry_interval = 30
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 25:3-25:22
  * templates = [ 'ssh', 'generic-service' ]
    % += modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 1:0-7:1
    % += modified in '/etc/icinga2/conf.d/templates.conf', lines 22:1-26:1
  * type = 'Service'
  * vars
    % += modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 6:3-6:19
    * sla = '24x7'
      % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 6:3-6:19

[...]

You can also filter by name and type:

# icinga2 object list --name *ssh* --type Service
Object 'localhost!ssh' of type 'Service':
  * __name = 'localhost!ssh'
  * check_command = 'ssh'
    % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 5:3-5:23
  * check_interval = 60
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 24:3-24:21
  * host_name = 'localhost'
    % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 4:3-4:25
  * max_check_attempts = 3
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 23:3-23:24
  * name = 'ssh'
  * retry_interval = 30
    % = modified in '/etc/icinga2/conf.d/templates.conf', lines 25:3-25:22
  * templates = [ 'ssh', 'generic-service' ]
    % += modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 1:0-7:1
    % += modified in '/etc/icinga2/conf.d/templates.conf', lines 22:1-26:1
  * type = 'Service'
  * vars
    % += modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 6:3-6:19
    * sla = '24x7'
      % = modified in '/etc/icinga2/conf.d/hosts/localhost/ssh.conf', lines 6:3-6:19

Found 1 Service objects.

[2014-10-15 14:27:19 +0200] information/cli: Parsed 175 objects.

Runtime modifications via the REST API are not immediately updated. Furthermore there is a known issue with group assign expressions which are not reflected in the host object output. You need to restart Icinga 2 in order to update the icinga2.debug cache file.

Where are the check command definitions?

Icinga 2 features a number of built-in check command definitions which are included with

include <itl>
include <plugins>

in the icinga2.conf configuration file. These files are not considered configuration files and will be overridden on upgrade, so please send modifications as proposed patches upstream. The default include path is set to LocalStateDir + "/share/icinga2/includes".

You should add your own command definitions to a new file in conf.d/ called commands.conf or similar.

Checks

Executed Command for Checks

  • Use the Icinga 2 API to query host/service objects for their check result containing the executed shell command.
  • Use the Icinga 2 console cli command to fetch the checkable object, its check result and the executed shell command.
  • Alternatively enable the debug log and look for the executed command.

Example for a service object query using a regex match on the name:

$ curl -k -s -u root:icinga -H 'Accept: application/json' -H 'X-HTTP-Method-Override: GET' -X POST 'https://localhost:5665/v1/objects/services' \
-d '{ "filter": "regex(pattern, service.name)", "filter_vars": { "pattern": "^http" }, "attrs": [ "__name", "last_check_result" ] }' | python -m json.tool
{
    "results": [
        {
            "attrs": {
                "__name": "example.localdomain!http",
                "last_check_result": {
                    "active": true,
                    "check_source": "example.localdomain",
                    "command": [
                        "/usr/local/sbin/check_http",
                        "-I",
                        "127.0.0.1",
                        "-u",
                        "/"
                    ],

  ...

                }
            },
            "joins": {},
            "meta": {},
            "name": "example.localdomain!http",
            "type": "Service"
        }
    ]
}

Example for using the icinga2 console CLI command evaluation functionality:

$ ICINGA2_API_PASSWORD=icinga icinga2 console --connect 'https://root@localhost:5665/' \
--eval 'get_service("example.localdomain", "http").last_check_result.command' | python -m json.tool
[
    "/usr/local/sbin/check_http",
    "-I",
    "127.0.0.1",
    "-u",
    "/"
]

Example for searching the debug log:

# icinga2 feature enable debuglog
# systemctl restart icinga2
# tail -f /var/log/icinga2/debug.log | grep "notice/Process"

Checks are not executed

  • Check the debug log to see if the check command gets executed.
  • Verify that failed depedencies do not prevent command execution.
  • Make sure that the plugin is executable by the Icinga 2 user (run a manual test).
  • Make sure the checker feature is enabled.
  • Use the Icinga 2 API event streams to receive live check result streams.

Examples:

# sudo -u icinga /usr/lib/nagios/plugins/check_ping -4 -H 127.0.0.1 -c 5000,100% -w 3000,80%

# icinga2 feature enable checker
The feature 'checker' is already enabled.

Fetch all check result events matching the event.service name random:

$ curl -k -s -u root:icinga -X POST 'https://localhost:5665/v1/events?queue=debugchecks&types=CheckResult&filter=match%28%22random*%22,event.service%29'

Check Fork Errors

We've learned that newer kernel versions introduce a fork limit for cgroups which is enabled in SLES 12 SP2+ for example. The default value for DefaultTasksMax in Systemd is set to 512.

Icinga 2 relies on forking child processes to execute commands and might therefore hit this limit in larger setups.

The error message could look like this:

2017-01-12T11:55:40.742685+01:00 icinga2-master1 kernel: [65567.582895] cgroup: fork rejected by pids controller in /system.slice/icinga2.service

In order to solve the problem, increase the value for DefaultTasksMax or set it to infinity:

[root@icinga2-master1.localdomain /]# vim /usr/lib/systemd/system/icinga2.service

[Service]

DefaultTasksMax=infinity

[root@icinga2-master1.localdomain /]# systemctl daemon-reload
[root@icinga2-master1.localdomain /]# systemctl restart icinga2

Please note that this setting is available since Systemd version 226.

Late Check Results

Icinga Web 2 provides a dashboard overview for overdue checks.

The REST API provides the [status] URL endpoint with some generic metrics on Icinga and its features.

# curl -k -s -u root:icinga 'https://localhost:5665/v1/status' | python -m json.tool | less

You can also calculate late check results via the REST API:

  • Fetch the last_check timestamp from each object
  • Compare the timestamp with the current time and add check_interval multiple times (change it to see which results are really late, like five times check_interval)

You can use the icinga2 console to connect to the instance, fetch all data and calculate the differences. More infos can be found in this blogpost.

# ICINGA2_API_USERNAME=root ICINGA2_API_PASSWORD=icinga icinga2 console --connect 'https://localhost:5665/'

<1> => var res = []; for (s in get_objects(Service).filter(s => s.last_check < get_time() - 2 * s.check_interval)) { res.add([s.__name, DateTime(s.last_check).to_string()]) }; res

[ [ "10807-host!10807-service", "2016-06-10 15:54:55 +0200" ], [ "mbmif.int.netways.de!disk /", "2016-01-26 16:32:29 +0100" ] ]

Or if you are just interested in numbers, call len on the result array res:

<2> => var res = []; for (s in get_objects(Service).filter(s => s.last_check < get_time() - 2 * s.check_interval)) { res.add([s.__name, DateTime(s.last_check).to_string()]) }; res.len()

2.000000

If you need to analyze that problem multiple times, just add the current formatted timestamp and repeat the commands.

<23> => DateTime(get_time()).to_string()

"2017-04-04 16:09:39 +0200"

<24> => var res = []; for (s in get_objects(Service).filter(s => s.last_check < get_time() - 2 * s.check_interval)) { res.add([s.__name, DateTime(s.last_check).to_string()]) }; res.len()

8287.000000

More details about the Icinga 2 DSL and its possibilities can be found in the language and library reference chapters.

Late Check Results in Distributed Environments

When it comes to a distributed HA setup, each node is responsible for a load-balanced amount of checks. Host and Service objects provide the attribute paused. If this is set to false, the current node actively attempts to schedule and execute checks. Otherwise the node does not feel responsible.

<3> => var res = {}; for (s in get_objects(Service).filter(s => s.last_check < get_time() - 2 * s.check_interval)) { res[s.paused] += 1 }; res
{
  @false = 2.000000
  @true = 1.000000
}

You may ask why this analysis is important? Fair enough - if the numbers are not inverted in a HA zone with two members, this may give a hint that the cluster nodes are in a split-brain scenario, or you've found a bug in the cluster.

If you are running a cluster setup where the master/satellite executes checks on the client via top down command endpoint mode, you might want to know which zones are affected.

This analysis assumes that clients which are not connected, have the string connected in their service check result output and their state is UNKNOWN.

<4> => var res = {}; for (s in get_objects(Service)) { if (s.state==3) { if (match("*connected*", s.last_check_result.output)) { res[s.zone] += [s.host_name] } } };  for (k => v in res) { res[k] = len(v.unique()) }; res

{
  Asia = 31.000000
  Europe = 214.000000
  USA = 207.000000
}

The result set shows the configured zones and their affected hosts in a unique list. The output also just prints the numbers but you can adjust this by omitting the len() call inside the for loop.

Notifications are not sent

  • Check the debug log to see if a notification is triggered.
  • If yes, verify that all conditions are satisfied.
  • Are any errors on the notification command execution logged?

Please ensure to add these details with your own description to any question or issue posted to the community channels.

Verify the following configuration:

  • Is the host/service enable_notifications attribute set, and if so, to which value?
  • Do the notification attributes states, types, period match the notification conditions?
  • Do the user attributes states, types, period match the notification conditions?
  • Are there any notification begin and end times configured?
  • Make sure the notification feature is enabled.
  • Does the referenced NotificationCommand work when executed as Icinga user on the shell?

If notifications are to be sent via mail, make sure that the mail program specified inside the NotificationCommand object exists. The name and location depends on the distribution so the preconfigured setting might have to be changed on your system.

Examples:

# icinga2 feature enable notification
The feature 'notification' is already enabled.

You can use the Icinga 2 API event streams to receive live notification streams:

$ curl -k -s -u root:icinga -X POST 'https://localhost:5665/v1/events?queue=debugnotifications&types=Notification'

Feature is not working

  • Make sure that the feature configuration is enabled by symlinking from features-available/ to features-enabled and that the latter is included in icinga2.conf.
  • Are the feature attributes set correctly according to the documentation?
  • Any errors on the logs?

Look up the object type for the required feature and verify it is enabled:

# icinga2 object list --type <feature object type>

Example for the graphite feature:

# icinga2 object list --type GraphiteWriter

Configuration is ignored

  • Make sure that the line(s) are not commented out (starting with // or #, or encapsulated by /* ... */).
  • Is the configuration file included in icinga2.conf?

Run the configuration validation and add notice as log severity. Search for the file which should be included i.e. using the grep CLI command.

# icinga2 daemon -C -x notice | grep command

Configuration attributes are inherited from

Icinga 2 allows you to import templates using the import keyword. If these templates contain additional attributes, your objects will automatically inherit them. You can override or modify these attributes in the current object.

The object list CLI command allows you to verify the attribute origin.

Configuration Value with Single Dollar Sign

In case your configuration validation fails with a missing closing dollar sign error message, you did not properly escape the single dollar sign preventing its usage as runtime macro.

critical/config: Error: Validation failed for Object 'ping4' (Type: 'Service') at /etc/icinga2/zones.d/global-templates/windows.conf:24: Closing $ not found in macro format string 'top-syntax=${list}'.

Correct the custom attribute value to

"top-syntax=$${list}"

Cluster and Clients Troubleshooting

This applies to any Icinga 2 node in a distributed monitoring setup.

You should configure the cluster health checks if you haven't done so already.

Note

Some problems just exist due to wrong file permissions or applied packet filters. Make sure to check these in the first place.

Cluster Troubleshooting Connection Errors

General connection errors could be one of the following problems:

  • Incorrect network configuration
  • Packet loss
  • Firewall rules preventing traffic

Use tools like netstat, tcpdump, nmap, etc. to make sure that the cluster communication works (default port is 5665).

# tcpdump -n port 5665 -i any

# netstat -tulpen | grep icinga

# nmap yourclusternode.localdomain

Cluster Troubleshooting SSL Errors

If the cluster communication fails with SSL error messages, make sure to check the following

  • File permissions on the SSL certificate files
  • Does the used CA match for all cluster endpoints?
    • Verify the Issuer being your trusted CA
    • Verify the Subject containing your endpoint's common name (CN)
    • Check the validity of the certificate itself

Steps on the client icinga2-node2.localdomain:

# ls -la /etc/icinga2/pki

# cd /etc/icinga2/pki/
# openssl x509 -in icinga2-node2.localdomain.crt -text
Certificate:
    Data:
        Version: 1 (0x0)
        Serial Number: 2 (0x2)
    Signature Algorithm: sha1WithRSAEncryption
        Issuer: C=DE, ST=Bavaria, L=Nuremberg, O=NETWAYS GmbH, OU=Monitoring, CN=Icinga CA
        Validity
            Not Before: Jan  7 13:17:38 2014 GMT
            Not After : Jan  5 13:17:38 2024 GMT
        Subject: C=DE, ST=Bavaria, L=Nuremberg, O=NETWAYS GmbH, OU=Monitoring, CN=icinga2-node2.localdomain
        Subject Public Key Info:
            Public Key Algorithm: rsaEncryption
                Public-Key: (4096 bit)
                Modulus:
                ...

Try to manually connect from icinga2-node2.localdomain to the master node icinga2-node1.localdomain:

# openssl s_client -CAfile /etc/icinga2/pki/ca.crt -cert /etc/icinga2/pki/icinga2-node2.localdomain.crt -key /etc/icinga2/pki/icinga2-node2.localdomain.key -connect icinga2-node1.localdomain:5665

CONNECTED(00000003)
---
...

If the connection attempt fails or your CA does not match, verify the master and client certificates.

Cluster Troubleshooting Unauthenticated Clients

Unauthenticated nodes are able to connect. This is required for client setups.

Master:

[2015-07-13 18:29:25 +0200] information/ApiListener: New client connection for identity 'icinga-client' (unauthenticated)

Client as command execution bridge:

[2015-07-13 18:29:26 +1000] notice/ApiEvents: Discarding 'execute command' message from 'icinga-master': Invalid endpoint origin (client not allowed).

If these messages do not go away, make sure to verify the master and client certificates.

Cluster Troubleshooting SSL Certificate Verification

Make sure to verify the client's certificate and its received ca.crt in /etc/icinga2/pki and ensure that both instances are signed by the same CA.

# openssl verify -verbose -CAfile /etc/icinga2/pki/ca.crt /etc/icinga2/pki/icinga2-node1.localdomain.crt
icinga2-node1.localdomain.crt: OK

# openssl verify -verbose -CAfile /etc/icinga2/pki/ca.crt /etc/icinga2/pki/icinga2-node2.localdomain.crt
icinga2-node2.localdomain.crt: OK

Fetch the ca.crt file from the client node and compare it to your master's ca.crt file:

# scp icinga2-node2:/etc/icinga2/pki/ca.crt test-client-ca.crt
# diff -ur /etc/icinga2/pki/ca.crt test-client-ca.crt

On SLES11 you'll need to use the openssl1 command instead of openssl.

Cluster Troubleshooting Message Errors

At some point, when the network connection is broken or gone, the Icinga 2 instances will be disconnected. If the connection can't be re-established between endpoints in the same HA zone, they remain in a Split-Brain-mode and history may differ.

Although the Icinga 2 cluster protocol stores historical events in a replay log for later synchronisation, you should make sure to check why the network connection failed.

Cluster Troubleshooting Command Endpoint Errors

Command endpoints can be used for clients as well as inside an High-Availability cluster.

There is no cli command for manually executing the check, but you can verify the following (e.g. by invoking a forced check from the web interface):

  • /var/log/icinga2/icinga2.log contains connection and execution errors.

  • The ApiListener is not enabled to accept commands.

  • CheckCommand definition not found on the remote client.

  • Referenced check plugin not found on the remote client.

  • Runtime warnings and errors, e.g. unresolved runtime macros or configuration problems.

  • Specific error messages are also populated into UNKNOWN check results including a detailed error message in their output.

  • Verify the check_source object attribute. This is populated by the node executing the check.

  • More verbose logs are found inside the debug log.

  • Use the Icinga 2 API event streams to receive live check result streams.

Fetch all check result events matching the event.service name remote-client:

$ curl -k -s -u root:icinga -X POST 'https://localhost:5665/v1/events?queue=debugcommandendpoint&types=CheckResult&filter=match%28%22remote-client*%22,event.service%29'

Cluster Troubleshooting Config Sync

If the cluster zones do not sync their configuration, make sure to check the following:

  • Within a config master zone, only one configuration master is allowed to have its config in /etc/icinga2/zones.d. ** The master syncs the configuration to /var/lib/icinga2/api/zones/ during startup and only syncs valid configuration to the other nodes. ** The other nodes receive the configuration into /var/lib/icinga2/api/zones/.
  • The icinga2.log log file in /var/log/icinga2 will indicate whether this ApiListener accepts config, or not.

Verify the object's version attribute on all nodes to check whether the config update and reload was succesful or not.

Cluster Troubleshooting Overdue Check Results

If your master does not receive check results (or any other events) from the child zones (satellite, clients, etc.), make sure to check whether the client sending in events is allowed to do so.

The distributed monitoring conventions apply. So, if there's a mismatch between your client node's endpoint name and its provided certificate's CN, the master will deny all events.

Tip

Icinga Web 2 provides a dashboard view for overdue check results.

Enable the debug log on the master for more verbose insights.

If the client cannot authenticate, it's a more general problem.

The client's endpoint is not configured on nor trusted by the master node:

Discarding 'check result' message from 'icinga2b': Invalid endpoint origin (client not allowed).

The check result message sent by the client does not belong to the zone the checkable object is in on the master:

Discarding 'check result' message from 'icinga2b': Unauthorized access.

Cluster Troubleshooting Replay Log

If your /var/lib/icinga2/api/log directory grows, it generally means that your cluster cannot replay the log on connection loss and re-establishment. A master node for example will store all events for not connected endpoints in the same and child zones.

Check the following:

  • All clients are connected? (e.g. cluster health check).
  • Check your connection in general.
  • Does the log replay work, e.g. are all events processed and the directory gets cleared up over time?
  • Decrease the log_duration attribute value for that specific endpoint.