diff --git a/ra/resource-agent-api.txt b/ra/resource-agent-api.txt index 60b8bbc..cb89469 100644 --- a/ra/resource-agent-api.txt +++ b/ra/resource-agent-api.txt @@ -1,416 +1,432 @@ DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT 0. Header Topic: Open Clustering Framework Resource Agent API Editor: Lars Marowsky-Brée Revision: $Id$ URL: http://www.opencf.org/standards/resource-agent-api.txt Copyright (c) 2002 by Lars Marowsky-Brée. This material may be distributed only subject to the terms and conditions set forth in the Open Publication License, v1.0 or later (the latest version is presently available at http://www.opencontent.org/openpub/). - 1. Abstract Resource Agents (RA) are the middle layer between the Resource Manager (RM) and the actual resources being managed. They aim to integrate the resource with the RM without any modifications to the actual resource provider itself, by encapsulating it carefully and thus making it movable between real nodes in a cluster. The RAs are obviously very specific to the resource type they are encapsulating, however there is no reason why they should be specific to a particular RM. The API described in this document should be general enough that a compliant Resource Agent can be used by all existing resource managers / fail-over systems who chose to implement this API either exclusively or in addition to their existing one. 1.1. Scope This document documents a common API for the RM to call the RAs so the pool of available RAs can be shared by the different clustering solutions. It does NOT define any libraries or helper functions which RAs might share with regard to common functionality like external command execution, cluster logging et cetera, as these are NOT specific to RA and are defined in the respective standards. 1.2. API version described This document currently describes version 1.1 of the API. 2. Terms used in this document 2.1. "Resource" A single physical or logical entity that provides a service to clients or other resources. For example, a resource can be a single disk volume, a particular network address, or an application such as a web server. A resource is generally available for use over time on two or more nodes in a cluster, although it usually can be allocated to only one node at any given time. Resources are identified by the combination of their type and name. The name is a special case of an instance parameter; the name/resource type combination is required to be unique in the cluster. A resource may also have instance parameters which provide additional information required for Resource Agent to control the resource. A resource may have dependencies on other resources or capabilities provided by other resources. Common examples include a dependency on an IP address being configured or a file system being mounted. These are special cases of instance parameters and are treated in much the same fashion. 2.2. "Resource types" A resource type represents a set of resources which share a common set of instance parameters and a common set of actions which can be performed on resource of the given type. 2.3. "Resource agent" A RA provides the actions ("member functions") for a a given type of resources; by providing the RA with the instance parameters, it is used to control a specific resource. They are usually implemented as shell scripts, but the API described here does not require this. Although this is somewhat similar to SystemV init scripts as described by the LSB, there are some differences explained below. 2.4. "Instance parameters" Instance parameters are the attributes which uniquely identify a given resource instance. It is recommended that the set of instance parameters for any given type of resources to be as minimal as possible. An instance parameter has a given name and value. They are both case sensitive and must satisfy the requirements of POSIX environment name/value combinations. 2.5. "Resource group" This is a term from the RM world, but it is explained in brief here for completeness. As explained above, a complex resource commonly has dependencies on other resources required for proper operation; all dependencies required to provide an actual service to the user are usually grouped into a "resource group" which is handled as an atomic unit by the cluster, as it isn't possible to move a resource without also moving its dependencies or only moving a resource but not the resources which depend on it. While the resource grouping is still commonly implemented by manual configuration, the information provided by the RA meta data should be sufficient for the RM to build the dependency tree to determine the ordering of resource startup and shutdown. 3. API 3.1. API Version Numbers The version number is of the form "x.y", where x and y are integer numbers greater than zero. x is referred to as the "major" number, and y the "minor" number. The major number must be increased if a _backwards incompatible_ change is made to the API. A major number mismatch between the RA and the RM must be reported as an error to the administrator. The minor number must be increased if _any_ change at all is made to the API. If the major is increased, the minor number is reset to "1". The minor number can be used by both sides to see whether a certain additional feature is supported by the other party. 3.2. Paths The Resource Agents are located in subdirectories under "/usr/ocf/resource.d"; the filename of the RA maps to the resource type provided and maybe a symlink to the real location. The subdirectories allow the installation of multiple RAs for the same type, but from different vendors or package versions: FailSafe -> FailSafe-1.1.0/ FailSafe-1.0.4/ FailSafe-1.1.0/ heartbeat -> heartbeat-0.4.9.1/ heartbeat-0.4.9.1/ How the RM decides on which of several RAs for a specific resource type installed it calls is implementation specific. 3.3. Execution syntax After the RM has identified the executable to call, it will be called in the following format: /usr/ocf/resource.d/... 3.4. Resource Agent actions A RA must be able to perform the following actions on a given resource on request by the RM; additional actions may be supported by the script for example for LSB compliance. In general, a RA should not assume it is the only RA of its type running at any given time because the RM might start several RA instances for multiple independent resource instances in parallel. _Mandatory_ actions must be supported; _not mandatory_ operations must be advertised in the meta data if supported. If the RM tries to call a not supported, not mandatory action, the RA should return an error. 3.4.1. start Mandatory. This brings the resource online and makes it available for use. It should NOT terminate before the resource has been fully started. It may try to implement recover actions for certain cases of startup failures. "start" must succeed if the resource instance is already running. 3.4.2. stop Mandatory. This stops the resource. After the "stop" command has completed, nothing should remain active of the resource and it must be possible to start it on the same node or another node. Only if this cannot be guaranteed should it report failure; stopping an already stopped resource must succeed. The "stop" request by the RM includes the authorization to bring down the resource even by force as long data integrity is maintained; breaking currently active transactions should be avoided, but the request to offline the resource has higher precedence than this. The "stop" action should also perform clean-ups of artifacts like leftover shared memory segments, semaphores, IPC message queues, lock files etc. 3.4.3. monitor Mandatory. Verifies whether a resource is working correctly. This should be "light-weight" query as it is called by the RM fairly often to poll the status of the resource. It is accepted practice to have additional instance parameters which are not strictly required to identify the resource instance but are needed to monitor it or customize of how intrusive this check is allowed to be. 3.4.4. restart Not mandatory. A special case of the "start" action, this should try to recover a resource locally. If this is not fully supported, it should be mapped to a stop/start action by the RM. An example includes "recovering" an IP address by moving it to another interface; this is much less costly than initiating a full resource group fail over to another node. 3.4.5. reload Not mandatory. Reload the configuration file without breaking currently connected users. 3.4.6. meta-data Mandatory. Returns the resource agent meta data via stdout. 3.4.7. validate-all Not mandatory. Validate the instance parameters provided. Perform a syntax check and if possible, a semantical check on the instance parameters. 3.5. Parameter passing The instance parameters and some additional attributes are passed in via the environment; this has been chosen because it does not reveal the parameters to an unprivileged user on the same system and environment variables can be easily accessed by all programming languages and shell scripts. 3.5.1. Syntax for instance parameters They are directly converted to environment variables; the name is prefixed with "OCF_RESKEY_". The instance parameter "force" with the value "yes" thus becomes: OCF_RESKEY_force=yes in the environment. See section 4. for a more formal explanation of instance parameters. 3.5.2. Special parameters The entire environment variable name space starting with OCF_ is considered to be reserved. Currently, the following additional parameters are defined: OCF_RA_VERSION_MAJ OCF_RA_VERSION_MIN Version number of the OCF Resource Agent API. If the script does not support this revision, it should report an error. See 3.1. for an explanation of the versioning scheme used. The version number is split into two numbers for ease of use in shell scripts. These two may be used by the RA to determine whether it is run under an OCF compliant RM. Example: OCF_RA_VERSION_MAJ=1 OCF_RA_VERSION_MIN=1 OCF_ROOT Referring to the root of the OCF directory hierarchy. Example: OCF_ROOT=/usr/ocf OCF_MONITOR_SKIP_QOS If set, the monitor action may skip expensive checks regarding the quality of the service. This effectively removes the distinction between the fine-grained states 0-119 and can be used if the RM is only interested in whether the resource is active at all - whether partially or fully - or not. 3.6. Exit codes These exit codes were largely modeled after the LSB 1.1.0 spec for compatibility, except for the "monitor" action. 3.6.1. "monitor" OK for fail-over / restart concerns: 0 - 29 running OK 30 - 59 running with warning / minor QoS offset NOT OK wrt fail-over / restart: 60 - 89 running with error / partially / QoS violation 90 - 119 not running, but with error / not cleaned up 120 - 149 not running at all (cleanly stopped) Hard errors: 150 - 179 Execution error (called with wrong syntax etc) 180 - 255 Reserved 3.6.2. "start", "stop", "restart", "reload" 0 No error, acton succeeded completely 1 generic or unspecified error (current practice) 2 invalid or excess argument(s) 3 unimplemented feature (for example, "reload") 4 user had insufficient privilege 5 program is not installed 6 program is not configured 7 program is not running 8-99 reserved for future LSB use 100-149 reserved for distribution use 150-199 reserved for application use 200-254 reserved 3.6.3. "validate-all" 0 No error 1 Semantical error 2 Syntactical error in at least one of the fields 8-99 reserved for future LSB use 100-149 reserved for distribution use 150-199 reserved for application use 200-254 reserved 3.6.4. "meta-data" 0 No error 1-254 Hard Resource Agent failure 4. Relation to the LSB It is required that the current LSB spec is fully supported by the system. The API tries to make it possible to have RA function both as a normal LSB init script and a cluster-aware RA, but this is not required functionality. The RAs could however use the helper functions defined for LSB init scripts. 5. RA meta data 5.1. Format We have the following requirements which are not fulfilled by the LSB way of embedding meta data into the beginning of the init scripts: - Independent of the language the RA is actually written in, - Extensible, - Structured, - Easy to parse from a variety of languages. This is why we use simple XML to describe the RA meta data. The DTD for this API can be found at http://www.opencf.org/standards/ra-api-1.dtd. 5.2. Semantics An example of a vs fully supported by the system. The API tries to make it possible to have RA function both as a normal LSB init script and a cluster-aware RA, but this is not required functionality. The RAs could however use the helper functions defined for LSB init scripts. 5. RA meta data 5.1. Format We have the following requirements which are not fulfilled by the LSB way of embedding meta data into the beginning of the init scripts: - Independent of the language the RA is actually written in, - Extensible, - Structured, - Easy to parse from a variety of languages. +This is why we use simple XML to describe the RA meta data. -This is why we use simple XML to describe the RA meta data. The DTD for this -API can be found at http://www.opencf.org/standards/ra-api-1 \ No newline at end of file +C. To-do list + +Move the terminology definitions out into a separate document common to all +OCF work. + +An interface where the RA asynchronously informs the RM of failures is planned +but not defined yet. + + +Contributors: + Greg Freemyer + Ragnar Kjørstad + Lars Marowsky-Brée + Alan Robertson + +B. Change Log + +DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT DRAFT