diff --git a/doc/sphinx/Pacemaker_Explained/acls.rst b/doc/sphinx/Pacemaker_Explained/acls.rst index 1786768a76..7385f97e70 100644 --- a/doc/sphinx/Pacemaker_Explained/acls.rst +++ b/doc/sphinx/Pacemaker_Explained/acls.rst @@ -1,335 +1,337 @@ +.. _acl: + Access Control Lists (ACLs) --------------------------- .. Convert_to_RST: anchor:ch-acls[Chapter 13, ACLs] indexterm:[access control list] indexterm:[ACL] By default, the +root+ user or any user in the +haclient+ group can modify Pacemaker's CIB without restriction. Pacemaker offers 'access control lists (ACLs)' to provide more fine-grained authorization. == ACL Prerequisites == In order to use ACLs: * The Pacemaker software must have been compiled with ACL support. If the output of the command `pacemakerd --features` contains `acls`, your installation supports ACLs. * Desired users must have user accounts in the +haclient+ group on all nodes in the cluster. * If your CIB was created before Pacemaker 1.1.12, it may need to be updated to the current schema using `cibadmin --upgrade` in order to use the syntax documented here. * The +enable-acl+ <> must be set to true. == ACL Configuration == ACLs are specified within an +acls+ element of the CIB. The +acls+ element may contain any number of +acl_role+, +acl_target+, and +acl_group+ elements. == ACL Roles == An ACL role is a collection of permissions allowing or denying access to particular portions of the CIB. .Properties of an ACL Role [width="95%",cols="1m,<3",options="header",align="center"] |==== |Attribute |Description |id |A unique name for the role (required) indexterm:[id,acl_role] indexterm:[access control list,acl_role,id] |description |Arbitrary text (not used by Pacemaker) indexterm:[description,acl_role] indexterm:[access control list,acl_role,description] |==== An +acl_role+ element may contain any number of +acl_permission+ elements. .Properties of an ACL Permission [width="95%",cols="1m,<3",options="header",align="center"] |==== |Attribute |Description |id |A unique name for the permission (required) indexterm:[id,acl_permission] indexterm:[access control list,acl_permission,id] |description |Arbitrary text (not used by Pacemaker) indexterm:[description,acl_permission] indexterm:[access control list,acl_permission,description] |kind |The access being granted. Allowed values are +read+, +write+, and +deny+. A value of +write+ grants both read and write access. indexterm:[kind,acl_permission] indexterm:[access control list,acl_permission,kind] |object-type |The name of an XML element in the CIB to which the permission applies. (Exactly one of +object-type+, +xpath+, and +reference+ must be specified for a permission.) indexterm:[object-type,acl_permission] indexterm:[access control list,acl_permission,object-type] |attribute |If specified, the permission applies only to +object-type+ elements that have this attribute set (to any value). If not specified, the permission applies to all +object-type+ elements. May only be used with +object-type+. indexterm:[attribute,acl_permission] indexterm:[access control list,acl_permission,attribute] |reference |The ID of an XML element in the CIB to which the permission applies. (Exactly one of +object-type+, +xpath+, and +reference+ must be specified for a permission.) indexterm:[reference,acl_permission] indexterm:[access control list,acl_permission,reference] |xpath |An https://www.w3.org/TR/xpath-10/[XPath] specification selecting an XML element in the CIB to which the permission applies. Attributes may be specified in the XPath to select particular elements, but the permissions apply to the entire element. (Exactly one of +object-type+, +xpath+, and +reference+ must be specified for a permission.) indexterm:[xpath,acl_permission] indexterm:[access control list,acl_permission,xpath] |==== [IMPORTANT] ==== * Permissions are applied to the selected XML element's entire XML subtree (all elements enclosed within it). * Write permission grants the ability to create, modify, or remove the element and its subtree, and also the ability to create any "scaffolding" elements (enclosing elements that do not have attributes other than an ID). * Permissions for more specific matches (more deeply nested elements) take precedence over more general ones. * If multiple permissions are configured for the same match (for example, in different roles applied to the same user), any +deny+ permission takes precedence, then +write+, then lastly +read+. ==== == ACL Targets and Groups == ACL targets correspond to user accounts on the system. .Properties of an ACL Target [width="95%",cols="1m,<3",options="header",align="center"] |==== |Attribute |Description |id |The name of a user on the system (required) indexterm:[id,acl_target] indexterm:[access control list,acl_target,id] |==== ACL groups may be specified, but are not currently used by Pacemaker. This is expected to change in a future version. .Properties of an ACL Group [width="95%",cols="1m,<3",options="header",align="center"] |==== |Attribute |Description |id |The name of a group on the system (required) indexterm:[id,acl_group] indexterm:[access control list,acl_group,id] |==== Each +acl_target+ and +acl_group+ element may contain any number of +role+ elements. .Properties of an ACL Role Reference [width="95%",cols="1m,<3",options="header",align="center"] |==== |Attribute |Description |id |The +id+ of an +acl_role+ element that specifies permissions granted to the enclosing target or group indexterm:[id,role] indexterm:[access control list,role,id] |==== [IMPORTANT] ==== The +root+ and +hacluster+ user accounts always have full access to the CIB, regardless of ACLs. For other user accounts, when +enable-acl+ is true, permission to all parts of the CIB is denied by default (permissions must be explicitly granted). ==== == ACL Examples == [source,XML] ---- ---- In the above example, the user +alice+ has the minimal permissions necessary to run basic Pacemaker CLI tools, including using `crm_mon` to view the cluster status, without being able to modify anything. The user +bob+ can view the entire configuration and status of the cluster, but not make any changes. The user +carol+ can read everything, and change selected cluster properties as well as resource roles and location constraints. Finally, +dave+ has full read and write access to the entire CIB. Looking at the +minimal+ role in more depth, it is designed to allow read access to the +cib+ tag itself, while denying access to particular portions of its subtree (which is the entire CIB). This is because the DC node is indicated in the +cib+ tag, so `crm_mon` will not be able to report the DC otherwise. However, this does change the security model to allow by default, since any portions of the CIB not explicitly denied will be readable. The +cib+ read access could be removed and replaced with read access to just the +crm_config+ and +status+ sections, for a safer approach at the cost of not seeing the DC in status output. For a simpler configuration, the +minimal+ role allows read access to the entire +crm_config+ section, which contains cluster properties. It would be possible to allow read access to specific properties instead (such as +stonith-enabled+, +dc-uuid+, +have-quorum+, and +cluster-name+) to restrict access further while still allowing status output, but cluster properties are unlikely to be considered sensitive. diff --git a/doc/sphinx/Pacemaker_Explained/advanced-options.rst b/doc/sphinx/Pacemaker_Explained/advanced-options.rst index 5ee5c9b5ed..ec064b33af 100644 --- a/doc/sphinx/Pacemaker_Explained/advanced-options.rst +++ b/doc/sphinx/Pacemaker_Explained/advanced-options.rst @@ -1,759 +1,776 @@ Advanced Configuration ---------------------- .. Convert_to_RST: [[s-recurring-start]] == Specifying When Recurring Actions are Performed == By default, recurring actions are scheduled relative to when the resource started. So if your resource was last started at 14:32 and you have a backup set to be performed every 24 hours, then the backup will always run in the middle of the business day -- hardly desirable. To specify a date and time that the operation should be relative to, set the operation's +interval-origin+. The cluster uses this point to calculate the correct +start-delay+ such that the operation will occur at _origin + (interval * N)_. So, if the operation's interval is 24h, its interval-origin is set to 02:00 and it is currently 14:32, then the cluster would initiate the operation with a start delay of 11 hours and 28 minutes. If the resource is moved to another node before 2am, then the operation is cancelled. The value specified for +interval+ and +interval-origin+ can be any date/time conforming to the http://en.wikipedia.org/wiki/ISO_8601[ISO8601 standard]. By way of example, to specify an operation that would run on the first Monday of 2009 and every Monday after that, you would add: .Specifying a Base for Recurring Action Intervals ===== [source,XML] ===== - - [[s-failure-handling]] - == Handling Resource Failure == + + +.. _failure-handling: + +Handling Resource Failure +######################### + + +.. Convert_to_RST_2: By default, Pacemaker will attempt to recover failed resources by restarting them. However, failure recovery is highly configurable. === Failure Counts === Pacemaker tracks resource failures for each combination of node, resource, and operation (start, stop, monitor, etc.). You can query the fail count for a particular node, resource, and/or operation using the `crm_failcount` command. For example, to see how many times the 10-second monitor for +myrsc+ has failed on +node1+, run: ---- # crm_failcount --query -r myrsc -N node1 -n monitor -I 10s ---- If you omit the node, `crm_failcount` will use the local node. If you omit the operation and interval, `crm_failcount` will display the sum of the fail counts for all operations on the resource. You can use `crm_resource --cleanup` or `crm_failcount --delete` to clear fail counts. For example, to clear the above monitor failures, run: ---- # crm_resource --cleanup -r myrsc -N node1 -n monitor -I 10s ---- If you omit the resource, `crm_resource --cleanup` will clear failures for all resources. If you omit the node, it will clear failures on all nodes. If you omit the operation and interval, it will clear the failures for all operations on the resource. [NOTE] ==== Even when cleaning up only a single operation, all failed operations will disappear from the status display. This allows us to trigger a re-check of the resource's current status. ==== Higher-level tools may provide other commands for querying and clearing fail counts. The `crm_mon` tool shows the current cluster status, including any failed operations. To see the current fail counts for any failed resources, call `crm_mon` with the `--failcounts` option. This shows the fail counts per resource (that is, the sum of any operation fail counts for the resource). === Failure Response === Normally, if a running resource fails, pacemaker will try to stop it and start it again. Pacemaker will choose the best location to start it each time, which may be the same node that it failed on. However, if a resource fails repeatedly, it is possible that there is an underlying problem on that node, and you might desire trying a different node in such a case. Pacemaker allows you to set your preference via the +migration-threshold+ resource meta-attribute. footnote:[ The naming of this option was perhaps unfortunate as it is easily confused with live migration, the process of moving a resource from one node to another without stopping it. Xen virtual guests are the most common example of resources that can be migrated in this manner. ] If you define +migration-threshold=pass:[N]+ for a resource, it will be banned from the original node after 'N' failures. [NOTE] ==== The +migration-threshold+ is per 'resource', even though fail counts are tracked per 'operation'. The operation fail counts are added together to compare against the +migration-threshold+. ==== By default, fail counts remain until manually cleared by an administrator using `crm_resource --cleanup` or `crm_failcount --delete` (hopefully after first fixing the failure's cause). It is possible to have fail counts expire automatically by setting the +failure-timeout+ resource meta-attribute. [IMPORTANT] ==== A successful operation does not clear past failures. If a recurring monitor operation fails once, succeeds many times, then fails again days later, its fail count is 2. Fail counts are cleared only by manual intervention or falure timeout. ==== For example, a setting of +migration-threshold=2+ and +failure-timeout=60s+ would cause the resource to move to a new node after 2 failures, and allow it to move back (depending on stickiness and constraint scores) after one minute. [NOTE] ==== +failure-timeout+ is measured since the most recent failure. That is, older failures do not individually time out and lower the fail count. Instead, all failures are timed out simultaneously (and the fail count is reset to 0) if there is no new failure for the timeout period. ==== There are two exceptions to the migration threshold concept: when a resource either fails to start or fails to stop. If the cluster property +start-failure-is-fatal+ is set to +true+ (which is the default), start failures cause the fail count to be set to +INFINITY+ and thus always cause the resource to move immediately. Stop failures are slightly different and crucial. If a resource fails to stop and STONITH is enabled, then the cluster will fence the node in order to be able to start the resource elsewhere. If STONITH is not enabled, then the cluster has no way to continue and will not try to start the resource elsewhere, but will try to stop it again after the failure timeout. == Moving Resources == indexterm:[Moving,Resources] indexterm:[Resource,Moving] === Moving Resources Manually === There are primarily two occasions when you would want to move a resource from its current location: when the whole node is under maintenance, and when a single resource needs to be moved. ==== Standby Mode ==== Since everything eventually comes down to a score, you could create constraints for every resource to prevent them from running on one node. While pacemaker configuration can seem convoluted at times, not even we would require this of administrators. Instead, one can set a special node attribute which tells the cluster "don't let anything run here". There is even a helpful tool to help query and set it, called `crm_standby`. To check the standby status of the current machine, run: ---- # crm_standby -G ---- A value of +on+ indicates that the node is _not_ able to host any resources, while a value of +off+ says that it _can_. You can also check the status of other nodes in the cluster by specifying the `--node` option: ---- # crm_standby -G --node sles-2 ---- To change the current node's standby status, use `-v` instead of `-G`: ---- # crm_standby -v on ---- Again, you can change another host's value by supplying a hostname with `--node`. A cluster node in standby mode will not run resources, but still contributes to quorum, and may fence or be fenced by nodes. ==== Moving One Resource ==== When only one resource is required to move, we could do this by creating location constraints. However, once again we provide a user-friendly shortcut as part of the `crm_resource` command, which creates and modifies the extra constraints for you. If +Email+ were running on +sles-1+ and you wanted it moved to a specific location, the command would look something like: ---- # crm_resource -M -r Email -H sles-2 ---- Behind the scenes, the tool will create the following location constraint: [source,XML] It is important to note that subsequent invocations of `crm_resource -M` are not cumulative. So, if you ran these commands ---- # crm_resource -M -r Email -H sles-2 # crm_resource -M -r Email -H sles-3 ---- then it is as if you had never performed the first command. To allow the resource to move back again, use: ---- # crm_resource -U -r Email ---- Note the use of the word _allow_. The resource can move back to its original location but, depending on +resource-stickiness+, it might stay where it is. To be absolutely certain that it moves back to +sles-1+, move it there before issuing the call to `crm_resource -U`: ---- # crm_resource -M -r Email -H sles-1 # crm_resource -U -r Email ---- Alternatively, if you only care that the resource should be moved from its current location, try: ---- # crm_resource -B -r Email ---- Which will instead create a negative constraint, like [source,XML] This will achieve the desired effect, but will also have long-term consequences. As the tool will warn you, the creation of a +-INFINITY+ constraint will prevent the resource from running on that node until `crm_resource -U` is used. This includes the situation where every other cluster node is no longer available! In some cases, such as when +resource-stickiness+ is set to +INFINITY+, it is possible that you will end up with the problem described in <>. The tool can detect some of these cases and deals with them by creating both positive and negative constraints. E.g. +Email+ prefers +sles-1+ with a score of +-INFINITY+ +Email+ prefers +sles-2+ with a score of +INFINITY+ which has the same long-term consequences as discussed earlier. === Moving Resources Due to Connectivity Changes === You can configure the cluster to move resources when external connectivity is lost in two steps. ==== Tell Pacemaker to Monitor Connectivity ==== First, add an *ocf:pacemaker:ping* resource to the cluster. The *ping* resource uses the system utility of the same name to a test whether list of machines (specified by DNS hostname or IPv4/IPv6 address) are reachable and uses the results to maintain a node attribute called +pingd+ by default. footnote:[ The attribute name is customizable, in order to allow multiple ping groups to be defined. ] [NOTE] =========== Older versions of Pacemaker used a different agent *ocf:pacemaker:pingd* which is now deprecated in favor of *ping*. If your version of Pacemaker does not contain the *ping* resource agent, download the latest version from https://github.com/ClusterLabs/pacemaker/tree/master/extra/resources/ping =========== Normally, the ping resource should run on all cluster nodes, which means that you'll need to create a clone. A template for this can be found below along with a description of the most interesting parameters. .Common Options for a 'ping' Resource [width="95%",cols="1m,<4",options="header",align="center"] |========================================================= |Field |Description |dampen |The time to wait (dampening) for further changes to occur. Use this to prevent a resource from bouncing around the cluster when cluster nodes notice the loss of connectivity at slightly different times. indexterm:[dampen,Ping Resource Option] indexterm:[Ping Resource,Option,dampen] |multiplier |The number of connected ping nodes gets multiplied by this value to get a score. Useful when there are multiple ping nodes configured. indexterm:[multiplier,Ping Resource Option] indexterm:[Ping Resource,Option,multiplier] |host_list |The machines to contact in order to determine the current connectivity status. Allowed values include resolvable DNS host names, IPv4 and IPv6 addresses. indexterm:[host_list,Ping Resource Option] indexterm:[Ping Resource,Option,host_list] |========================================================= .An example ping cluster resource that checks node connectivity once every minute ===== [source,XML] ------------ ------------ ===== [IMPORTANT] =========== You're only half done. The next section deals with telling Pacemaker how to deal with the connectivity status that +ocf:pacemaker:ping+ is recording. =========== ==== Tell Pacemaker How to Interpret the Connectivity Data ==== [IMPORTANT] ====== Before attempting the following, make sure you understand <>. ====== There are a number of ways to use the connectivity data. The most common setup is for people to have a single ping target (e.g. the service network's default gateway), to prevent the cluster from running a resource on any unconnected node. .Don't run a resource on unconnected nodes ===== [source,XML] ------- ------- ===== A more complex setup is to have a number of ping targets configured. You can require the cluster to only run resources on nodes that can connect to all (or a minimum subset) of them. .Run only on nodes connected to three or more ping targets. ===== [source,XML] ------- ... ... ... ------- ===== Alternatively, you can tell the cluster only to _prefer_ nodes with the best connectivity. Just be sure to set +multiplier+ to a value higher than that of +resource-stickiness+ (and don't set either of them to +INFINITY+). .Prefer the node with the most connected ping nodes ===== [source,XML] ------- ------- ===== It is perhaps easier to think of this in terms of the simple constraints that the cluster translates it into. For example, if *sles-1* is connected to all five ping nodes but *sles-2* is only connected to two, then it would be as if you instead had the following constraints in your configuration: .How the cluster translates the above location constraint ===== [source,XML] ------- ------- ===== The advantage is that you don't have to manually update any constraints whenever your network connectivity changes. You can also combine the concepts above into something even more complex. The example below shows how you can prefer the node with the most connected ping nodes provided they have connectivity to at least three (again assuming that +multiplier+ is set to 1000). .A more complex example of choosing a location based on connectivity ===== [source,XML] ------- ------- ===== - - [[s-migrating-resources]] - === Migrating Resources === + + +.. _live-migration: + +Migrating Resources +___________________ + + +.. Convert_to_RST_3: Normally, when the cluster needs to move a resource, it fully restarts the resource (i.e. stops the resource on the current node and starts it on the new node). However, some types of resources, such as Xen virtual guests, are able to move to another location without loss of state (often referred to as live migration or hot migration). In pacemaker, this is called resource migration. Pacemaker can be configured to migrate a resource when moving it, rather than restarting it. Not all resources are able to migrate; see the Migration Checklist below, and those that can, won't do so in all situations. Conceptually, there are two requirements from which the other prerequisites follow: * The resource must be active and healthy at the old location; and * everything required for the resource to run must be available on both the old and new locations. The cluster is able to accommodate both 'push' and 'pull' migration models by requiring the resource agent to support two special actions: +migrate_to+ (performed on the current location) and +migrate_from+ (performed on the destination). In push migration, the process on the current location transfers the resource to the new location where is it later activated. In this scenario, most of the work would be done in the +migrate_to+ action and, if anything, the activation would occur during +migrate_from+. Conversely for pull, the +migrate_to+ action is practically empty and +migrate_from+ does most of the work, extracting the relevant resource state from the old location and activating it. There is no wrong or right way for a resource agent to implement migration, as long as it works. .Migration Checklist * The resource may not be a clone. * The resource must use an OCF style agent. * The resource must not be in a failed or degraded state. * The resource agent must support +migrate_to+ and +migrate_from+ actions, and advertise them in its metadata. * The resource must have the +allow-migrate+ meta-attribute set to +true+ (which is not the default). If an otherwise migratable resource depends on another resource via an ordering constraint, there are special situations in which it will be restarted rather than migrated. For example, if the resource depends on a clone, and at the time the resource needs to be moved, the clone has instances that are stopping and instances that are starting, then the resource will be restarted. The scheduler is not yet able to model this situation correctly and so takes the safer (if less optimal) path. Also, if a migratable resource depends on a non-migratable resource, and both need to be moved, the migratable resource will be restarted. + + +.. _node-health: + +Tracking Node Health +#################### - [[s-node-health]] - == Tracking Node Health == +.. Convert_to_RST_4: A node may be functioning adequately as far as cluster membership is concerned, and yet be "unhealthy" in some respect that makes it an undesirable location for resources. For example, a disk drive may be reporting SMART errors, or the CPU may be highly loaded. Pacemaker offers a way to automatically move resources off unhealthy nodes. === Node Health Attributes === Pacemaker will treat any node attribute whose name starts with +#health+ as an indicator of node health. Node health attributes may have one of the following values: .Allowed Values for Node Health Attributes [width="95%",cols="1,<3",options="header",align="center"] |========================================================= |Value |Intended significance |+red+ |This indicator is unhealthy indexterm:[Node health,red] |+yellow+ |This indicator is becoming unhealthy indexterm:[Node health,yellow] |+green+ |This indicator is healthy indexterm:[Node health,green] |'integer' |A numeric score to apply to all resources on this node (0 or positive is healthy, negative is unhealthy) indexterm:[Node health,score] |========================================================= === Node Health Strategy === Pacemaker assigns a node health score to each node, as the sum of the values of all its node health attributes. This score will be used as a location constraint applied to this node for all resources. The +node-health-strategy+ cluster option controls how Pacemaker responds to changes in node health attributes, and how it translates +red+, +yellow+, and +green+ to scores. Allowed values are: .Node Health Strategies [width="95%",cols="1m,<3",options="header",align="center"] |========================================================= |Value |Effect |none |Do not track node health attributes at all. indexterm:[Node health,none] |migrate-on-red |Assign the value of +-INFINITY+ to +red+, and 0 to +yellow+ and +green+. This will cause all resources to move off the node if any attribute is +red+. indexterm:[Node health,migrate-on-red] |only-green |Assign the value of +-INFINITY+ to +red+ and +yellow+, and 0 to +green+. This will cause all resources to move off the node if any attribute is +red+ or +yellow+. indexterm:[Node health,only-green] |progressive |Assign the value of the +node-health-red+ cluster option to +red+, the value of +node-health-yellow+ to +yellow+, and the value of +node-health-green+ to +green+. Each node is additionally assigned a score of +node-health-base+ (this allows resources to start even if some attributes are +yellow+). This strategy gives the administrator finer control over how important each value is. indexterm:[Node health,progressive] |custom |Track node health attributes using the same values as +progressive+ for +red+, +yellow+, and +green+, but do not take them into account. The administrator is expected to implement a policy by defining rules (see <>) referencing node health attributes. indexterm:[Node health,custom] |========================================================= === Measuring Node Health === Since Pacemaker calculates node health based on node attributes, any method that sets node attributes may be used to measure node health. The most common ways are resource agents or separate daemons. Pacemaker provides examples that can be used directly or as a basis for custom code. The +ocf:pacemaker:HealthCPU+ and +ocf:pacemaker:HealthSMART+ resource agents set node health attributes based on CPU and disk parameters. The +ipmiservicelogd+ daemon sets node health attributes based on IPMI values (the +ocf:pacemaker:SystemHealth+ resource agent can be used to manage the daemon as a cluster resource). In order to take advantage of this feature - firstly add the resource to your cluster, preferably as a cloned resource to constantly measure health on all nodes: ===== [source,XML] ------------ ------------ ===== This way attrd_updater will set proper status for each node running this resource. Any attribute matching "#health-[a-zA-z]+" will force cluster to migrate all resources from unhealthy node and place it on other nodes according to all constraints defined in your cluster. When the node is no longer faulty you can force the cluster to restart the cloned resource on faulty node and make it available to take resources, in this case since we are using HealthIOWait provider: ---- # attrd_updater -n "#health-iowait" -U "green" --node="" -d "60s" ---- == Reloading Services After a Definition Change == The cluster automatically detects changes to the definition of services it manages. The normal response is to stop the service (using the old definition) and start it again (with the new definition). This works well, but some services are smarter and can be told to use a new set of options without restarting. To take advantage of this capability, the resource agent must: . Accept the +reload+ operation and perform any required actions. _The actions here depend completely on your application!_ + .The DRBD agent's logic for supporting +reload+ ===== [source,Bash] ------- case $1 in start) drbd_start ;; stop) drbd_stop ;; reload) drbd_reload ;; monitor) drbd_monitor ;; *) drbd_usage exit $OCF_ERR_UNIMPLEMENTED ;; esac exit $? ------- ===== . Advertise the +reload+ operation in the +actions+ section of its metadata + .The DRBD Agent Advertising Support for the +reload+ Operation ===== [source,XML] ------- 1.1 Master/Slave OCF Resource Agent for DRBD ... ------- ===== . Advertise one or more parameters that can take effect using +reload+. + Any parameter with the +unique+ set to 0 is eligible to be used in this way. + .Parameter that can be changed using reload ===== [source,XML] ------- Full path to the drbd.conf file. Path to drbd.conf ------- ===== Once these requirements are satisfied, the cluster will automatically know to reload the resource (instead of restarting) when a non-unique field changes. [NOTE] ====== Metadata will not be re-read unless the resource needs to be started. This may mean that the resource will be restarted the first time, even though you changed a parameter with +unique=0+. ====== [NOTE] ====== If both a unique and non-unique field are changed simultaneously, the resource will still be restarted. ====== diff --git a/doc/sphinx/Pacemaker_Explained/resources.rst b/doc/sphinx/Pacemaker_Explained/resources.rst index 6e24844550..592e299121 100644 --- a/doc/sphinx/Pacemaker_Explained/resources.rst +++ b/doc/sphinx/Pacemaker_Explained/resources.rst @@ -1,964 +1,980 @@ +.. _resource: + Cluster Resources ----------------- .. Convert_to_RST: [[s-resource-primitive]] == What is a Cluster Resource? == indexterm:[Resource] A resource is a service made highly available by a cluster. The simplest type of resource, a 'primitive' resource, is described in this chapter. More complex forms, such as groups and clones, are described in later chapters. Every primitive resource has a 'resource agent'. A resource agent is an external program that abstracts the service it provides and present a consistent view to the cluster. This allows the cluster to be agnostic about the resources it manages. The cluster doesn't need to understand how the resource works because it relies on the resource agent to do the right thing when given a `start`, `stop` or `monitor` command. For this reason, it is crucial that resource agents are well-tested. Typically, resource agents come in the form of shell scripts. However, they can be written using any technology (such as C, Python or Perl) that the author is comfortable with. [[s-resource-supported]] == Resource Classes == indexterm:[Resource,class] Pacemaker supports several classes of agents: * OCF * LSB * Upstart * Systemd * Service * Fencing * Nagios Plugins === Open Cluster Framework === indexterm:[Resource,OCF] indexterm:[OCF,Resources] indexterm:[Open Cluster Framework,Resources] The OCF standard footnote:[See https://github.com/ClusterLabs/OCF-spec/tree/master/ra . The Pacemaker implementation has been somewhat extended from the OCF specs.] is basically an extension of the Linux Standard Base conventions for init scripts to: * support parameters, * make them self-describing, and * make them extensible OCF specs have strict definitions of the exit codes that actions must return. footnote:[ The resource-agents source code includes the `ocf-tester` script, which can be useful in this regard. ] The cluster follows these specifications exactly, and giving the wrong exit code will cause the cluster to behave in ways you will likely find puzzling and annoying. In particular, the cluster needs to distinguish a completely stopped resource from one which is in some erroneous and indeterminate state. Parameters are passed to the resource agent as environment variables, with the special prefix +OCF_RESKEY_+. So, a parameter which the user thinks of as +ip+ will be passed to the resource agent as +OCF_RESKEY_ip+. The number and purpose of the parameters is left to the resource agent; however, the resource agent should use the `meta-data` command to advertise any that it supports. The OCF class is the most preferred as it is an industry standard, highly flexible (allowing parameters to be passed to agents in a non-positional manner) and self-describing. For more information, see the http://www.linux-ha.org/wiki/OCF_Resource_Agents[reference] and the 'Resource Agents' chapter of 'Pacemaker Administration'. === Linux Standard Base === indexterm:[Resource,LSB] indexterm:[LSB,Resources] indexterm:[Linux Standard Base,Resources] 'LSB' resource agents are more commonly known as 'init scripts'. If a full path is not given, they are assumed to be located in +/etc/init.d+. Commonly, they are provided by the OS distribution. In order to be used with a Pacemaker cluster, they must conform to the LSB specification. footnote:[ See http://refspecs.linux-foundation.org/LSB_3.0.0/LSB-Core-generic/LSB-Core-generic/iniscrptact.html for the LSB Spec as it relates to init scripts. ] [WARNING] ==== Many distributions or particular software packages claim LSB compliance but ship with broken init scripts. For details on how to check whether your init script is LSB-compatible, see the 'Resource Agents' chapter of 'Pacemaker Administration'. Common problematic violations of the LSB standard include: * Not implementing the +status+ operation at all * Not observing the correct exit status codes for +start+/+stop+/+status+ actions * Starting a started resource returns an error * Stopping a stopped resource returns an error ==== [IMPORTANT] ==== Remember to make sure the computer is _not_ configured to start any services at boot time -- that should be controlled by the cluster. ==== [[s-resource-supported-systemd]] === Systemd === indexterm:[Resource,Systemd] indexterm:[Systemd,Resources] Some newer distributions have replaced the old http://en.wikipedia.org/wiki/Init#SysV-style["SysV"] style of initialization daemons and scripts with an alternative called http://www.freedesktop.org/wiki/Software/systemd[Systemd]. Pacemaker is able to manage these services _if they are present_. Instead of init scripts, systemd has 'unit files'. Generally, the services (unit files) are provided by the OS distribution, but there are online guides for converting from init scripts. footnote:[For example, http://0pointer.de/blog/projects/systemd-for-admins-3.html] [IMPORTANT] ==== Remember to make sure the computer is _not_ configured to start any services at boot time -- that should be controlled by the cluster. ==== === Upstart === indexterm:[Resource,Upstart] indexterm:[Upstart,Resources] Some newer distributions have replaced the old http://en.wikipedia.org/wiki/Init#SysV-style["SysV"] style of initialization daemons (and scripts) with an alternative called http://upstart.ubuntu.com/[Upstart]. Pacemaker is able to manage these services _if they are present_. Instead of init scripts, upstart has 'jobs'. Generally, the services (jobs) are provided by the OS distribution. [IMPORTANT] ==== Remember to make sure the computer is _not_ configured to start any services at boot time -- that should be controlled by the cluster. ==== === System Services === indexterm:[Resource,System Services] indexterm:[System Service,Resources] Since there are various types of system services (+systemd+, +upstart+, and +lsb+), Pacemaker supports a special +service+ alias which intelligently figures out which one applies to a given cluster node. This is particularly useful when the cluster contains a mix of +systemd+, +upstart+, and +lsb+. In order, Pacemaker will try to find the named service as: . an LSB init script . a Systemd unit file . an Upstart job === STONITH === indexterm:[Resource,STONITH] indexterm:[STONITH,Resources] The STONITH class is used exclusively for fencing-related resources. This is discussed later in <>. === Nagios Plugins === indexterm:[Resource,Nagios Plugins] indexterm:[Nagios Plugins,Resources] Nagios Plugins footnote:[The project has two independent forks, hosted at https://www.nagios-plugins.org/ and https://www.monitoring-plugins.org/. Output from both projects' plugins is similar, so plugins from either project can be used with pacemaker.] allow us to monitor services on remote hosts. Pacemaker is able to do remote monitoring with the plugins _if they are present_. A common use case is to configure them as resources belonging to a resource container (usually a virtual machine), and the container will be restarted if any of them has failed. Another use is to configure them as ordinary resources to be used for monitoring hosts or services via the network. The supported parameters are same as the long options of the plugin. [[primitive-resource]] == Resource Properties == These values tell the cluster which resource agent to use for the resource, where to find that resource agent and what standards it conforms to. .Properties of a Primitive Resource [width="95%",cols="1m,<6",options="header",align="center"] |========================================================= |Field |Description |id |Your name for the resource indexterm:[id,Resource] indexterm:[Resource,Property,id] |class |The standard the resource agent conforms to. Allowed values: +lsb+, +nagios+, +ocf+, +service+, +stonith+, +systemd+, +upstart+ indexterm:[class,Resource] indexterm:[Resource,Property,class] |type |The name of the Resource Agent you wish to use. E.g. +IPaddr+ or +Filesystem+ indexterm:[type,Resource] indexterm:[Resource,Property,type] |provider |The OCF spec allows multiple vendors to supply the same resource agent. To use the OCF resource agents supplied by the Heartbeat project, you would specify +heartbeat+ here. indexterm:[provider,Resource] indexterm:[Resource,Property,provider] |========================================================= The XML definition of a resource can be queried with the `crm_resource` tool. For example: ---- # crm_resource --resource Email --query-xml ---- might produce: .A system resource definition ===== [source,XML] ===== [NOTE] ===== One of the main drawbacks to system services (LSB, systemd or Upstart) resources is that they do not allow any parameters! ===== //// See https://tools.ietf.org/html/rfc5737 for choice of example IP address //// .An OCF resource definition ===== [source,XML] ------- ------- ===== .. _resource_options: Resource Options ################ .. Convert_to_RST_2: Resources have two types of options: 'meta-attributes' and 'instance attributes'. Meta-attributes apply to any type of resource, while instance attributes are specific to each resource agent. === Resource Meta-Attributes === Meta-attributes are used by the cluster to decide how a resource should behave and can be easily set using the `--meta` option of the `crm_resource` command. .Meta-attributes of a Primitive Resource [width="95%",cols="2m,2,<5",options="header",align="center"] |========================================================= |Field |Default |Description |priority |0 |If not all resources can be active, the cluster will stop lower priority resources in order to keep higher priority ones active. indexterm:[priority,Resource Option] indexterm:[Resource,Option,priority] |target-role |Started a|What state should the cluster attempt to keep this resource in? Allowed values: * +Stopped:+ Force the resource to be stopped * +Started:+ Allow the resource to be started (and in the case of <>, promoted to master if appropriate) * +Slave:+ Allow the resource to be started, but only in Slave mode if the resource is <> * +Master:+ Equivalent to +Started+ indexterm:[target-role,Resource Option] indexterm:[Resource,Option,target-role] |is-managed |TRUE |Is the cluster allowed to start and stop the resource? Allowed values: +true+, +false+ indexterm:[is-managed,Resource Option] indexterm:[Resource,Option,is-managed] |maintenance |FALSE |Similar to the +maintenance-mode+ <>, but for a single resource. If true, the resource will not be started, stopped, or monitored on any node. This differs from +is-managed+ in that monitors will not be run. Allowed values: +true+, +false+ indexterm:[maintenance,Resource Option] indexterm:[Resource,Option,maintenance] .. _resource-stickiness: placeholder .. Convert_to_RST_3: |resource-stickiness |1 for individual clone instances, 0 for all other resources |A score that will be added to the current node when a resource is already active. This allows running resources to stay where they are, even if they would be placed elsewhere if they were being started from a stopped state. indexterm:[resource-stickiness,Resource Option] indexterm:[Resource,Option,resource-stickiness] - + + +.. _requires: + + placeholder + +.. Convert_to_RST_4: + |requires |+quorum+ for resources with a +class+ of +stonith+, otherwise +unfencing+ if unfencing is active in the cluster, otherwise +fencing+ if +stonith-enabled+ is true, otherwise +quorum+ a|Conditions under which the resource can be started Allowed values: * +nothing:+ can always be started * +quorum:+ The cluster can only start this resource if a majority of the configured nodes are active * +fencing:+ The cluster can only start this resource if a majority of the configured nodes are active _and_ any failed or unknown nodes have been <> * +unfencing:+ The cluster can only start this resource if a majority of the configured nodes are active _and_ any failed or unknown nodes have been fenced _and_ only on nodes that have been <> indexterm:[requires,Resource Option] indexterm:[Resource,Option,requires] |migration-threshold |INFINITY |How many failures may occur for this resource on a node, before this node is marked ineligible to host this resource. A value of 0 indicates that this feature is disabled (the node will never be marked ineligible); by constrast, the cluster treats INFINITY (the default) as a very large but finite number. This option has an effect only if the failed operation specifies +on-fail+ as +restart+ (the default), and additionally for failed +start+ operations, if the cluster property +start-failure-is-fatal+ is +false+. indexterm:[migration-threshold,Resource Option] indexterm:[Resource,Option,migration-threshold] |failure-timeout |0 |How many seconds to wait before acting as if the failure had not occurred, and potentially allowing the resource back to the node on which it failed. A value of 0 indicates that this feature is disabled. indexterm:[failure-timeout,Resource Option] indexterm:[Resource,Option,failure-timeout] |multiple-active |stop_start a|What should the cluster do if it ever finds the resource active on more than one node? Allowed values: * +block:+ mark the resource as unmanaged * +stop_only:+ stop all active instances and leave them that way * +stop_start:+ stop all active instances and start the resource in one location only indexterm:[multiple-active,Resource Option] indexterm:[Resource,Option,multiple-active] |allow-migrate |TRUE for ocf:pacemaker:remote resources, FALSE otherwise |Whether the cluster should try to "live migrate" this resource when it needs to be moved (see <>) |container-attribute-target | |Specific to bundle resources; see <> |remote-node | |The name of the Pacemaker Remote guest node this resource is associated with, if any. If specified, this both enables the resource as a guest node and defines the unique name used to identify the guest node. The guest must be configured to run the Pacemaker Remote daemon when it is started. +WARNING:+ This value cannot overlap with any resource or node IDs. |remote-port |3121 |If +remote-node+ is specified, the port on the guest used for its Pacemaker Remote connection. The Pacemaker Remote daemon on the guest must be configured to listen on this port. |remote-addr |value of +remote-node+ |If +remote-node+ is specified, the IP address or hostname used to connect to the guest via Pacemaker Remote. The Pacemaker Remote daemon on the guest must be configured to accept connections on this address. |remote-connect-timeout |60s |If +remote-node+ is specified, how long before a pending guest connection will time out. |========================================================= As an example of setting resource options, if you performed the following commands on an LSB Email resource: ------- # crm_resource --meta --resource Email --set-parameter priority --parameter-value 100 # crm_resource -m -r Email -p multiple-active -v block ------- the resulting resource definition might be: .An LSB resource with cluster options ===== [source,XML] ------- ------- ===== In addition to the cluster-defined meta-attributes described above, you may also configure arbitrary meta-attributes of your own choosing. Most commonly, this would be done for use in <>. For example, an IT department might define a custom meta-attribute to indicate which company department each resource is intended for. To reduce the chance of name collisions with cluster-defined meta-attributes added in the future, it is recommended to use a unique, organization-specific prefix for such attributes. [[s-resource-defaults]] === Setting Global Defaults for Resource Meta-Attributes === To set a default value for a resource option, add it to the +rsc_defaults+ section with `crm_attribute`. For example, ---- # crm_attribute --type rsc_defaults --name is-managed --update false ---- would prevent the cluster from starting or stopping any of the resources in the configuration (unless of course the individual resources were specifically enabled by having their +is-managed+ set to +true+). === Resource Instance Attributes === The resource agents of some resource classes (lsb, systemd and upstart 'not' among them) can be given parameters which determine how they behave and which instance of a service they control. If your resource agent supports parameters, you can add them with the `crm_resource` command. For example, ---- # crm_resource --resource Public-IP --set-parameter ip --parameter-value 192.0.2.2 ---- would create an entry in the resource like this: .An example OCF resource with instance attributes ===== [source,XML] ------- ------- ===== For an OCF resource, the result would be an environment variable called +OCF_RESKEY_ip+ with a value of +192.0.2.2+. The list of instance attributes supported by an OCF resource agent can be found by calling the resource agent with the `meta-data` command. The output contains an XML description of all the supported attributes, their purpose and default values. .Displaying the metadata for the Dummy resource agent template ===== ---- # export OCF_ROOT=/usr/lib/ocf # $OCF_ROOT/resource.d/pacemaker/Dummy meta-data ---- [source,XML] ------- 1.0 This is a Dummy Resource Agent. It does absolutely nothing except keep track of whether its running or not. Its purpose in life is for testing and to serve as a template for RA writers. NB: Please pay attention to the timeouts specified in the actions section below. They should be meaningful for the kind of resource the agent manages. They should be the minimum advised timeouts, but they shouldn't/cannot cover _all_ possible resource instances. So, try to be neither overly generous nor too stingy, but moderate. The minimum timeouts should never be below 10 seconds. Example stateless resource agent Location to store the resource state in. State file Fake attribute that can be changed to cause a reload Fake attribute that can be changed to cause a reload Number of seconds to sleep during operations. This can be used to test how the cluster reacts to operation timeouts. Operation sleep duration in seconds. ------- ===== - - == Resource Operations == + + +.. _operation: + +Resource Operations +################### + + +.. Convert_to_RST_5: indexterm:[Resource,Action] 'Operations' are actions the cluster can perform on a resource by calling the resource agent. Resource agents must support certain common operations such as start, stop, and monitor, and may implement any others. Operations may be explicitly configured for two purposes: to override defaults for options (such as timeout) that the cluster will use whenever it initiates the operation, and to run an operation on a recurring basis (for example, to monitor the resource for failure). .An OCF resource with a non-default start timeout ===== [source,XML] ------- ------- ===== Pacemaker identifies operations by a combination of name and interval, so this combination must be unique for each resource. That is, you should not configure two operations for the same resource with the same name and interval. .. _operation_properties: Operation Properties ____________________ -.. Convert_to_RST_4: +.. Convert_to_RST_6: Operation properties may be specified directly in the +op+ element as XML attributes, or in a separate +meta_attributes+ block as +nvpair+ elements. XML attributes take precedence over +nvpair+ elements if both are specified. .Properties of an Operation [width="95%",cols="2m,3,<6",options="header",align="center"] |========================================================= |Field |Default |Description |id | |A unique name for the operation. indexterm:[id,Action Property] indexterm:[Action,Property,id] |name | |The action to perform. This can be any action supported by the agent; common values include +monitor+, +start+, and +stop+. indexterm:[name,Action Property] indexterm:[Action,Property,name] |interval |0 |How frequently (in seconds) to perform the operation. A value of 0 means "when needed". A positive value defines a 'recurring action', which is typically used with <>. indexterm:[interval,Action Property] indexterm:[Action,Property,interval] |timeout | |How long to wait before declaring the action has failed indexterm:[timeout,Action Property] indexterm:[Action,Property,timeout] |on-fail a|Varies by action: * +stop+: +fence+ if +stonith-enabled+ is true or +block+ otherwise * +demote+: +on-fail+ of the +monitor+ action with +role+ set to +Master+, if present, enabled, and configured to a value other than +demote+, or +restart+ otherwise * all other actions: +restart+ a|The action to take if this action ever fails. Allowed values: * +ignore:+ Pretend the resource did not fail. * +block:+ Don't perform any further operations on the resource. * +stop:+ Stop the resource and do not start it elsewhere. * +demote:+ Demote the resource, without a full restart. This is valid only for +promote+ actions, and for +monitor+ actions with both a nonzero +interval+ and +role+ set to +Master+; for any other action, a configuration error will be logged, and the default behavior will be used. * +restart:+ Stop the resource and start it again (possibly on a different node). * +fence:+ STONITH the node on which the resource failed. * +standby:+ Move _all_ resources away from the node on which the resource failed. indexterm:[on-fail,Action Property] indexterm:[Action,Property,on-fail] |enabled |TRUE |If +false+, ignore this operation definition. This is typically used to pause a particular recurring +monitor+ operation; for instance, it can complement the respective resource being unmanaged (+is-managed=false+), as this alone will <>. Disabling the operation does not suppress all actions of the given type. Allowed values: +true+, +false+. indexterm:[enabled,Action Property] indexterm:[Action,Property,enabled] |record-pending |TRUE |If +true+, the intention to perform the operation is recorded so that GUIs and CLI tools can indicate that an operation is in progress. This is best set as an _operation default_ (see <>). Allowed values: +true+, +false+. indexterm:[enabled,Action Property] indexterm:[Action,Property,enabled] |role | |Run the operation only on node(s) that the cluster thinks should be in the specified role. This only makes sense for recurring +monitor+ operations. Allowed (case-sensitive) values: +Stopped+, +Started+, and in the case of <>, +Slave+ and +Master+. indexterm:[role,Action Property] indexterm:[Action,Property,role] |========================================================= [NOTE] ==== When +on-fail+ is set to +demote+, recovery from failure by a successful demote causes the cluster to recalculate whether and where a new instance should be promoted. The node with the failure is eligible, so if master scores have not changed, it will be promoted again. There is no direct equivalent of +migration-threshold+ for the master role, but the same effect can be achieved with a location constraint using a <> with a node attribute expression for the resource's fail count. For example, to immediately ban the master role from a node with any failed promote or master monitor: [source,XML] ---- ---- This example assumes that there is a promotable clone of the +my_primitive+ resource (note that the primitive name, not the clone name, is used in the rule), and that there is a recurring 10-second-interval monitor configured for the master role (fail count attributes specify the interval in milliseconds). ==== [[s-resource-monitoring]] === Monitoring Resources for Failure === When Pacemaker first starts a resource, it runs one-time +monitor+ operations (referred to as 'probes') to ensure the resource is running where it's supposed to be, and not running where it's not supposed to be. (This behavior can be affected by the +resource-discovery+ location constraint property.) Other than those initial probes, Pacemaker will 'not' (by default) check that the resource continues to stay healthy. footnote:[Currently, anyway. Automatic monitoring operations may be added in a future version of Pacemaker.] You must configure +monitor+ operations explicitly to perform these checks. .An OCF resource with a recurring health check ===== [source,XML] ------- ------- ===== By default, a +monitor+ operation will ensure that the resource is running where it is supposed to. The +target-role+ property can be used for further checking. For example, if a resource has one +monitor+ operation with +interval=10 role=Started+ and a second +monitor+ operation with +interval=11 role=Stopped+, the cluster will run the first monitor on any nodes it thinks 'should' be running the resource, and the second monitor on any nodes that it thinks 'should not' be running the resource (for the truly paranoid, who want to know when an administrator manually starts a service by mistake). [NOTE] ==== Currently, monitors with +role=Stopped+ are not implemented for <> resources. ==== [[s-monitoring-unmanaged]] === Monitoring Resources When Administration is Disabled === Recurring +monitor+ operations behave differently under various administrative settings: * When a resource is unmanaged (by setting +is-managed=false+): No monitors will be stopped. + If the unmanaged resource is stopped on a node where the cluster thinks it should be running, the cluster will detect and report that it is not, but it will not consider the monitor failed, and will not try to start the resource until it is managed again. + Starting the unmanaged resource on a different node is strongly discouraged and will at least cause the cluster to consider the resource failed, and may require the resource's +target-role+ to be set to +Stopped+ then +Started+ to be recovered. * When a node is put into standby: All resources will be moved away from the node, and all +monitor+ operations will be stopped on the node, except those specifying +role+ as +Stopped+ (which will be newly initiated if appropriate). * When the cluster is put into maintenance mode: All resources will be marked as unmanaged. All monitor operations will be stopped, except those specifying +role+ as +Stopped+ (which will be newly initiated if appropriate). As with single unmanaged resources, starting a resource on a node other than where the cluster expects it to be will cause problems. [[s-operation-defaults]] === Setting Global Defaults for Operations === You can change the global default values for operation properties in a given cluster. These are defined in an +op_defaults+ section of the CIB's +configuration+ section, and can be set with `crm_attribute`. For example, ---- # crm_attribute --type op_defaults --name timeout --update 20s ---- would default each operation's +timeout+ to 20 seconds. If an operation's definition also includes a value for +timeout+, then that value would be used for that operation instead. === When Implicit Operations Take a Long Time === The cluster will always perform a number of implicit operations: +start+, +stop+ and a non-recurring +monitor+ operation used at startup to check whether the resource is already active. If one of these is taking too long, then you can create an entry for them and specify a longer timeout. .An OCF resource with custom timeouts for its implicit actions ===== [source,XML] ------- ------- ===== === Multiple Monitor Operations === Provided no two operations (for a single resource) have the same name and interval, you can have as many +monitor+ operations as you like. In this way, you can do a superficial health check every minute and progressively more intense ones at higher intervals. To tell the resource agent what kind of check to perform, you need to provide each monitor with a different value for a common parameter. The OCF standard creates a special parameter called +OCF_CHECK_LEVEL+ for this purpose and dictates that it is "made available to the resource agent without the normal +OCF_RESKEY+ prefix". Whatever name you choose, you can specify it by adding an +instance_attributes+ block to the +op+ tag. It is up to each resource agent to look for the parameter and decide how to use it. .An OCF resource with two recurring health checks, performing different levels of checks specified via +OCF_CHECK_LEVEL+. ===== [source,XML] ------- ------- ===== === Disabling a Monitor Operation === The easiest way to stop a recurring monitor is to just delete it. However, there can be times when you only want to disable it temporarily. In such cases, simply add +enabled=false+ to the operation's definition. .Example of an OCF resource with a disabled health check ===== [source,XML] ------- ------- ===== This can be achieved from the command line by executing: ---- # cibadmin --modify --xml-text '' ---- Once you've done whatever you needed to do, you can then re-enable it with ---- # cibadmin --modify --xml-text '' ---- diff --git a/doc/sphinx/Pacemaker_Explained/utilization.rst b/doc/sphinx/Pacemaker_Explained/utilization.rst index 4bcd777d62..3aa67adcac 100644 --- a/doc/sphinx/Pacemaker_Explained/utilization.rst +++ b/doc/sphinx/Pacemaker_Explained/utilization.rst @@ -1,232 +1,232 @@ +.. _utilization: + Utilization and Placement Strategy ---------------------------------- .. Convert_to_RST: - [[s-utilization]] - Pacemaker decides where to place a resource according to the resource allocation scores on every node. The resource will be allocated to the node where the resource has the highest score. If the resource allocation scores on all the nodes are equal, by the default placement strategy, Pacemaker will choose a node with the least number of allocated resources for balancing the load. If the number of resources on each node is equal, the first eligible node listed in the CIB will be chosen to run the resource. Often, in real-world situations, different resources use significantly different proportions of a node's capacities (memory, I/O, etc.). We cannot balance the load ideally just according to the number of resources allocated to a node. Besides, if resources are placed such that their combined requirements exceed the provided capacity, they may fail to start completely or run with degraded performance. To take these factors into account, Pacemaker allows you to configure: . The capacity a certain node provides. . The capacity a certain resource requires. . An overall strategy for placement of resources. == Utilization attributes == To configure the capacity that a node provides or a resource requires, you can use 'utilization attributes' in +node+ and +resource+ objects. You can name utilization attributes according to your preferences and define as many name/value pairs as your configuration needs. However, the attributes' values must be integers. .Specifying CPU and RAM capacities of two nodes ==== [source,XML] ---- ---- ==== .Specifying CPU and RAM consumed by several resources ==== [source,XML] ---- ---- ==== A node is considered eligible for a resource if it has sufficient free capacity to satisfy the resource's requirements. The nature of the required or provided capacities is completely irrelevant to Pacemaker -- it just makes sure that all capacity requirements of a resource are satisfied before placing a resource to a node. == Placement Strategy == After you have configured the capacities your nodes provide and the capacities your resources require, you need to set the +placement-strategy+ in the global cluster options, otherwise the capacity configurations have 'no effect'. Four values are available for the +placement-strategy+: +default+:: Utilization values are not taken into account at all. Resources are allocated according to allocation scores. If scores are equal, resources are evenly distributed across nodes. +utilization+:: Utilization values are taken into account 'only' when deciding whether a node is considered eligible (i.e. whether it has sufficient free capacity to satisfy the resource's requirements). Load-balancing is still done based on the number of resources allocated to a node. +balanced+:: Utilization values are taken into account when deciding whether a node is eligible to serve a resource 'and' when load-balancing, so an attempt is made to spread the resources in a way that optimizes resource performance. +minimal+:: Utilization values are taken into account 'only' when deciding whether a node is eligible to serve a resource. For load-balancing, an attempt is made to concentrate the resources on as few nodes as possible, thereby enabling possible power savings on the remaining nodes. Set +placement-strategy+ with `crm_attribute`: ---- # crm_attribute --name placement-strategy --update balanced ---- Now Pacemaker will ensure the load from your resources will be distributed evenly throughout the cluster, without the need for convoluted sets of colocation constraints. == Allocation Details == === Which node is preferred to get consumed first when allocating resources? === - The node with the highest node weight gets consumed first. Node weight is a score maintained by the cluster to represent node health. - If multiple nodes have the same node weight: * If +placement-strategy+ is +default+ or +utilization+, the node that has the least number of allocated resources gets consumed first. ** If their numbers of allocated resources are equal, the first eligible node listed in the CIB gets consumed first. * If +placement-strategy+ is +balanced+, the node that has the most free capacity gets consumed first. ** If the free capacities of the nodes are equal, the node that has the least number of allocated resources gets consumed first. *** If their numbers of allocated resources are equal, the first eligible node listed in the CIB gets consumed first. * If +placement-strategy+ is +minimal+, the first eligible node listed in the CIB gets consumed first. === Which node has more free capacity? === If only one type of utilization attribute has been defined, free capacity is a simple numeric comparison. If multiple types of utilization attributes have been defined, then the node that is numerically highest in the the most attribute types has the most free capacity. For example: - If +nodeA+ has more free +cpus+, and +nodeB+ has more free +memory+, then their free capacities are equal. - If +nodeA+ has more free +cpus+, while +nodeB+ has more free +memory+ and +storage+, then +nodeB+ has more free capacity. === Which resource is preferred to be assigned first? === - The resource that has the highest +priority+ (see <>) gets allocated first. - If their priorities are equal, check whether they are already running. The resource that has the highest score on the node where it's running gets allocated first, to prevent resource shuffling. - If the scores above are equal or the resources are not running, the resource has the highest score on the preferred node gets allocated first. - If the scores above are equal, the first runnable resource listed in the CIB gets allocated first. == Limitations and Workarounds == The type of problem Pacemaker is dealing with here is known as the http://en.wikipedia.org/wiki/Knapsack_problem[knapsack problem] and falls into the http://en.wikipedia.org/wiki/NP-complete[NP-complete] category of computer science problems -- a fancy way of saying "it takes a really long time to solve". Clearly in a HA cluster, it's not acceptable to spend minutes, let alone hours or days, finding an optimal solution while services remain unavailable. So instead of trying to solve the problem completely, Pacemaker uses a 'best effort' algorithm for determining which node should host a particular service. This means it arrives at a solution much faster than traditional linear programming algorithms, but by doing so at the price of leaving some services stopped. In the contrived example at the start of this chapter: - +rsc-small+ would be allocated to +node1+ - +rsc-medium+ would be allocated to +node2+ - +rsc-large+ would remain inactive Which is not ideal. There are various approaches to dealing with the limitations of pacemaker's placement strategy: Ensure you have sufficient physical capacity.:: It might sound obvious, but if the physical capacity of your nodes is (close to) maxed out by the cluster under normal conditions, then failover isn't going to go well. Even without the utilization feature, you'll start hitting timeouts and getting secondary failures. Build some buffer into the capabilities advertised by the nodes.:: Advertise slightly more resources than we physically have, on the (usually valid) assumption that a resource will not use 100% of the configured amount of CPU, memory and so forth 'all' the time. This practice is sometimes called 'overcommit'. Specify resource priorities.:: If the cluster is going to sacrifice services, it should be the ones you care about (comparatively) the least. Ensure that resource priorities are properly set so that your most important resources are scheduled first.