Scroll to navigation

CRM(8) crmsh documentation CRM(8)


crm - Pacemaker command line interface for configuration and management




The crm shell is a command-line based cluster configuration and management tool. Its goal is to assist as much as possible with the configuration and maintenance of Pacemaker-based High Availability clusters.

For more information on Pacemaker itself, see

crm works both as a command-line tool to be called directly from the system shell, and as an interactive shell with extensive tab completion and help.

The primary focus of the crm shell is to provide a simplified and consistent interface to Pacemaker, but it also provides tools for managing the creation and configuration of High Availability clusters from scratch. To learn more about this aspect of crm, see the cluster section below.

The crm shell can be used to manage every aspect of configuring and maintaining a cluster. It provides a simplified line-based syntax on top of the XML configuration format used by Pacemaker, commands for starting and stopping resources, tools for exploring the history of a cluster including log scraping and a set of cluster scripts useful for automating the setup and installation of services on the cluster nodes.

The crm shell is line oriented: every command must start and finish on the same line. It is possible to use a continuation character (\) to write one command in two or more lines. The continuation character is commonly used when displaying configurations.


-f, --file=FILE

Load commands from the given file. If a dash - is used in place of a file name, crm will read commands from the shell standard input (stdin).

-c, --cib=CIB

Start the session using the given shadow CIB file. Equivalent to cib use <CIB>.

-D, --display=OUTPUT_TYPE

Choose one of the output options: plain, color-always, color, or uppercase. The default is color if the terminal emulation supports colors. Otherwise, plain is used.

-F, --force

Make crm proceed with applying changes where it would normally ask the user to confirm before proceeding. This option is mainly useful in scripts, and should be used with care.

-w, --wait

Make crm wait for the cluster transition to finish (for the changes to take effect) after each processed line.

-H, --history=DIR|FILE|SESSION

A directory or file containing a cluster report to load into the history commands, or the name of a previously saved history session.

-h, --help

Print help page.


Print crmsh version and build information (Mercurial Hg changeset hash).

-d, --debug

Print verbose debugging information.

-R, --regression-tests

Enables extra verbose trace logging used by the regression tests. Logs all external calls made by crmsh.


Extra directory where crm looks for cluster scripts, or a list of directories separated by semi-colons (e.g. /dir1;/dir2;etc.).

-o, --opt=OPTION=VALUE

Set crmsh option temporarily. If the options are saved using options save then the value passed here will also be saved. Multiple options can be set by using -o multiple times.


This section of the user guide covers general topics about the user interface and describes some of the features of crmsh in detail.

User interface

The main purpose of crmsh is to provide a simple yet powerful interface to the cluster stack. There are two main modes of operation with the user interface of crmsh:

•Command line (single-shot) use - Use crm as a regular UNIX command from your usual shell. crm has full bash completion built in, so using it in this manner should be as comfortable and familiar as using any other command-line tool.

•Interactive mode - By calling crm without arguments, or by calling it with only a sublevel as argument, crm enters the interactive mode. In this mode, it acts as its own command shell, which remembers which sublevel you are currently in and allows for rapid and convenient execution of multiple commands within the same sublevel. This mode also has full tab completion, as well as built-in interactive help and syntax highlighting.

Here are a few examples of using crm both as a command-line tool and as an interactive shell:

Command line (one-shot) use:.

# crm resource stop www_app

Interactive use:.

# crm
crm(live)# resource
crm(live)resource# unmanage tetris_1
crm(live)resource# up
crm(live)# node standby node4

Cluster configuration:.

# crm configure<<EOF

# resources
primitive disk0 iscsi \
params portal=
primitive fs0 Filesystem \
params device=/dev/disk/by-label/disk0 directory=/disk0 fstype=ext3
primitive internal_ip IPaddr params ip=
primitive apache apache \
params configfile=/disk0/etc/apache2/site0.conf
primitive apcfence stonith:apcsmart \
params ttydev=/dev/ttyS0 hostlist="node1 node2" \
op start timeout=60s
primitive pingd ocf:pacemaker:ping \
params name=pingd dampen=5s multiplier=100 host_list="r1 r2"
# monitor apache and the UPS
monitor apache 60s:30s
monitor apcfence 120m:60s
# cluster layout
group internal_www \
disk0 fs0 internal_ip apache
clone fence apcfence \
meta globally-unique=false clone-max=2 clone-node-max=1
clone conn pingd \
meta globally-unique=false clone-max=2 clone-node-max=1
location node_pref internal_www \
rule 50: #uname eq node1 \
rule pingd: defined pingd
# cluster properties
property stonith-enabled=true
commit EOF

The crm interface is hierarchical, with commands organized into separate levels by functionality. To list the available levels and commands, either execute help <level>, or, if at the top level of the shell, simply typing help will provide an overview of all available levels and commands.

The (live) string in the crm prompt signifies that the current CIB in use is the cluster live configuration. It is also possible to work with so-called shadow CIBs. These are separate, inactive configurations stored in files, that can be applied and thereby replace the live configuration at any time.

Tab completion

The crm makes extensive use of tab completion. The completion is both static (i.e. for crm commands) and dynamic. The latter takes into account the current status of the cluster or information from installed resource agents. Sometimes, completion may also be used to get short help on resource parameters. Here are a few examples:

crm(live)resource# <TAB><TAB>
ban           demote        maintenance   param         scores        trace
cd            failcount     manage        promote       secret        unmanage
cleanup       help          meta          quit          start         untrace
clear         locate        move          refresh       status        up
constraints   ls            operations    restart       stop          utilization
crm(live)configure# primitive fence-1 <TAB><TAB>
lsb:      ocf:      service:  stonith:  systemd:
crm(live)configure# primitive fence-1 stonith:<TAB><TAB>
apcmaster                external/ippower9258     fence_legacy
apcmastersnmp            external/kdumpcheck      ibmhmc
apcsmart                 external/libvirt         ipmilan
crm(live)configure# primitive fence-1 stonith:ipmilan params <TAB><TAB>
auth=      hostname=  ipaddr=    login=     password=  port=      priv=
crm(live)configure# primitive fence-1 stonith:ipmilan params auth=<TAB><TAB>
auth* (string)

The authorization type of the IPMI session ("none", "straight", "md2", or "md5")

crmsh also comes with bash completion usable directly from the system shell. This should be installed automatically with the command itself.

Shorthand syntax

When using the crm shell to manage clusters, you will end up typing a lot of commands many times over. Clear command names like configure help in understanding and learning to use the cluster shell, but is easy to misspell and is tedious to type repeatedly. The interactive mode and tab completion both help with this, but the crm shell also has the ability to understand a variety of shorthand aliases for all of the commands.

For example, instead of typing crm status, you can type crm st or crm stat. Instead of crm configure you can type crm cfg or even crm cf. crm resource can be shorted as crm rsc, and so on.

The exact list of accepted aliases is too long to print in full, but experimentation and typos should help in discovering more of them.


The feature set of crmsh covers a wide range of functionality, and understanding how and when to use the various features of the shell can be difficult. This section of the guide describes some of the features and use cases of crmsh in more depth. The intention is to provide a deeper understanding of these features, but also to serve as a guide to using them.

Shadow CIB usage

A Shadow CIB is a normal cluster configuration stored in a file. They may be manipulated in much the same way as the live CIB, with the key difference that changes to a shadow CIB have no effect on the actual cluster resources. An administrator may choose to apply any of them to the cluster, thus replacing the running configuration with the one found in the shadow CIB.

The crm prompt always contains the name of the configuration which is currently in use, or the string live if using the live cluster configuration.

When editing the configuration in the configure level, no changes are actually applied until the commit command is executed. It is possible to start editing a configuration as usual, but instead of committing the changes to the active CIB, save them to a shadow CIB.

The following example configure session demonstrates how this can be done:

crm(live)configure# cib new test-2
INFO: test-2 shadow CIB created
crm(test-2)configure# commit

Configuration semantic checks

Resource definitions may be checked against the meta-data provided with the resource agents. These checks are currently carried out:

•are required parameters set

•existence of defined parameters

•timeout values for operations

The parameter checks are obvious and need no further explanation. Failures in these checks are treated as configuration errors.

The timeouts for operations should be at least as long as those recommended in the meta-data. Too short timeout values are a common mistake in cluster configurations and, even worse, they often slip through if cluster testing was not thorough. Though operation timeouts issues are treated as warnings, make sure that the timeouts are usable in your environment. Note also that the values given are just advisory minimum---your resources may require longer timeouts.

User may tune the frequency of checks and the treatment of errors by the check-frequency and check-mode preferences.

Note that if the check-frequency is set to always and the check-mode to strict, errors are not tolerated and such configuration cannot be saved.

Configuration templates

Deprecation note

Configuration templates have been deprecated in favor of the more capable cluster scripts. To learn how to use cluster scripts, see the dedicated documentation on the crmsh website at, or in the Script section.

Configuration templates are ready made configurations created by cluster experts. They are designed in such a way so that users may generate valid cluster configurations with minimum effort. If you are new to Pacemaker, templates may be the best way to start.

We will show here how to create a simple yet functional Apache configuration:

# crm configure
crm(live)configure# template
crm(live)configure template# list templates
apache       filesystem   virtual-ip
crm(live)configure template# new web <TAB><TAB>
apache       filesystem   virtual-ip
crm(live)configure template# new web apache
INFO: pulling in template apache
INFO: pulling in template virtual-ip
crm(live)configure template# list
web2-d       web2     vip2     web3     vip      web

We enter the template level from configure. Use the list command to show templates available on the system. The new command creates a configuration from the apache template. You can use tab completion to pick templates. Note that the apache template depends on a virtual IP address which is automatically pulled along. The list command shows the just created web configuration, among other configurations (I hope that you, unlike me, will use more sensible and descriptive names).

The show command, which displays the resulting configuration, may be used to get an idea about the minimum required changes which have to be done. All ERROR messages show the line numbers in which the respective parameters are to be defined:

crm(live)configure template# show
ERROR: 23: required parameter ip not set
ERROR: 61: required parameter id not set
ERROR: 65: required parameter configfile not set
crm(live)configure template# edit

The edit command invokes the preferred text editor with the web configuration. At the top of the file, the user is advised how to make changes. A good template should require from the user to specify only parameters. For example, the web configuration we created above has the following required and optional parameters (all parameter lines start with %%):

$ grep -n ^%% ~/.crmconf/web
23:%% ip
31:%% netmask
35:%% lvs_support
61:%% id
65:%% configfile
71:%% options
76:%% envfiles

These lines are the only ones that should be modified. Simply append the parameter value at the end of the line. For instance, after editing this template, the result could look like this (we used tabs instead of spaces to make the values stand out):

$ grep -n ^%% ~/.crmconf/web
23:%% ip
31:%% netmask
35:%% lvs_support
61:%% id        websvc
65:%% configfile    /etc/apache2/httpd.conf
71:%% options
76:%% envfiles

As you can see, the parameter line format is very simple:

%% <name> <value>

After editing the file, use show again to display the configuration:

crm(live)configure template# show
primitive virtual-ip IPaddr \

params ip= primitive apache apache \
params configfile="/etc/apache2/httpd.conf" monitor apache 120s:60s group websvc \
apache virtual-ip

The target resource of the apache template is a group which we named websvc in this sample session.

This configuration looks exactly as you could type it at the configure level. The point of templates is to save you some typing. It is important, however, to understand the configuration produced.

Finally, the configuration may be applied to the current crm configuration (note how the configuration changed slightly, though it is still equivalent, after being digested at the configure level):

crm(live)configure template# apply
crm(live)configure template# cd ..
crm(live)configure# show
node xen-b
node xen-c
primitive apache apache \

params configfile="/etc/apache2/httpd.conf" \
op monitor interval=120s timeout=60s primitive virtual-ip IPaddr \
params ip= group websvc apache virtual-ip

Note that this still does not commit the configuration to the CIB which is used in the shell, either the running one (live) or some shadow CIB. For that you still need to execute the commit command.

To complete our example, we should also define the preferred node to run the service:

crm(live)configure# location websvc-pref websvc 100: xen-b

If you are not happy with some resource names which are provided by default, you can rename them now:

crm(live)configure# rename virtual-ip intranet-ip
crm(live)configure# show
node xen-b
node xen-c
primitive apache apache \

params configfile="/etc/apache2/httpd.conf" \
op monitor interval=120s timeout=60s primitive intranet-ip IPaddr \
params ip= group websvc apache intranet-ip location websvc-pref websvc 100: xen-b

To summarize, working with templates typically consists of the following steps:

•new: create a new configuration from templates

•edit: define parameters, at least the required ones

•show: see if the configuration is valid

•apply: apply the configuration to the configure level

Resource testing

The amount of detail in a cluster makes all configurations prone to errors. By far the largest number of issues in a cluster is due to bad resource configuration. The shell can help quickly diagnose such problems. And considerably reduce your keyboard wear.

Let’s say that we entered the following configuration:

node xen-b
node xen-c
node xen-d
primitive fencer stonith:external/libvirt \

params hypervisor_uri="qemu+tcp://" \
hostlist="xen-b xen-c xen-d" \
op monitor interval=2h primitive svc Xinetd \
params service=systat \
op monitor interval=30s primitive intranet-ip IPaddr2 \
params ip= \
op monitor interval=30s primitive apache apache \
params configfile="/etc/apache2/httpd.conf" \
op monitor interval=120s timeout=60s group websvc apache intranet-ip location websvc-pref websvc 100: xen-b

Before typing commit to submit the configuration to the cib we can make sure that all resources are usable on all nodes:

crm(live)configure# rsctest websvc svc fencer

It is important that resources being tested are not running on any nodes. Otherwise, the rsctest command will refuse to do anything. Of course, if the current configuration resides in a CIB shadow, then a commit is irrelevant. The point being that resources are not running on any node. Note on stopping all resources

Alternatively to not committing a configuration, it is also possible to tell Pacemaker not to start any resources:

crm(live)configure# property stop-all-resources=yes

Almost none---resources of class stonith are still started. But shell is not as strict when it comes to stonith resources.

Order of resources is significant insofar that a resource depends on all resources to its left. In most configurations, it’s probably practical to test resources in several runs, based on their dependencies.

Apart from groups, crm does not interpret constraints and therefore knows nothing about resource dependencies. It also doesn’t know if a resource can run on a node at all in case of an asymmetric cluster. It is up to the user to specify a list of eligible nodes if a resource is not meant to run on every node.

Access Control Lists (ACL)

Note on ACLs in Pacemaker 1.1.12

The support for ACLs has been revised in Pacemaker version 1.1.12 and up. Depending on which version you are using, the information in this section may no longer be accurate. Look for the acl_target configuration element for more details on the new syntax.

By default, the users from the haclient group have full access to the cluster (or, more precisely, to the CIB). Access control lists allow for finer access control to the cluster.

Access control lists consist of an ordered set of access rules. Each rule allows read or write access or denies access completely. Rules are typically combined to produce a specific role. Then, users may be assigned a role.

For instance, this is a role which defines a set of rules allowing management of a single resource:

role bigdb_admin \

write meta:bigdb:target-role \
write meta:bigdb:is-managed \
write location:bigdb \
read ref:bigdb

The first two rules allow modifying the target-role and is-managed meta attributes which effectively enables users in this role to stop/start and manage/unmanage the resource. The constraints write access rule allows moving the resource around. Finally, the user is granted read access to the resource definition.

For proper operation of all Pacemaker programs, it is advisable to add the following role to all users:

role read_all \

read cib

For finer grained read access try with the rules listed in the following role:

role basic_read \

read node attribute:uname \
read node attribute:type \
read property \
read status

It is however possible that some Pacemaker programs (e.g. ptest) may not function correctly if the whole CIB is not readable.

Some of the ACL rules in the examples above are expanded by the shell to XPath specifications. For instance, meta:bigdb:target-role expands to:


You can see the expansion by showing XML:

crm(live) configure# show xml bigdb_admin

<acl_role id="bigdb_admin">
<write id="bigdb_admin-write"

Many different XPath expressions can have equal meaning. For instance, the following two are equal, but only the first one is going to be recognized as shortcut:


XPath is a powerful language, but you should try to keep your ACL xpaths simple and the builtin shortcuts should be used whenever possible.

Syntax: Resource sets

Using resource sets can be a bit confusing unless one knows the details of the implementation in Pacemaker as well as how to interpret the syntax provided by crmsh.

Three different types of resource sets are provided by crmsh, and each one implies different values for the two resource set attributes, sequential and require-all.


If false, the resources in the set do not depend on each other internally. Setting sequential to true implies a strict order of dependency within the set.


If false, only one resource in the set is required to fulfil the requirements of the set. The set of A, B and C with require-all set to false is be read as "A OR B OR C" when its dependencies are resolved.

The three types of resource sets modify the attributes in the following way:

1.Implicit sets (no brackets). sequential=true, require-all=true

2.Parenthesis set (( ... )). sequential=false, require-all=true

3.Bracket set ([ ... ]). sequential=false, require-all=false

To create a set with the properties sequential=true and require-all=false, explicitly set sequential in a bracketed set, [ A B C sequential=true ].

To create multiple sets with both sequential and require-all set to true, explicitly set sequential in a parenthesis set: A B ( C D sequential=true ).

Syntax: Attribute list references

Attribute lists are used to set attributes and parameters for resources, constraints and property definitions. For example, to set the virtual IP used by an IPAddr2 resource the attribute ip can be set in an attribute list for that resource.

Attribute lists can have identifiers that name them, and other resources can reuse the same attribute list by referring to that name using an $id-ref. For example, the following statement defines a simple dummy resource with an attribute list which sets the parameter state to the value 1 and sets the identifier for the attribute list to on-state:

primitive dummy-1 Dummy params $id=on-state state=1

To refer to this attribute list from a different resource, refer to the on-state name using an id-ref:

primitive dummy-2 Dummy params $id-ref=on-state

The resource dummy-2 will now also have the parameter state set to the value 1.

Syntax: Attribute references

In some cases, referencing complete attribute lists is too coarse-grained, for example if two different parameters with different names should have the same value set. Instead of having to copy the value in multiple places, it is possible to create references to individual attributes in attribute lists.

To name an attribute in order to be able to refer to it later, prefix the attribute name with a $ character (as seen above with the special names $id and $id-ref:

primitive dummy-1 Dummy params $state=1

The identifier state can now be used to refer to this attribute from other primitives, using the @<id> syntax:

primitive dummy-2 Dummy params @state

In some cases, using the attribute name as the identifier doesn’t work due to name clashes. In this case, the syntax $<id>:<name>=<value> can be used to give the attribute a different identifier:

primitive dummy-1 params $dummy-state-on:state=1
primitive dummy-2 params @dummy-state-on

There is also the possibility that two resources both use the same attribute value but with different names. For example, a web server may have a parameter server_ip for setting the IP address where it listens for incoming requests, and a virtual IP resource may have a parameter called ip which sets the IP address it creates. To configure these two resources with an IP without repeating the value, the reference can be given a name using the syntax @<id>:<name>.


primitive virtual-ip IPaddr2 params $vip:ip=
primitive webserver apache params @vip:server_ip

Syntax: Rule expressions

Many of the configuration commands in crmsh now support the use of rule expressions, which can influence what attributes apply to a resource or under which conditions a constraint is applied, depending on changing conditions like date, time, the value of attributes and more.

Here is an example of a simple rule expression used to apply a a different resource parameter on the node named node1:

primitive my_resource Special \

params 2: rule #uname eq node1 interface=eth1 \
params 1: interface=eth0

This primitive resource has two lists of parameters with descending priority. The parameter list with the highest priority is applied first, but only if the rule expressions for that parameter list all apply. In this case, the rule #uname eq node1 limits the parameter list so that it is only applied on node1.

Note that rule expressions are not terminated and are immediately followed by the data to which the rule is applied. In this case, the name-value pair interface=eth1.

Rule expressions can contain multiple expressions connected using the boolean operator or and and. The full syntax for rule expressions is listed below.

rules ::

rule [id_spec] [$role=<role>] <score>: <expression>
[rule [id_spec] [$role=<role>] <score>: <expression> ...] id_spec :: $id=<id> | $id-ref=<id> score :: <number> | <attribute> | [-]inf expression :: <simple_exp> [<bool_op> <simple_exp> ...] bool_op :: or | and simple_exp :: <attribute> [type:]<binary_op> <value>
| <unary_op> <attribute>
| date <date_expr> type :: <string> | <version> | <number> binary_op :: lt | gt | lte | gte | eq | ne unary_op :: defined | not_defined date_expr :: lt <end>
| gt <start>
| in start=<start> end=<end>
| in start=<start> <duration>
| spec <date_spec> duration|date_spec ::
| monthdays=<value>
| weekdays=<value>
| yearsdays=<value>
| months=<value>
| weeks=<value>
| years=<value>
| weekyears=<value>
| moon=<value>


Lifetimes can be specified in the ISO 8601 time format or the ISO 8601 duration format. To distinguish between months and minutes, use the PT prefix before specifying minutes. The duration format is one of PnYnMnDTnHnMnS, PnW, P<date>T<time>.

P = duration. Y = year. M = month. W = week. D = day. T = time. H = hour. M = minute. S = second.


PT5M = 5 minutes later.
3D = 3 days later.
PT1H = 1 hour later.

The cluster checks lifetimes at an interval defined by the cluster-recheck-interval property (default 15 minutes).


The commands are structured to be compatible with the shell command line. Sometimes, the underlying Pacemaker grammar uses characters that have special meaning in bash, that will need to be quoted. This includes the hash or pound sign (#), single and double quotes, and any significant whitespace.

Whitespace is also significant when assigning values, meaning that key=value is different from key = value.

Commands can be referenced using short-hand as long as the short-hand is unique. This can be either a prefix of the command name or a prefix string of characters found in the name.

For example, status can be abbreviated as st or su, and configure as conf or cfg.

The syntax for the commands is given below in an informal, BNF-like grammar.

•<value> denotes a string.

•[value] means that the construct is optional.

•The ellipsis (...) signifies that the previous construct may be repeated.

•first|second means either first or second.

•The rest are literals (strings, :, =).


Show cluster status. The status is displayed by crm_mon. Supply additional arguments for more information or different format. See crm_mon(8) for more details.


status simple
status full


status [<option> ...]
option :: full

| bynode
| inactive
| ops
| timing
| failcounts
| verbose
| quiet
| html
| xml
| simple
| tickets
| noheaders
| detail
| brief


Performs basic checks for the cluster configuration and current status, reporting potential issues.

See crm_verify(8) and crm_simulate(8) for more details.


verify scores


verify [scores]

cluster - Cluster setup and management

Whole-cluster configuration management with High Availability awareness.

The commands on the cluster level allows configuration and modification of the underlying cluster infrastructure, and also supplies tools to do whole-cluster systems management.

These commands enable easy installation and maintenance of a HA cluster, by providing support for package installation, configuration of the cluster messaging layer, file system setup and more.


Copy file to other cluster nodes.

Copies the given file to all other nodes unless given a list of nodes to copy to as argument.


copy <filename> [nodes ...]


copy /etc/motd


Displays the difference, if any, between a given file on different nodes. If the second argument is --checksum, a checksum of the file will be calculated and displayed for each node.


diff <file> [--checksum] [nodes...]


diff /etc/crm/crm.conf node2
diff /etc/resolv.conf --checksum


See "crm cluster help disable" or "crm cluster disable --help"


See "crm cluster help enable" or "crm cluster enable --help


See "crm cluster help geo_init" or "crm cluster geo_init --help"


See "crm cluster help geo_init_arbitrator" or "crm cluster geo_init_arbitrator --help"


See "crm cluster help geo_join" or "crm cluster geo_join --help"


Runs a larger set of tests and queries on all nodes in the cluster to verify the general system health and detect potential problems.




See "crm cluster help init" or "crm cluster init --help"


See "crm cluster help join" or "crm cluster join --help"


See "crm cluster help remove" or "crm cluster remove --help"


See "crm cluster help crash_test" or "crm cluster crash_test --help"


See "crm cluster help restart" or "crm cluster restart --help"


Rename the cluster name


rename <new_cluster_name>


This command takes a shell statement as argument, executes that statement on all nodes in the cluster or a specific node, and reports the result.


run <command> [node ...]


run "cat /proc/uptime"
run "ls" node1 node2


See "crm cluster help start" or "crm cluster start --help"


Reports the status for the cluster messaging layer on the local node.




See "crm cluster help stop" or "crm cluster stop --help"


Mostly useful in scripts or automated workflows, this command will attempt to connect to the local cluster node repeatedly. The command will keep trying until the cluster node responds, or the timeout elapses. The timeout can be changed by supplying a value in seconds as an argument.



script - Cluster script management

A big part of the configuration and management of a cluster is collecting information about all cluster nodes and deploying changes to those nodes. Often, just performing the same procedure on all nodes will encounter problems, due to subtle differences in the configuration.

For example, when configuring a cluster for the first time, the software needs to be installed and configured on all nodes before the cluster software can be launched and configured using crmsh. This process is cumbersome and error-prone, and the goal is for scripts to make this process easier.

Scripts are implemented using the python parallax package which provides a thin wrapper on top of SSH. This allows the scripts to function through the usual SSH channels used for system maintenance, requiring no additional software to be installed or maintained.


This command provides a JSON API for the cluster scripts, intended for use in user interface tools that want to interact with the cluster via scripts.

The command takes a single argument, which should be a JSON array with the first member identifying the command to perform.

The output is line-based: Commands that return multiple results will return them line-by-line, ending with a terminator value: "end".

When providing parameter values to this command, they should be provided as nested objects, so virtual-ip:ip= on the command line becomes the JSON object {"virtual-ip":{"ip":""}}.


=> [{name, shortdesc, category}]
["show", <name>]
=> [{name, shortdesc, longdesc, category, <<steps>>}]
<<steps>> := [{name, shortdesc], longdesc, required, parameters, steps}]
<<params>> := [{name, shortdesc, longdesc, required, unique, advanced,

type, value, example}] ["verify", <name>, <<values>>] => [{shortdesc, longdesc, text, nodes}] ["run", <name>, <<values>>] => [{shortdesc, rc, output|error}]


Lists the available scripts, sorted by category. Scripts that have the special Script category are hidden by default, since they are mainly used by other scripts or commands. To also show these, pass all as argument.

To get a flat list of script names, not sorted by category, pass names as an extra argument.


list [all] [names]


list all names


Given a list of parameter values, this command will execute the actions specified by the cluster script. The format for the parameter values is the same as for the verify command.

Can optionally take at least two parameters: * nodes=<nodes>: List of nodes that the script runs over * dry_run=yes|no: If set, the script will not perform any modifications.

Additional parameters may be available depending on the script.

Use the show command to see what parameters are available.


run <script> [args...]


run apache install=true
run sbd id=sbd-1 node=node1 sbd_device=/dev/disk/by-uuid/F00D-CAFE


Prints a description and short summary of the script, with descriptions of the accepted parameters.

Advanced parameters are hidden by default. To show the complete list of parameters accepted by the script, pass all as argument.


show <script> [all]


show virtual-ip


Checks the given parameter values, and returns a list of actions that will be executed when running the script if provided the same list of parameter values.


verify <script> [args...]


verify sbd id=sbd-1 node=node1 sbd_device=/dev/disk/by-uuid/F00D-CAFE

corosync - Corosync management

Corosync is the underlying messaging layer for most HA clusters. This level provides commands for editing and managing the corosync configuration.


Adds a node to the corosync configuration. This is used with the udpu type configuration in corosync.

A nodeid for the added node is generated automatically.

Note that this command assumes that only a single ring is used, and sets only the address for ring0.


add-node <addr> [name]


Removes a node from the corosync configuration. The argument given is the ring0_addr address set in the configuration file.


del-node <addr>


Diffs the corosync configurations on different nodes. If no nodes are given as arguments, the corosync configurations on all nodes in the cluster are compared.

diff takes an option argument --checksum, to display a checksum for each file instead of calculating a diff.


diff [--checksum] [node...]


Opens the Corosync configuration file in an editor.




Returns the value configured in corosync.conf, which is not necessarily the value used in the running configuration. See reload for telling corosync about configuration changes.

The argument is the complete dot-separated path to the value.

If there are multiple values configured with the same path, the command returns all values for that path. For example, to get all configured ring0_addr values, use this command:


get nodelist.node.ring0_addr


Opens the log file specified in the corosync configuration file. If no log file is configured, this command returns an error.

The pager used can be configured either using the PAGER environment variable or in crm.conf.




Gets the corosync configuration from another node and copies it to this node.


pull <node>


Pushes the corosync configuration file on this node to the list of nodes provided. If no target nodes are given, the configuration is pushed to all other nodes in the cluster.

It is recommended to use csync2 to distribute the cluster configuration files rather than relying on this command.


push [node] ...


push node-2 node-3


Tells all instances of corosync in this cluster to reload corosync.conf.

After pushing a new configuration to all cluster nodes, call this command to make corosync use the new configuration.




Sets the value identified by the given path. If the value does not exist in the configuration file, it will be added. However, if the section containing the value does not exist, the command will fail.


set quorum.expected_votes 2


Displays the corosync configuration on the current node.



Displays the corosync ring status(default), also can display quorum/qdevice/qnetd status.


status [ring|quorum|qdevice|qnetd]

cib - CIB shadow management

This level is for management of shadow CIBs. It is available both at the top level and the configure level.

All the commands are implemented using cib_shadow(8) and the CIB_shadow environment variable. The user prompt always includes the name of the currently active shadow or the live CIB.


Enter edit and manage the CIB status section level. See the CIB status management section.


Apply a shadow CIB to the cluster. If the shadow name is omitted then the current shadow CIB is applied.

Temporary shadow CIBs are removed automatically on commit.


commit [<cib>]


Delete an existing shadow CIB.


delete <cib>


Print differences between the current cluster configuration and the active shadow CIB.




At times it may be useful to create a shadow file from the existing CIB. The CIB may be specified as file or as a PE input file number. The shell will look up files in the local directory first and then in the PE directory (typically /var/lib/pengine). Once the CIB file is found, it is copied to a shadow and this shadow is immediately available for use at both configure and cibstatus levels.

If the shadow name is omitted then the target shadow is named after the input CIB file.

Note that there are often more than one PE input file, so you may need to specify the full name.


import {<file>|<number>} [<shadow>]


import pe-warn-2222
import 2289 issue2


List existing shadow CIBs.




Create a new shadow CIB. The live cluster configuration and status is copied to the shadow CIB.

If the name of the shadow is omitted, we create a temporary CIB shadow. It is useful if multiple level sessions are desired without affecting the cluster. A temporary CIB shadow is short lived and will be removed either on commit or on program exit. Note that if the temporary shadow is not committed all changes in the temporary shadow are lost.

Specify withstatus if you want to edit the status section of the shadow CIB (see the cibstatus section). Add force to force overwriting the existing shadow CIB.

To start with an empty configuration that is not copied from the live CIB, specify the empty keyword. (This also allows a shadow CIB to be created in case no cluster is running.)


new [<cib>] [withstatus] [force] [empty]


Copy the current cluster configuration into the shadow CIB.


reset <cib>


Choose a CIB source. If you want to edit the status from the shadow CIB specify withstatus (see cibstatus). Leave out the CIB name to switch to the running CIB.


use [<cib>] [withstatus]

ra - Resource Agents (RA) lists and documentation

This level contains commands which show various information about the installed resource agents. It is available both at the top level and at the configure level.


Print all resource agents' classes and, where appropriate, a list of available providers.



info (meta)

Show the meta-data of a resource agent type. This is where users can find information on how to use a resource agent. It is also possible to get information from some programs: pengine, crmd, cib, and stonithd. Just specify the program name instead of an RA.


info [<class>:[<provider>:]]<type>
info <type> <class> [<provider>] (obsolete)


info apache
info ocf:pacemaker:Dummy
info stonith:ipmilan
info pengine


List available resource agents for the given class. If the class is ocf, supply a provider to get agents which are available only from that provider.


list <class> [<provider>]


list ocf pacemaker


List providers for a resource agent type. The class parameter defaults to ocf.


providers <type> [<class>]


providers apache


If the resource agent supports the validate-all action, this calls the action with the given parameters, printing any warnings or errors reported by the agent.


validate <agent> [<key>=<value> ...]

resource - Resource management

At this level resources may be managed.

All (or almost all) commands are implemented with the CRM tools such as crm_resource(8).


Ban a resource from running on a certain node. If no node is given as argument, the resource is banned from the current location.

See move for details on other arguments.


ban <rsc> [<node>] [<lifetime>] [force]


If resource has any past failures, clear its history and fail count. Typically done after the resource has temporarily failed.

If a node is omitted, cleanup on all nodes.

(Pacemaker 1.1.14) Pass force to cleanup the resource itself, otherwise the cleanup command will apply to the parent resource (if any).


cleanup [<rsc>] [<node>] [force]

clear (unmove, unmigrate, unban)

Remove any relocation constraint created by the move, migrate or ban command.


clear <rsc>
unmigrate <rsc>
unban <rsc>


Display the location and colocation constraints affecting the resource.


constraints <rsc>


Demote a promotable resource using the target-role attribute.


demote <rsc>


Show/edit/delete the failcount of a resource. When set a non-zero value, operation and interval should be provided when multiple operation failcount entries exist. interval is a value in seconds.


failcount <rsc> set <node> <value> [operation] [interval]
failcount <rsc> delete <node>
failcount <rsc> show <node>


failcount fs_0 delete node2


Show the current location of one or more resources.


locate [<rsc> ...]


Enables or disables the per-resource maintenance mode. When this mode is enabled, no monitor operations will be triggered for the resource. maintenance attribute conflicts with the is-managed. When setting the maintenance attribute, the user is proposed to remove the is-managed attribute if it exists.


maintenance <resource> [on|off|true|false]


maintenance rsc1
maintenance rsc2 off


Manage a resource using the is-managed attribute. If there are multiple meta attributes sets, the attribute is set in all of them. If the resource is a clone, all is-managed attributes are removed from the children resources. is-managed attribute conflicts with the maintenance. When setting the is-managed attribute, the user is proposed to remove the maintenance attribute if it exists.

For details on group management see options manage-children.


manage <rsc>


Show/edit/delete a meta attribute of a resource. Currently, all meta attributes of a resource may be managed with other commands such as resource stop.


meta <rsc> set <attr> <value>
meta <rsc> delete <attr>
meta <rsc> show <attr>


meta ip_0 set target-role stopped

move (migrate)

Move a resource away from its current location.

If the destination node is left out, the resource is migrated by creating a constraint which prevents it from running on the current node. For this type of constraint to be created, the force argument is required.

A lifetime may be given for the constraint. Once it expires, the location constraint will no longer be active.


move <rsc> [<node>] [<lifetime>] [force]


Show active operations, optionally filtered by resource and node.


operations [<rsc>] [<node>]


Show/edit/delete a parameter of a resource.


param <rsc> set <param> <value>
param <rsc> delete <param>
param <rsc> show <param>


param ip_0 show ip


Promote a promotable resource using the target-role attribute.


promote <rsc>


Delete resource’s history (including failures) so its current state is rechecked.


refresh [<rsc>] [<node>] [force]


Restart one or more resources. This is essentially a shortcut for resource stop followed by a start. The shell is first going to wait for the stop to finish, that is for all resources to really stop, and only then to order the start action. Due to this command entailing a whole set of operations, informational messages are printed to let the user see some progress.

For details on group management see options manage-children.


restart <rsc> [<rsc> ...]


# crm resource restart g_webserver
INFO: ordering g_webserver to stop
waiting for stop to finish .... done
INFO: ordering g_webserver to start


Display the allocation scores for all resources.




Sensitive parameters can be kept in local files rather than CIB in order to prevent accidental data exposure. Use the secret command to manage such parameters. stash and unstash move the value from the CIB and back to the CIB respectively. The set subcommand sets the parameter to the provided value. delete removes the parameter completely. show displays the value of the parameter from the local file. Use check to verify if the local file content is valid.


secret <rsc> set <param> <value>
secret <rsc> stash <param>
secret <rsc> unstash <param>
secret <rsc> delete <param>
secret <rsc> show <param>
secret <rsc> check <param>


secret fence_1 show password
secret fence_1 stash password
secret fence_1 set password secret_value


Start one or more resources by setting the target-role attribute. If there are multiple meta attributes sets, the attribute is set in all of them. If the resource is a clone, all target-role attributes are removed from the children resources.

For details on group management see options manage-children.


start <rsc> [<rsc> ...]

status (show, list)

Print resource status. More than one resource can be shown at once. If the resource parameter is left out, the status of all resources is printed.


status [<rsc> ...]


Stop one or more resources using the target-role attribute. If there are multiple meta attributes sets, the attribute is set in all of them. If the resource is a clone, all target-role attributes are removed from the children resources.

For details on group management see options manage-children.


stop <rsc> [<rsc> ...]


Start tracing RA for the given operation. When [<log-dir>] is not specified the trace files are stored in $HA_VARLIB/trace_ra. If the operation to be traced is monitor, note that the number of trace files can grow very quickly.

If no operation name is given, crmsh will attempt to trace all operations for the RA. This includes any configured operations, start and stop as well as promote/demote for multistate resources.

To trace the probe operation which exists for all resources, either set a trace for monitor with interval 0, or use probe as the operation name.

Note: RA tracing is only supported by OCF resource agents; The pacemaker-execd daemon does not log recurring monitor operations unless an error occurred.


trace <rsc> [<op> [<interval>] [<log-dir>]]


trace fs start
trace webserver
trace webserver probe
trace fs monitor 0 /var/log/foo/bar


Unmanage a resource using the is-managed attribute. If there are multiple meta attributes sets, the attribute is set in all of them. If the resource is a clone, all is-managed attributes are removed from the children resources.

For details on group management see options manage-children.


unmanage <rsc>


Stop tracing RA for the given operation. If no operation name is given, crmsh will attempt to stop tracing all operations in resource.


untrace <rsc> [<op> [<interval>] ]


untrace fs start
untrace webserver


Show/edit/delete a utilization attribute of a resource. These attributes describe hardware requirements. By setting the placement-strategy cluster property appropriately, it is possible then to distribute resources based on resource requirements and node size. See also node utilization attributes.


utilization <rsc> set <attr> <value>
utilization <rsc> delete <attr>
utilization <rsc> show <attr>


utilization xen1 set memory 4096

node - Node management

Node management and status commands.


Edit node attributes. This kind of attribute should refer to relatively static properties, such as memory size.


attribute <node> set <attr> <value>
attribute <node> delete <attr>
attribute <node> show <attr>


attribute node_1 set memory_size 4096


Resets and clears the state of the specified node. This node is afterwards assumed clean and offline. This command can be used to manually confirm that a node has been fenced (e.g., powered off).

Be careful! This can cause data corruption if you confirm that a node is down that is, in fact, not cleanly down - the cluster will proceed as if the fence had succeeded, possibly starting resources multiple times.


clearstate <node>


Delete a node. This command will remove the node from the CIB and, in case the cluster stack is running, use the appropriate program (crm_node or hb_delnode) to remove the node from the membership.

If the node is still listed as active and a member of our partition we refuse to remove it. With the global force option (-F) we will try to delete the node anyway.


delete <node>


Make CRM fence a node. This functionality depends on stonith resources capable of fencing the specified node. No such stonith resources, no fencing will happen.


fence <node>


Set the node status to maintenance. This is equivalent to the cluster-wide maintenance-mode property but puts just one node into the maintenance mode. If there are maintenaned resources on the node, the user will be proposed to remove the maintenance property from them.

The node parameter defaults to the node where the command is run.


maintenance [<node>]


See "crm node help online" or "crm node online --help"


Set the node’s maintenance status to off. The node should be now again fully operational and capable of running resource operations.

The node parameter defaults to the node where the command is run.


ready [<node>]


Remote nodes may have a configured server address which should be used when contacting the node. This command prints the server address if configured, else the node name.

If no parameter is given, the addresses or names for all nodes are printed.


server [<node> ...]


Show a node definition. If the node parameter is omitted then all nodes are shown.


show [<node>]


See "crm node help standby" or "crm node standby --help"


Edit node attributes which are in the CIB status section, i.e. attributes which hold properties of a more volatile nature. One typical example is attribute generated by the pingd utility.


status-attr <node> set <attr> <value>
status-attr <node> delete <attr>
status-attr <node> show <attr>


status-attr node_1 show pingd


Edit node utilization attributes. These attributes describe hardware characteristics as integer numbers such as memory size or the number of CPUs. By setting the placement-strategy cluster property appropriately, it is possible then to distribute resources based on resource requirements and node size. See also resource utilization attributes.


utilization <node> set <attr> <value>
utilization <node> delete <attr>
utilization <node> show <attr>


utilization node_1 set memory 16384
utilization node_1 show cpu

site - GEO clustering site support

A cluster may consist of two or more subclusters in different and distant locations. This set of commands supports such setups.


Tickets are cluster-wide attributes. They can be managed at the site where this command is executed.

It is then possible to constrain resources depending on the ticket availability (see the rsc_ticket command for more details).


ticket {grant|revoke|standby|activate|show|time|delete} <ticket>


ticket grant ticket1

options - User preferences

The user may set various options for the crm shell itself.


The shell (as in /bin/sh) parser strips quotes from the command line. This may sometimes make it really difficult to type values which contain white space. One typical example is the configure filter command. The crm shell will supply extra quotes around arguments which contain white space. The default is yes. Note on quotes use

Adding quotes around arguments automatically has been introduced with version 1.2.2 and it is technically a regression. Being a regression is the only reason the add-quotes option exists. If you have custom shell scripts which would break, just set the add-quotes option to no.

For instance, with adding quotes enabled, it is possible to do the following:

# crm configure primitive d1 Dummy \

meta description="some description here" # crm configure filter 'sed "s/hostlist=./&node-c /"' fencing


Semantic check of the CIB or elements modified or created may be done on every configuration change (always), when verifying (on-verify) or never. It is by default set to always. Experts may want to change the setting to on-verify.

The checks require that resource agents are present. If they are not installed at the configuration time set this preference to never.

See Configuration semantic checks for more details.


Semantic check of the CIB or elements modified or created may be done in the strict mode or in the relaxed mode. In the former certain problems are treated as configuration errors. In the relaxed mode all are treated as warnings. The default is strict.

See Configuration semantic checks for more details.


With output set to color, a comma separated list of colors from this option are used to emphasize:


•object ids

•attribute names

•attribute values


•resource references

crm can show colors only if there is curses support for python installed (usually provided by the python-curses package). The colors are whatever is available in your terminal. Use normal if you want to keep the default foreground color.

This user preference defaults to yellow,normal,cyan,red,green,magenta which is good for terminals with dark background. You may want to change the color scheme and save it in the preferences file for other color setups.


colorscheme yellow,normal,blue,red,green,magenta


The edit command invokes an editor. Use this to specify your preferred editor program. If not set, it will default to either the value of the EDITOR environment variable or to one of the standard UNIX editors (vi,emacs,nano).


editor program


editor vim


Some resource management commands, such as resource stop, when the target resource is a group, may not always produce desired result. Each element, group and the primitive members, can have a meta attribute and those attributes may end up with conflicting values. Consider the following construct:

crm(live)# configure show svc fs virtual-ip
primitive fs Filesystem \

params device="/dev/drbd0" directory="/srv/nfs" fstype=ext3 \
op monitor interval=10s \
meta target-role=Started primitive virtual-ip IPaddr2 \
params ip= iflabel=1 \
op monitor interval=10s \
op start interval=0 \
meta target-role=Started group svc fs virtual-ip \
meta target-role=Stopped

Even though the element svc should be stopped, the group is actually running because all its members have the target-role set to Started:

crm(live)# resource show svc
resource svc is running on: xen-f

Hence, if the user invokes resource stop svc the intention is not clear. This preference gives the user an opportunity to better control what happens if attributes of group members have values which are in conflict with the same attribute of the group itself.

Possible values are ask (the default), always, and never. If set to always, the crm shell removes all children attributes which have values different from the parent. If set to never, all children attributes are left intact. Finally, if set to ask, the user will be asked for each member what is to be done.


crm can adorn configurations in two ways: in color (similar to for instance the ls --color command) and by showing keywords in upper case. Possible values are plain, color-always, color, and uppercase. It is possible to combine uppercase with one of the color values in order to get an upper case xmass tree. Just set this option to color,uppercase or color-always,uppercase. In case you need color codes in pipes, color-always forces color codes even in case the terminal is not a tty (just like ls --color=always).


The view command displays text through a pager. Use this to specify your preferred pager program. If not set, it will default to either the value of the PAGER environment variable or to one of the standard UNIX system pagers (less,more,pg).


This command resets all user options to the defaults. If used as a single-shot command, the rc file ($HOME/.config/crm/rc) is reset to the defaults too.


Save current settings to the rc file ($HOME/.config/crm/rc). On further crm runs, the rc file is automatically read and parsed.


Sets the value of an option. Takes the fully qualified name of the option as argument, as displayed by show all.

The modified option value is stored in the user-local configuration file, usually found in ~/.config/crm/crm.conf.


set <option> <value>


set color.warn "magenta bold"
set editor nano


Display all current settings.

Given an option name as argument, show will display only the value of that argument.

Given all as argument, show displays all available user options.


show [all|<option>]


show skill-level
show all


Based on the skill-level setting, the user is allowed to use only a subset of commands. There are three levels: operator, administrator, and expert. The operator level allows only commands at the resource and node levels, but not editing or deleting resources. The administrator may do that and may also configure the cluster at the configure level and manage the shadow CIBs. The expert may do all.


skill-level <level>
level :: operator | administrator | expert
Note on security

The skill-level option is advisory only. There is nothing stopping any users change their skill level (see Access Control Lists (ACL) on how to enforce access control).


crm by default sorts CIB elements. If you want them appear in the order they were created, set this option to no.


sort-elements {yes|no}


sort-elements no


Sufficient privileges are necessary in order to manage a cluster: programs such as crm_verify or crm_resource and, ultimately, cibadmin have to be run either as root or as the CRM owner user (typically hacluster). You don’t have to worry about that if you run crm as root. A more secure way is to run the program with your usual privileges, set this option to the appropriate user (such as hacluster), and setup the sudoers file.


user system-user


user hacluster


In normal operation, crm runs a command and gets back immediately to process other commands or get input from the user. With this option set to yes it will wait for the started transition to finish. In interactive mode dots are printed to indicate progress.


wait {yes|no}


wait yes

configure - CIB configuration

This level enables all CIB object definition commands.

The configuration may be logically divided into four parts: nodes, resources, constraints, and (cluster) properties and attributes. Each of these commands support one or more basic CIB objects.

Nodes and attributes describing nodes are managed using the node command.

Commands for resources are:




•clone (promotable clones)

•ms/master (master-slave) (deprecated)

In order to streamline large configurations, it is possible to define a template which can later be referenced in primitives:


In that case the primitive inherits all attributes defined in the template.

There are three types of constraints:




It is possible to define fencing order (stonith resource priorities):


Finally, there are the cluster properties, resource meta attributes defaults, and operations defaults. All are just a set of attributes. These attributes are managed by the following commands:




In addition to the cluster configuration, the Access Control Lists (ACL) can be setup to allow access to parts of the CIB for users other than root and hacluster. The following commands manage ACL:



In Pacemaker 1.1.12 and up, this command replaces the user command for handling ACLs:


The changes are applied to the current CIB only on ending the configuration session or using the commit command.

Comments start with # in the first line. The comments are tied to the element which follows. If the element moves, its comments will follow.


Defines an ACL target.


acl_target <tid> [<role> ...]


acl_target joe resource_admin constraint_editor


Version note

This feature is only available in Pacemaker 1.1.15+.

Event-driven alerts enables calling scripts whenever interesting events occur in the cluster (nodes joining or leaving, resources starting or stopping, etc.).

The path is an arbitrary file path to an alert script. Existing external scripts used with ClusterMon resources can be used as alert scripts, since the interface is compatible.

Each alert may have a number of receipients configured. These will be passed to the script as arguments. The first recipient will also be passed as the CRM_alert_recipient environment variable, for compatibility with existing scripts that only support one recipient.

The available meta attributes are timeout (default 30s) and timestamp-format (default "%H:%M:%S.%06N").

Some configurations may require each recipient to be delimited by brackets, to avoid ambiguity. In the example alert-2 below, the meta attribute for timeout is defined after the recipient, so the brackets are used to ensure that the meta attribute is set for the alert and not just the recipient. This can be avoided by setting any alert attributes before defining the recipients.


alert <id> <path> \

[attributes <nvpair> ...] \
[meta <nvpair> ...] \
[select [nodes | fencing | resources | attributes '{' <attribute> ... '}' ] ...] \
[to [{] <recipient>
[attributes <nvpair> ...] \
[meta <nvpair> ...] [}] \


alert alert-1 /srv/pacemaker/ \

to /var/log/cluster-alerts.log alert alert-2 /srv/pacemaker/ \
meta timeout=60s \
to { /var/log/cluster-alerts.log } alert alert-3 /srv/pacemaker/ \
select fencing \
to { /var/log/fencing-alerts.log }


A bundle is a single resource specifying the settings, networking requirements, and storage requirements for any number of containers generated from the same container image.

Pacemaker bundles support Docker (since version 1.1.17) and rkt (since version 1.1.18) container technologies.

A bundle must contain exactly one docker or rkt element.

The bundle definition may contain a reference to a primitive resource which defining the resource running inside the container.


primitive httpd-apache ocf:heartbeat:apache
bundle httpd \

docker image=pcmk:httpd replicas=3 \
network ip-range-start= host-netmask=24 \
port-mapping port=80 \
storage \
storage-mapping target-dir=/var/www/html source-dir=/srv/www options=rw \
primitive httpd-apache


This level is for management of shadow CIBs. It is available at the configure level to enable saving intermediate changes to a shadow CIB instead of to the live cluster. This short excerpt shows how:

crm(live)configure# cib new test-2
INFO: test-2 shadow CIB created
crm(test-2)configure# commit

Note how the current CIB in the prompt changed from live to test-2 after issuing the cib new command. See also the CIB shadow management for more information.


Enter edit and manage the CIB status section level. See the CIB status management section.


The clone command creates a resource clone. It may contain a single primitive resource or one group of resources.

Promotable clones are clone resources with the promotable=true meta attribute for the given promotable resources. It’s used to deprecate the master-slave resources.


clone <name> <rsc>

[meta <attr_list>]
[params <attr_list>] attr_list :: [$id=<id>] <attr>=<val> [<attr>=<val>...] | $id-ref=<id>


clone cl_fence apc_1 \

meta clone-node-max=1 globally-unique=false clone disk1 drbd1 \
meta promotable=true notify=true globally-unique=false

colocation (collocation)

This constraint expresses the placement relation between two or more resources. If there are more than two resources, then the constraint is called a resource set.

The score is used to indicate the priority of the constraint. A positive score indicates that the resources should run on the same node. A negative score that they should not run on the same node. Values of positive or negative infinity indicate a mandatory constraint.

In the two resource form, the cluster will place <with-rsc> first, and then decide where to put the <rsc> resource.

Collocation resource sets have an extra attribute (sequential) to allow for sets of resources which don’t depend on each other in terms of state. The shell syntax for such sets is to put resources in parentheses.

Sets cannot be nested.

The optional node-attribute can be used to colocate resources on a set of nodes and not necessarily on the same node. For example, by setting a node attribute color on all nodes and setting the node-attribute value to color as well, the colocated resources will be placed on any node that has the same color.

For more details on how to configure resource sets, see Syntax: Resource sets.


colocation <id> <score>: <rsc>[:<role>] <with-rsc>[:<role>]

[node-attribute=<node_attr>] colocation <id> <score>: <resource_sets>
[node-attribute=<node_attr>] resource_sets :: <resource_set> [<resource_set> ...] resource_set :: ["("|"["] <rsc>[:<role>] [<rsc>[:<role>] ...] \
[<attributes>] [")"|"]"] attributes :: [require-all=(true|false)] [sequential=(true|false)]


colocation never_put_apache_with_dummy -inf: apache dummy
colocation c1 inf: A ( B C )


Commit the current configuration to the CIB in use. As noted elsewhere, commands in a configure session don’t have immediate effect on the CIB. All changes are applied at one point in time, either using commit or when the user leaves the configure level. In case the CIB in use changed in the meantime, presumably by somebody else, the crm shell will refuse to apply the changes.

If you know that it’s fine to still apply them, add force to the command line.

To disable CIB patching and apply the changes by replacing the CIB completely, add replace to the command line. Note that this can lead to previous changes being overwritten if some other process concurrently modifies the CIB.


commit [force] [replace]


This command takes the timeouts from the actions section of the resource agent meta-data and sets them for the operations of the primitive.


default-timeouts <id> [<id>...]
Note on default-timeouts

The use of this command is discouraged in favor of manually determining the best timeouts required for the particular configuration. Relying on the resource agent to supply appropriate timeouts can cause the resource to fail at the worst possible moment.

Appropriate timeouts for resource actions are context-sensitive, and should be carefully considered with the whole configuration in mind.


Delete one or more objects. If an object to be deleted belongs to a container object, such as a group, and it is the only resource in that container, then the container is deleted as well. Any related constraints are removed as well.

If the object is a started resource, it will not be deleted unless the --force flag is passed to the command, or the force option is set.


delete [--force] <id> [<id>...]


This command invokes the editor with the object description. As with the show command, the user may choose to edit all objects or a set of objects.

If the user insists, he or she may edit the XML edition of the object. If you do that, don’t modify any id attributes.


edit [xml] [<id> ...]
edit [xml] changed
Note on renaming element ids

The edit command sometimes cannot properly handle modifying element ids. In particular for elements which belong to group or ms resources. Group and ms resources themselves also cannot be renamed. Please use the rename command instead.


The erase clears all configuration. Apart from nodes. To remove nodes, you have to specify an additional keyword nodes.

Note that removing nodes from the live cluster may have some strange/interesting/unwelcome effects.


erase [nodes]


If multiple fencing (stonith) devices are available capable of fencing a node, their order may be specified by fencing_topology. The order is specified per node.

Stonith resources can be separated by , in which case all of them need to succeed. If they fail, the next stonith resource (or set of resources) is used. In other words, use comma to separate resources which all need to succeed and whitespace for serial order. It is not allowed to use whitespace around comma.

If the node is left out, the order is used for all nodes. That should reduce the configuration size in some stonith setups.

From Pacemaker version 1.1.14, it is possible to use a node attribute as the target in a fencing topology. The syntax for this usage is described below.

From Pacemaker version 1.1.14, it is also possible to use regular expression patterns as the target in a fencing topology. The configured fencing sequence then applies to all devices matching the pattern.


fencing_topology <stonith_resources> [<stonith_resources> ...]
fencing_topology <fencing_order> [<fencing_order> ...]
fencing_order :: <target> <stonith_resources> [<stonith_resources> ...]
stonith_resources :: <rsc>[,<rsc>...]
target :: <node>: | attr:<node-attribute>=<value> | pattern:<pattern>


# Only kill the power if poison-pill fails
fencing_topology poison-pill power
# As above for node-a, but a different strategy for node-b
fencing_topology \

node-a: poison-pill power \
node-b: ipmi serial # Fencing anything on rack 1 requires fencing via both APC 1 and 2, # to defeat the redundancy provided by two separate UPS units. fencing_topology attr:rack=1 apc01,apc02 # Fencing for all machines named green.* is done using the pear # fencing device first, while all machines named red.* are fenced # using the apple fencing device first. fencing_topology \
pattern:green.* pear apple \
pattern:red.* apple pear


This command filters the given CIB elements through an external program. The program should accept input on stdin and send output to stdout (the standard UNIX filter conventions). As with the show command, the user may choose to filter all or just a subset of elements.

It is possible to filter the XML representation of objects, but probably not as useful as the configuration language. The presentation is somewhat different from what would be displayed by the show command---each element is shown on a single line, i.e. there are no backslashes and no other embelishments.

Don’t forget to put quotes around the filter if it contains spaces.


filter <prog> [xml] [<id> ...]
filter <prog> [xml] changed


filter "sed '/^primitive/s/target-role=[^ ]*//'"
# crm configure filter "sed '/^primitive/s/target-role=[^ ]*//'"
crm configure <<END

filter "sed '/threshold=\"1\"/s/=\"1\"/=\"0\"/g'" END
Note on quotation marks

Filter commands which feature a blend of quotation marks can be difficult to get right, especially when used directly from bash, since bash does its own quotation parsing. In these cases, it can be easier to supply the filter command as standard input. See the last example above.


Show the value of the given property. If the value is not set, the command will print the default value for the property, if known.

If no property name is passed to the command, the list of known cluster properties is printed.

If the property is set multiple times, for example using multiple property sets with different rule expressions, the output of this command is undefined.

Pass the argument -t or --true to get-property to translate the argument value into true or false. If the value is not set, the command will print false.


get-property [-t|--true] [<name>]


get-property stonith-enabled
get-property -t maintenance-mode


Create a graphviz graphical layout from the current cluster configuration.

Currently, only dot (directed graph) is supported. It is essentially a visualization of resource ordering.

The graph may be saved to a file which can be used as source for various graphviz tools (by default it is displayed in the user’s X11 session). Optionally, by specifying the format, one can also produce an image instead.

For more or different graphviz attributes, it is possible to save the default set of attributes to an ini file. If this file exists it will always override the builtin settings. The exportsettings subcommand also prints the location of the ini file.


graph [<gtype> [<file> [<img_format>]]]
graph exportsettings
gtype :: dot
img_format :: `dot` output format (see the +-T+ option)


graph dot
graph dot
graph dot clu1.conf.svg svg


The group command creates a group of resources. This can be useful when resources depend on other resources and require that those resources start in order on the same node. A common use of resource groups is to ensure that a server and a virtual IP are located together, and that the virtual IP is started before the server.

Grouped resources are started in the order they appear in the group, and stopped in the reverse order. If a resource in the group cannot run anywhere, resources following it in the group will not start.

group can be passed the "container" meta attribute, to indicate that it is to be used to group VM resources monitored using Nagios. The resource referred to by the container attribute must be of type ocf:heartbeat:Xen, ocf:heartbeat:VirtualDomain or ocf:heartbeat:lxc.


group <name> <rsc> [<rsc>...]

[meta attr_list]
[params attr_list] attr_list :: [$id=<id>] <attr>=<val> [<attr>=<val>...] | $id-ref=<id>


group internal_www disk0 fs0 internal_ip apache \

meta target_role=stopped group vm-and-services vm vm-sshd meta container="vm"


Load a part of configuration (or all of it) from a local file or a network URL. The replace method replaces the current configuration with the one from the source. The update method tries to import the contents into the current configuration. The push method imports the contents into the current configuration and removes any lines that are not present in the given configuration. The file may be a CLI file or an XML file.

If the URL is -, the configuration is read from standard input.


load [xml] <method> URL
method :: replace | update | push


load xml update myfirstcib.xml
load xml replace
load xml push smallcib.xml


location defines the preference of nodes for the given resource. The location constraints consist of one or more rules which specify a score to be awarded if the rule matches.

The resource referenced by the location constraint can be one of the following:

•Plain resource reference: location loc1 webserver 100: node1

•Resource set in curly brackets: location loc1 { virtual-ip webserver } 100: node1

•Tag containing resource ids: location loc1 tag1 100: node1

•Resource pattern: location loc1 /web.*/ 100: node1

The resource-discovery attribute allows probes to be selectively enabled or disabled per resource and node.

The syntax for resource sets is described in detail for colocation.

For more details on how to configure resource sets, see Syntax: Resource sets.

For more information on rule expressions, see Syntax: Rule expressions.


location <id> <rsc> [<attributes>] {<node_pref>|<rules>}
rsc :: /<rsc-pattern>/

| { resource_sets }
| <rsc> attributes :: role=<role> | resource-discovery=always|never|exclusive node_pref :: <score>: <node> rules ::
rule [id_spec] [$role=<role>] <score>: <expression>
[rule [id_spec] [$role=<role>] <score>: <expression> ...] id_spec :: $id=<id> | $id-ref=<id> score :: <number> | <attribute> | [-]inf expression :: <simple_exp> [<bool_op> <simple_exp> ...] bool_op :: or | and simple_exp :: <attribute> [type:]<binary_op> <value>
| <unary_op> <attribute>
| date <date_expr> type :: string | version | number binary_op :: lt | gt | lte | gte | eq | ne unary_op :: defined | not_defined date_expr :: lt <end>
| gt <start>
| in start=<start> end=<end>
| in start=<start> <duration>
| spec <date_spec> duration|date_spec ::
| monthdays=<value>
| weekdays=<value>
| yearsdays=<value>
| months=<value>
| weeks=<value>
| years=<value>
| weekyears=<value>
| moon=<value>


location conn_1 internal_www 100: node1
location conn_1 internal_www \

rule 50: #uname eq node1 \
rule pingd: defined pingd location conn_2 dummy_float \
rule -inf: not_defined pingd or pingd number:lte 0 # never probe for rsc1 on node1 location no-probe rsc1 resource-discovery=never -inf: node1


Add or remove primitives in a group. The add subcommand appends the new group member by default. Should it go elsewhere, there are after and before clauses.


modgroup <id> add <id> [after <id>|before <id>]
modgroup <id> remove <id>


modgroup share1 add storage2 before share1-fs


Monitor is by far the most common operation. It is possible to add it without editing the whole resource. Also, long primitive definitions may be a bit uncluttered. In order to make this command as concise as possible, less common operation attributes are not available. If you need them, then use the op part of the primitive command.


monitor <rsc>[:<role>] <interval>[:<timeout>]


monitor apcfence 60m:60s

Note that after executing the command, the monitor operation may be shown as part of the primitive definition.

ms (master)

The ms command creates a master/slave resource type. It may contain a single primitive resource or one group of resources.


ms <name> <rsc>

[meta attr_list]
[params attr_list] attr_list :: [$id=<id>] <attr>=<val> [<attr>=<val>...] | $id-ref=<id>


ms disk1 drbd1 \

meta notify=true globally-unique=false
Note on ms deprecated

From Pacemaker-2.0, the resource type referred to as "master/slave", "stateful", or "multi-state" is no longer a separate resource type, but a variation of clone now referred to as a "promotable clone". For backward compatibility, above configurations are also accepted.

clone disk1 drbd1 \

meta promotable=true notify=true globally-unique=false
Note on id-ref usage

Instance or meta attributes (‘params` and meta) may contain a reference to another set of attributes. In that case, no other attributes are allowed. Since attribute sets’ ids, though they do exist, are not shown in the crm, it is also possible to reference an object instead of an attribute set. crm will automatically replace such a reference with the right id:

crm(live)configure# primitive a2 www-2 meta $id-ref=a1
crm(live)configure# show a2
primitive a2 apache \

meta $id-ref=a1-meta_attributes

It is advisable to give meaningful names to attribute sets which are going to be referenced.


The node command describes a cluster node. Nodes in the CIB are commonly created automatically by the CRM. Hence, you should not need to deal with nodes unless you also want to define node attributes. Note that it is also possible to manage node attributes at the node level.


node [$id=<id>] <uname>[:<type>]

[attributes [$id=<id>] [<score>:] [rule...]
<param>=<value> [<param>=<value>...]] | $id-ref=<ref>
[utilization [$id=<id>] [<score>:] [rule...]
<param>=<value> [<param>=<value>...]] | $id-ref=<ref> type :: normal | member | ping | remote


node node1
node big_node attributes memory=64


Set defaults for the operations meta attributes.

For more information on rule expressions, see Syntax: Rule expressions.


op_defaults [$id=<set_id>] [rule ...] <option>=<value> [<option>=<value> ...]


op_defaults record-pending=true


This constraint expresses the order of actions on two resources or more resources. If there are more than two resources, then the constraint is called a resource set.

Ordered resource sets have an extra attribute to allow for sets of resources whose actions may run in parallel. The shell syntax for such sets is to put resources in parentheses.

If the subsequent resource can start or promote after any one of the resources in a set has done, enclose the set in brackets ([ and ]).

Sets cannot be nested.

Three strings are reserved to specify a kind of order constraint: Mandatory, Optional, and Serialize. It is preferred to use one of these settings instead of score. Previous versions mapped scores 0 and inf to keywords advisory and mandatory. That is still valid but deprecated.

For more details on how to configure resource sets, see Syntax: Resource sets.


order <id> [kind:] first then [symmetrical=<bool>]
order <id> [kind:] resource_sets [symmetrical=<bool>]
kind :: Mandatory | Optional | Serialize
first :: <rsc>[:<action>]
then ::  <rsc>[:<action>]
resource_sets :: resource_set [resource_set ...]
resource_set :: ["["|"("] <rsc>[:<action>] [<rsc>[:<action>] ...] \

[attributes] ["]"|")"] attributes :: [require-all=(true|false)] [sequential=(true|false)]


order o-1 Mandatory: apache:start ip_1
order o-2 Serialize: A ( B C )
order o-4 first-resource then-resource


The primitive command describes a resource. It may be referenced only once in group, or clone objects. If it’s not referenced, then it is placed as a single resource in the CIB.

Operations may be specified anonymously, as a group or by reference:

•"Anonymous", as a list of op specifications. Use this method if you don’t need to reference the set of operations elsewhere. This is the most common way to define operations.

•If reusing operation sets is desired, use the operations keyword along with an id to give the operations set a name. Use the operations keyword and an id-ref value set to the id of another operations set, to apply the same set of operations to this primitive.

Operation attributes which are not recognized are saved as instance attributes of that operation. A typical example is OCF_CHECK_LEVEL.

For multistate resources, roles are specified as role=<role>. The Master/Slave resources are deprecated and replaced by Promoted/Unpromoted promotable resources if desired.

A template may be defined for resources which are of the same type and which share most of the configuration. See rsc_template for more information.

Attributes containing time values, such as the interval attribute on operations, are configured either as a plain number, which is interpreted as a time in seconds, or using one of the following suffixes:

•s, sec - time in seconds (same as no suffix)

•ms, msec - time in milliseconds

•us, usec - time in microseconds

•m, min - time in minutes

•h, hr - time in hours


primitive <rsc> {[<class>:[<provider>:]]<type>|@<template>}

[[params] attr_list]
[meta attr_list]
[utilization attr_list]
[operations id_spec]
[op op_type [<attribute>=<value>...]
[[op_params] attr_list]
[op_meta attr_list] ...] attr_list :: [$id=<id>] [<score>:] [rule...]
<attr>=<val> [<attr>=<val>...]] | $id-ref=<id> id_spec :: $id=<id> | $id-ref=<id> op_type :: start | stop | monitor


primitive apcfence stonith:apcsmart \

params ttydev=/dev/ttyS0 hostlist="node1 node2" \
op start timeout=60s \
op monitor interval=30m timeout=60s primitive www8 apache \
configfile=/etc/apache/www8.conf \
operations $id-ref=apache_ops primitive db0 mysql \
params config=/etc/mysql/db0.conf \
op monitor interval=60s \
op monitor interval=300s OCF_CHECK_LEVEL=10 primitive r0 ocf:linbit:drbd \
params drbd_resource=r0 \
op monitor role=Promoted interval=60s \
op monitor role=Unpromoted interval=300s primitive xen0 @vm_scheme1 xmfile=/etc/xen/vm/xen0 primitive mySpecialRsc Special \
params 3: rule #uname eq node1 interface=eth1 \
params 2: rule #uname eq node2 interface=eth2 port=8888 \
params 1: interface=eth0 port=9999 primitive A ocf:pacemaker:Dummy \
op start \
op_meta 2: rule #ra-version version:gt 1.0 timeout=120s \
op_meta 1: timeout=60s


Set cluster configuration properties. To list the available cluster configuration properties, use the ra info command with pengine, crmd, cib and stonithd as arguments. When setting the maintenance-mode property, it will inform the user if there are nodes or resources that have the maintenance property.

For more information on rule expressions, see Syntax: Rule expressions.


property [<set_id>:] [rule ...] <option>=<value> [<option>=<value> ...]


property stonith-enabled=true
property rule date spec years=2014 stonith-enabled=false

ptest (simulate)

Show PE (Policy Engine) motions using ptest(8) or crm_simulate(8).

A CIB is constructed using the current user edited configuration and the status from the running CIB. The resulting CIB is run through ptest (or crm_simulate) to show changes which would happen if the configuration is committed.

The status section may be loaded from another source and modified using the cibstatus level commands. In that case, the ptest command will issue a message informing the user that the Policy Engine graph is not calculated based on the current status section and therefore won’t show what would happen to the running but some imaginary cluster.

If you have graphviz installed and X11 session, dotty(1) is run to display the changes graphically.

Add a string of v characters to increase verbosity. ptest can also show allocation scores. utilization turns on information about the remaining capacity of nodes. With the actions option, ptest will print all resource actions.

The ptest program has been replaced by crm_simulate in newer Pacemaker versions. In some installations both could be installed. Use simulate to enfore using crm_simulate.


ptest [nograph] [v...] [scores] [actions] [utilization]


ptest scores
ptest vvvvv
simulate actions


Refresh the internal structures from the CIB. All changes made during this session are lost.




Rename an object. It is recommended to use this command to rename a resource, because it will take care of updating all related constraints and a parent resource. Changing ids with the edit command won’t have the same effect.

If you want to rename a resource, it must be in the stopped state.


rename <old_id> <new_id>


An ACL role is a set of rules which describe access rights to CIB. Rules consist of an access right read, write, or deny and a specification denoting part of the configuration to which the access right applies. The specification can be an XPath or a combination of tag and id references. If an attribute is appended, then the specification applies only to that attribute of the matching element.

There is a number of shortcuts for XPath specifications. The meta, params, and utilization shortcuts reference resource meta attributes, parameters, and utilization respectively. The location may be used to specify location constraints most of the time to allow resource move and unmove commands. The property references cluster properties. The node allows reading node attributes. nodeattr and nodeutil reference node attributes and node capacity (utilization). The status shortcut references the whole status section of the CIB. Read access to status is necessary for various monitoring tools such as crm_mon(8) (aka crm status).

For more information on rule expressions, see Syntax: Rule expressions.


role <role-id> rule [rule ...]
rule :: acl-right cib-spec [attribute:<attribute>]
acl-right :: read | write | deny
cib-spec :: xpath-spec | tag-ref-spec
xpath-spec :: xpath:<xpath> | shortcut
tag-ref-spec :: tag:<tag> | ref:<id> | tag:<tag> ref:<id>
shortcut :: meta:<rsc>[:<attr>]



role app1_admin \

write meta:app1:target-role \
write meta:app1:is-managed \
write location:app1 \
read ref:app1


Set defaults for the resource meta attributes.

For more information on rule expressions, see Syntax: Rule expressions.


rsc_defaults [<set_id>:] [rule ...] <option>=<value> [<option>=<value> ...]


rsc_defaults failure-timeout=3m


The rsc_template command creates a resource template. It may be referenced in primitives. It is used to reduce large configurations with many similar resources.


rsc_template <name> [<class>:[<provider>:]]<type>

[params attr_list]
[meta attr_list]
[utilization attr_list]
[operations id_spec]
[op op_type [<attribute>=<value>...] ...] attr_list :: [$id=<id>] <attr>=<val> [<attr>=<val>...] | $id-ref=<id> id_spec :: $id=<id> | $id-ref=<id> op_type :: start | stop | monitor


rsc_template public_vm Xen \

op start timeout=300s \
op stop timeout=300s \
op monitor interval=30s timeout=60s \
op migrate_from timeout=600s \
op migrate_to timeout=600s primitive xen0 @public_vm \
params xmfile=/etc/xen/xen0 primitive xen1 @public_vm \
params xmfile=/etc/xen/xen1


This constraint expresses dependency of resources on cluster-wide attributes, also known as tickets. Tickets are mainly used in geo-clusters, which consist of multiple sites. A ticket may be granted to a site, thus allowing resources to run there.

The loss-policy attribute specifies what happens to the resource (or resources) if the ticket is revoked. The default is either stop or demote depending on whether a resource is multi-state.

See also the site set of commands.


rsc_ticket <id> <ticket_id>: <rsc>[:<role>] [<rsc>[:<role>] ...]

[loss-policy=<loss_policy_action>] loss_policy_action :: stop | demote | fence | freeze


rsc_ticket ticket-A_public-ip ticket-A: public-ip
rsc_ticket ticket-A_bigdb ticket-A: bigdb loss-policy=fence
rsc_ticket ticket-B_storage ticket-B: drbd-a:Promoted drbd-b:Promoted


Test resources with current resource configuration. If no nodes are specified, tests are run on all known nodes.

The order of resources is significant: it is assumed that later resources depend on earlier ones.

If a resource is multi-state, it is assumed that the role on which later resources depend is master.

Tests are run sequentially to prevent running the same resource on two or more nodes. Tests are carried out only if none of the specified nodes currently run any of the specified resources. However, it won’t verify whether resources run on the other nodes.

Superuser privileges are obviously required: either run this as root or setup the sudoers file appropriately.

Note that resource testing may take some time.


rsctest <rsc_id> [<rsc_id> ...] [<node_id> ...]


rsctest my_ip websvc
rsctest websvc nodeB


Save the current configuration to a file. Optionally, as XML. Use - instead of file name to write the output to stdout.

The save command accepts the same selection arguments as the show command. See the help section for show for more details.


save [xml] [<id> | type:<type | tag:<tag> |

related:<obj> | changed ...] <file>


save myfirstcib.txt
save web-server server-config.txt


CIB’s content is validated by a RNG schema. Pacemaker supports several, depending on version. At least the following schemas are accepted by crmsh:






Use this command to display or switch to another RNG schema.


schema [<schema>]


schema pacemaker-1.1


Set the value of a configured attribute. The attribute must configured previously, and can be an agent parameter, meta attribute, utilization value or operation value.

The first argument to the command is a path to an attribute. This is a dot-separated sequence beginning with the name of the resource or object, and ending with the name of the attribute to set. To set operation value, op_type should be specified; when multi operations exist like multi monitors, interval should be specified.


set <path> <value>
path:: id.[op_type.][interval.]name


set vip1.ip
set vm-a.force_stop 1
set vip1.monitor.on-fail ignore
set drbd.monitor.10s.interval 20s


The show command displays CIB objects. Without any argument, it displays all objects in the CIB, but the set of objects displayed by show can be limited to only objects with the given IDs or by using one or more of the special prefixes described below.

The XML representation for the objects can be displayed by passing xml as the first argument.

To show one or more specific objects, pass the object IDs as arguments.

To show all objects of a certain type, use the type: prefix.

To show all objects in a tag, use the tag: prefix.

To show all constraints related to a primitive, or to show all objects of a certain RA type, use the related: prefix.

To show all modified objects, pass the argument changed.

The prefixes can be used together on a single command line. For example, to show both the tag itself and the objects tagged by it the following combination can be used: show tag:my-tag my-tag.

To refine a selection of objects using multiple modifiers, the keywords and and or can be used. For example, to select all primitives tagged foo, the following combination can be used: show type:primitive and tag:foo.

To hide values when displaying the configuration, use the obscure:<glob> argument. This can be useful when sending the configuration over a public channel, to avoid exposing potentially sensitive information. The <glob> argument is a bash-style pattern matching attribute keys.

In /etc/crm/crm.conf, obscure_pattern option is the persisent configuration of CLI. Example, for the high security concern,

obscure_pattern = passw* | ip

Which makes crm configure show is equal to

node-1:~ # crm configure show obscure:passw* obscure:ip
node 1084783297: node1
primitive fence_device stonith:fence_ilo5 \

params password="******" primitive ip IPaddr2 \
params ip="******"

The default suggestion is passw* If you don’t want to obscure, change the value to blank.


show [xml] [<id>

| changed
| type:<type>
| tag:<id>
| related:<obj>
| obscure:<glob>
...] type :: node | primitive | group | clone | ms | rsc_template
| location | colocation | order
| rsc_ticket
| property | rsc_defaults | op_defaults
| fencing_topology
| role | user | acl_target
| tag


show webapp
show type:primitive
show xml tag:db tag:fs
show related:webapp
show related:IPaddr2
show related:ocf:heartbeat:Dummy
show type:primitive obscure:passwd


Define a resource tag. A tag is an id referring to one or more resources, without implying any constraints between the tagged resources. This can be useful for grouping conceptually related resources.


tag <tag-name>: <rsc> [<rsc> ...]
tag <tag-name> <rsc> [<rsc> ...]


tag web: p-webserver p-vip
tag ips server-vip admin-vip


The specified template is loaded into the editor. It’s up to the user to make a good CRM configuration out of it. See also the template section.


template [xml] url


template two-apaches.txt


Attempts to upgrade the CIB to validate with the current version. Commonly, this is required if the error CIB not supported occurs. It typically means that the active CIB version is coming from an older release.

As a safety precaution, the force argument is required if the validation-with attribute is set to anything other than 0.6. Thus in most cases, it is required.


upgrade [force]


upgrade force


Users which normally cannot view or manage cluster configuration can be allowed access to parts of the CIB. The access is defined by a set of read, write, and deny rules as in role definitions or by referencing roles. The latter is considered best practice.

For more information on rule expressions, see Syntax: Rule expressions.


user <uid> {roles|rules}
roles :: role:<role-ref> [role:<role-ref> ...]
rules :: rule [rule ...]


user joe \

role:app1_admin \


Call the validate-all action for the resource, if possible.


•The resource agent must implement the validate-all action.

•The current user must be root.

•The primitive resource must not use nvpair references.


validate-all <rsc>


Verify the contents of the CIB which would be committed.




Even though we promissed no xml, it may happen, but hopefully very very seldom, that an element from the CIB cannot be rendered in the configuration language. In that case, the element will be shown as raw xml, prefixed by this command. That element can then be edited like any other. If the shell finds out that after the change it can digest it, then it is going to be converted into the normal configuration language. Otherwise, there is no need to use xml for configuration.


xml <xml>

template - Import configuration from templates

User may be assisted in the cluster configuration by templates prepared in advance. Templates consist of a typical ready configuration which may be edited to suit particular user needs.

This command enters a template level where additional commands for configuration/template management are available.


Copy the current or given configuration to the current CIB. By default, the CIB is replaced, unless the method is set to "update".


apply [<method>] [<config>]
method :: replace | update


Remove a configuration. The loaded (active) configuration may be removed by force.


delete <config> [force]


Edit current or given configuration using your favourite editor.


edit [<config>]


When called with no argument, lists existing templates and configurations.

Given the argument templates, lists the available templates.

Given the argument configs, lists the available configurations.


list [templates|configs]


Load an existing configuration. Further edit, show, and apply commands will refer to this configuration.


load <config>


Create a new configuration from one or more templates. Note that configurations and templates are kept in different places, so it is possible to have a configuration name equal a template name.

If you already know which parameters are required, you can set them directly on the command line.

The parameter name id is set by default to the name of the configuration.

If no parameters are being set and you don’t want a particular name for your configuration, you can call this command with a template name as the only parameter. A unique configuration name based on the template name will be generated.


new [<config>] <template> [<template> ...] [params name=value ...]


new vip virtual-ip
new bigfs ocfs2 params device=/dev/sdx8 directory=/bigfs
new apache


Process the current or given configuration and display the result.


show [<config>]

cibstatus - CIB status management and editing

The status section of the CIB keeps the current status of nodes and resources. It is modified only on events, i.e. when some resource operation is run or node status changes. For obvious reasons, the CRM has no user interface with which it is possible to affect the status section. From the user’s point of view, the status section is essentially a read-only part of the CIB. The current status is never even written to disk, though it is available in the PE (Policy Engine) input files which represent the history of cluster motions. The current status may be read using the cibadmin -Q command.

It may sometimes be of interest to see how status changes would affect the Policy Engine. The set of ‘cibstatus` level commands allow the user to load status sections from various sources and then insert or modify resource operations or change nodes’ state.

The effect of those changes may then be observed by running the ptest command at the configure level or simulate and run commands at this level. The ptest runs with the user edited CIB whereas the latter two commands run with the CIB which was loaded along with the status section.

The simulate and run commands as well as all status modification commands are implemented using crm_simulate(8).


Load a status section from a file, a shadow CIB, or the running cluster. By default, the current (live) status section is modified. Note that if the live status section is modified it is not going to be updated if the cluster status changes, because that would overwrite the user changes. To make crm drop changes and resume use of the running cluster status, run load live.

All CIB shadow configurations contain the status section which is a snapshot of the status section taken at the time the shadow was created. Obviously, this status section doesn’t have much to do with the running cluster status, unless the shadow CIB has just been created. Therefore, the ptest command by default uses the running cluster status section.


load {<file>|shadow:<cib>|live}


load bug-12299.xml
load shadow:test1


Change the node status. It is possible to throw a node out of the cluster, make it a member, or set its state to unclean.


Set the node_state crmd attribute to online and the expected and join attributes to member. The effect is that the node becomes a cluster member.


Set the node_state crmd attribute to offline and the expected attribute to empty. This makes the node cleanly removed from the cluster.


Set the node_state crmd attribute to offline and the expected attribute to member. In this case the node has unexpectedly disappeared.


node <node> {online|offline|unclean}


node xen-b unclean


Edit the outcome of a resource operation. This way you can tell CRM that it ran an operation and that the resource agent returned certain exit code. It is also possible to change the operation’s status. In case the operation status is set to something other than done, the exit code is effectively ignored.


op <operation> <resource> <exit_code> [<op_status>] [<node>]
operation :: probe | monitor[:<n>] | start | stop |

promote | demote | notify | migrate_to | migrate_from exit_code :: <rc> | success | generic | args |
unimplemented | perm | installed | configured | not_running |
master | failed_master op_status :: pending | done | cancelled | timeout | notsupported | error n :: the monitor interval in seconds; if omitted, the first
recurring operation is referenced rc :: numeric exit code in range 0..9


op start d1 xen-b generic
op start d1 xen-b 1
op monitor d1 xen-b not_running
op stop d1 xen-b 0 timeout


Show the origin of the status section currently in use. This essentially shows the latest load argument.




Set the quorum value.


quorum <bool>


quorum false


Run the policy engine with the edited status section.

Add a string of v characters to increase verbosity. Specify scores to see allocation scores also. utilization turns on information about the remaining capacity of nodes.

If you have graphviz installed and X11 session, dotty(1) is run to display the changes graphically.


run [nograph] [v...] [scores] [utilization]




The current internal status section with whatever modifications were performed can be saved to a file or shadow CIB.

If the file exists and contains a complete CIB, only the status section is going to be replaced and the rest of the CIB will remain intact. Otherwise, the current user edited configuration is saved along with the status section.

Note that all modifications are saved in the source file as soon as they are run.


save [<file>|shadow:<cib>]


save bug-12299.xml


Show the current status section in the XML format. Brace yourself for some unreadable output. Add changed option to get a human readable output of all changes.


show [changed]


Run the policy engine with the edited status section and simulate the transition.

Add a string of v characters to increase verbosity. Specify scores to see allocation scores also. utilization turns on information about the remaining capacity of nodes.

If you have graphviz installed and X11 session, dotty(1) is run to display the changes graphically.


simulate [nograph] [v...] [scores] [utilization]




Modify the ticket status. Tickets can be granted and revoked. Granted tickets could be activated or put in standby.


ticket <ticket> {grant|revoke|activate|standby}


ticket ticketA grant

assist - Configuration assistant

The assist sublevel is a collection of helper commands that create or modify resources and constraints, to simplify the creation of certain configurations.

For more information on individual commands, see the help text for those commands.


This command takes a list of primitives as argument, and creates a new rsc_template for these primitives. It can only do this if the primitives do not already share a template and are of the same type.


template primitive-1 primitive-2 primitive-3


A colocation between a group of resources says that the resources should be located together, but it also means that those resources are dependent on each other. If one of the resources fails, the others will be restarted.

If this is not desired, it is possible to circumvent: By placing the resources in a non-sequential set and colocating the set with a dummy resource which is not monitored, the resources will be placed together but will have no further dependency on each other.

This command creates both the constraint and the dummy resource needed for such a colocation.


weak-bond resource-1 resource-2

maintenance - Maintenance mode commands

Maintenance mode commands are commands that manipulate resources directly without going through the cluster infrastructure. Therefore, it is essential to ensure that the cluster does not attempt to monitor or manipulate the resources while these commands are being executed.

To ensure this, these commands require that maintenance mode is set either for the particular resource, or for the whole cluster.


Invokes the given action for the resource. This is done directly via the resource agent, so the command must be issued while the cluster or the resource is in maintenance mode.

Unless the action is start or monitor, the action must be invoked on the same node as where the resource is running. If the resource is running on multiple nodes, the command will fail.

To use SSH for executing resource actions on multiple nodes, append ssh after the action name. This requires SSH access to be configured between the nodes and the parallax python package to be installed.


action <rsc> <action>
action <rsc> <action> ssh


action webserver reload
action webserver monitor ssh


Disables maintenances mode, either for the whole cluster or for the given resource.


off <rsc>


off rsc1


Enables maintenances mode, either for the whole cluster or for the given resource.


on <rsc>


on rsc1

history - Cluster history

Examining Pacemaker’s history is a particularly involved task. The number of subsystems to be considered, the complexity of the configuration, and the set of various information sources, most of which are not exactly human readable, keep analyzing resource or node problems accessible to only the most knowledgeable. Or, depending on the point of view, to the most persistent. The following set of commands has been devised in hope to make cluster history more accessible.

Of course, looking at all history could be time consuming regardless of how good the tools at hand are. Therefore, one should first say which period he or she wants to analyze. If not otherwise specified, the last hour is considered. Logs and other relevant information is collected using crm report. Since this process takes some time and we always need fresh logs, information is refreshed in a much faster way using the python parallax module. If python-parallax is not found on the system, examining a live cluster is still possible — though not as comfortable.

Apart from examining a live cluster, events may be retrieved from a report generated by crm report (see also the -H option). In that case we assume that the period stretching the whole report needs to be investigated. Of course, it is still possible to further reduce the time range.

If you have discovered an issue that you want to show someone else, you can use the session pack command to save the current session as a tarball, similar to those generated by crm report.

In order to minimize the size of the tarball, and to make it easier for others to find the interesting events, it is recommended to limit the time frame which the saved session covers. This can be done using the timeframe command (example below).

It is also possible to name the saved session using the session save command.


crm(live)history# limit "Jul 18 12:00" "Jul 18 12:30"
crm(live)history# session save strange_restart
crm(live)history# session pack
Report saved in .../strange_restart.tar.bz2


How much detail to show from the logs. Valid detail levels are either 0 or 1, where 1 is the highest detail level. The default detail level is 0.


detail <detail_level>
detail_level :: small integer (defaults to 0)


detail 1


A transition represents a change in cluster configuration or state. Use diff to see what has changed between two transitions.

If you want to specify the current cluster configuration and status, use the string live.

Normally, the first transition specified should be the one which is older, but we are not going to enforce that.

Note that a single configuration update may result in more than one transition.


diff <pe> <pe> [status] [html]
pe :: <number>|<index>|<file>|live


diff 2066 2067
diff pe-input-2080.bz2 live status


By analysing the log output and looking for particular patterns, the events command helps sifting through the logs to find when particular events like resources changing state or node failure may have occurred.

This can be used to generate a combined list of events from all nodes.






If a log is infested with irrelevant messages, those messages may be excluded by specifying a regular expression. The regular expressions used are Python extended. This command is additive. To drop all regular expressions, use exclude clear. Run exclude only to see the current list of regular expressions. Excludes are saved along with the history sessions.


exclude [<regex>|clear]


exclude kernel.*ocfs2


Create a graphviz graphical layout from the PE file (the transition). Every transition contains the cluster configuration which was active at the time. See also generate a directed graph from configuration.


graph <pe> [<gtype> [<file> [<img_format>]]]
gtype :: dot
img_format :: `dot` output format (see the +-T+ option)


graph -1
graph 322 dot
graph 322 dot clu1.conf.svg svg


The info command provides a summary of the information source, which can be either a live cluster snapshot or a previously generated report.






The latest command shows a bit of recent history, more precisely whatever happened since the last cluster change (the latest transition). If the transition is running, the shell will first wait until it finishes.





limit (timeframe)

This command can be used to modify the time span to examine. All history commands look at events within a certain time span.

For the live source, the default time span is the last hour.

There is no time span limit for the crm report source.

The time period is parsed by the dateutil python module. It covers a wide range of date formats. For instance:

•3:00 (today at 3am)

•15:00 (today at 3pm)

•2010/9/1 2pm (September 1st 2010 at 2pm)

For more examples of valid time/date statements, please refer to the python-dateutil documentation:


If the dateutil module is not available, then the time is parsed using strptime and only the kind as printed by date(1) is allowed:

•Tue Sep 15 20:46:27 CEST 2010


limit [<from_time>] [<to_time>]


limit 10:15
limit 15h22m 16h
limit "Sun 5 20:46" "Sun 5 22:00"


Show messages logged on one or more nodes. Leaving out a node name produces combined logs of all nodes. Messages are sorted by time and, if the terminal emulations supports it, displayed in different colours depending on the node to allow for easier reading.

The sorting key is the timestamp as written by syslog which normally has the maximum resolution of one second. Obviously, messages generated by events which share the same timestamp may not be sorted in the same way as they happened. Such close events may actually happen fairly often.


log [<node> [<node> ...] ]


log node-a


Show important events that happened on a node. Important events are node lost and join, standby and online, and fence. Use either node names or extended regular expressions.


node <node> [<node> ...]


node node1


Every event in the cluster results in generating one or more Policy Engine (PE) files. These files describe future motions of resources. The files are listed as full paths in the current report directory. Add v to also see the creation time stamps.


peinputs [{<range>|<number>} ...] [v]
range :: <n1>:<n2>


peinputs 440:444 446
peinputs v


This command makes sense only for the live source and makes crm collect the latest logs and other relevant information from the logs. If you want to make a completely new report, specify force.


refresh [force]


Show actions and any failures that happened on all specified resources on all nodes. Normally, one gives resource names as arguments, but it is also possible to use extended regular expressions. Note that neither groups nor clones or master/slave names are ever logged. The resource command is going to expand all of these appropriately, so that clone instances or resources which are part of a group are shown.


resource <rsc> [<rsc> ...]


resource bigdb public_ip
resource my_.*_db2
resource ping_clone


Sometimes you may want to get back to examining a particular history period or bug report. In order to make that easier, the current settings can be saved and later retrieved.

If the current history being examined is coming from a live cluster the logs, PE inputs, and other files are saved too, because they may disappear from nodes. For the existing reports coming from crm report, only the directory location is saved (not to waste space).

A history session may also be packed into a tarball which can then be sent to support.

Leave out subcommand to see the current session.


session [{save|load|delete} <name> | pack [<name>] | update | list]


session save bnc966622
session load rsclost-2
session list


In case the host this program runs on is not part of the cluster, it is necessary to set the list of nodes.


setnodes node <node> [<node> ...]


setnodes node_a node_b


Every transition is saved as a PE file. Use this command to render that PE file either as configuration or status. The configuration output is the same as crm configure show.


show <pe> [status]
pe :: <number>|<index>|<file>|live


show 2066
show pe-input-2080.bz2 status


Events to be examined can come from the current cluster or from a crm report report. This command sets the source. source live sets source to the running cluster and system logs. If no source is specified, the current source information is printed.

In case a report source is specified as a file reference, the file is going to be unpacked in place where it resides. This directory is not removed on exit.


source [<dir>|<file>|live]


source live
source /tmp/customer_case_22.tar.bz2
source /tmp/customer_case_22


This command will print actions planned by the PE and run graphviz (dotty) to display a graphical representation of the transition. Of course, for the latter an X11 session is required. This command invokes ptest(8) in background.

The showdot subcommand runs graphviz (dotty) to display a graphical representation of the .dot file which has been included in the report. Essentially, it shows the calculation produced by pengine which is installed on the node where the report was produced. In optimal case this output should not differ from the one produced by the locally installed pengine.

The log subcommand shows the full log for the duration of the transition.

A transition can also be saved to a CIB shadow for further analysis or use with cib or configure commands (use the save subcommand). The shadow file name defaults to the name of the PE input file.

If the PE input file number is not provided, it defaults to the last one, i.e. the last transition. The last transition can also be referenced with number 0. If the number is negative, then the corresponding transition relative to the last one is chosen.

If there are warning and error PE input files or different nodes were the DC in the observed timeframe, it may happen that PE input file numbers collide. In that case provide some unique part of the path to the file.

After the ptest output, logs about events that happened during the transition are printed.

The tags subcommand scans the logs for the transition and return a list of key events during that transition. For example, the tag error will be returned if there are any errors logged during the transition.


transition [<number>|<index>|<file>] [nograph] [v...] [scores] [actions] [utilization]
transition showdot [<number>|<index>|<file>]
transition log [<number>|<index>|<file>]
transition save [<number>|<index>|<file> [name]]
transition tags [<number>|<index>|<file>]


transition 444
transition -1
transition pe-error-3.bz2
transition node-a/pengine/pe-input-2.bz2
transition showdot 444
transition log
transition save 0 enigma-22


A transition represents a change in cluster configuration or state. This command lists the transitions in the current timeframe.






A transition represents a change in cluster configuration or state. Use wdiff to see what has changed between two transitions as word differences on a line-by-line basis.

If you want to specify the current cluster configuration and status, use the string live.

Normally, the first transition specified should be the one which is older, but we are not going to enforce that.

Note that a single configuration update may result in more than one transition.


wdiff <pe> <pe> [status]
pe :: <number>|<index>|<file>|live


wdiff 2066 2067
wdiff pe-input-2080.bz2 live status


Interface to a tool for creating a cluster report. A report is an archive containing log files, configuration files, system information and other relevant data for a given time period. This is a useful tool for collecting data to attach to bug reports, or for detecting the root cause of errors resulting in resource failover, for example.

See crmsh_crm_report(8) for more details on arguments, or call crm report -h


report -f {time|"cts:"testnum} [-t time] [-u user] [-l file]

[-n nodes] [-E files] [-p patt] [-L patt] [-e prog]
[-MSDZAVsvhd] [dest]


report -f 2pm report_1
report -f "2007/9/5 12:30" -t "2007/9/5 14:00" report_2
report -f 1:00 -t 3:00 -l /var/log/cluster/ha-debug report_3
report -f "09sep07 2:00" -u hbadmin report_4
report -f 18:00 -p "usern.*" -p "admin.*" report_5
report -f cts:133 ctstest_133

end (cd, up)

The end command ends the current level and the user moves to the parent level. This command is available everywhere.




The help command prints help for the current level or for the specified topic (command). This command is available everywhere.


help [<topic>]

quit (exit, bye)

Leave the program.




YAML file /etc/crm/profiles.yml contains Corosync, SBD and Pacemaker parameters for different platforms.

crmsh bootstrap detects system environment and load the corresponding parameters predefined in this file.



key_name: value

The valid profile names are: "microsoft-azure", "google-cloud-platform", "amazon-web-services", "s390", "default"

key_name is a known Corosync, SBD, or Pacemaker parameters, like corosync.totem.token or sbd.watchdog_timeout.

More details about the parameter definitions please refer to the man page of corosync.conf(5), sbd(8).



corosync.totem.crypto_hash: sha1
corosync.totem.crypto_cipher: aes256
corosync.totem.token: 5000
corosync.totem.join: 60
corosync.totem.max_messages: 20
corosync.totem.token_retransmits_before_loss_const: 10
sbd.watchdog_timeout: 15 microsoft-azure:
corosync.totem.token: 30000
sbd.watchdog_timeout: 60

How the content of the file is interpreted

The profiles has the following properties:

•Profiles are only loaded on bootstrap init node.

•The "default" profile is loaded in the beginning.

•Specific profiles will override the corresponding values in the "default" profile (if the specific environment is detected).

•Users could customize the "default" profile for their needs. For example, those on-premise environments which is not defined yet.


Even though all sensible configurations (and most of those that are not) are going to be supported by the crm shell, I suspect that it may still happen that certain XML constructs may confuse the tool. When that happens, please file a bug report.

The crm shell will not try to update the objects it does not understand. Of course, it is always possible to edit such objects in the XML format.


Dejan Muhamedagic, <> Kristoffer Gronlund <> and many OTHERS


crm_resource(8), crm_attribute(8), crm_mon(8), cib_shadow(8), ptest(8), dotty(1), crm_simulate(8), cibadmin(8)


Copyright (C) 2008-2013 Dejan Muhamedagic. Copyright (C) 2013 Kristoffer Gronlund.

Free use of this software is granted under the terms of the GNU General Public License (GPL).

03/03/2023 crm 4.0.0