Automatic class assignment in Workload Manager
The automatic assignment of processes to classes uses a set of class-assignment rules specified by a WLM administrator.
There are two levels of assignment rules:
- A set of assignment rules at the WLM configuration level used to determine which superclass a given process is assigned to.
- Each superclass with subclasses defined, in turn has a set of assignment rules used to determine which subclass of the superclass the process is assigned to.
The assignment rules at both levels are based on the values of a set of process attributes. These attributes are as follows:
- Process user ID
- Process group ID
- Path name of the application (program) executed
- Type of the process (
32bitor64bit, for example) - Process tag.
The tag is a process attribute, defined as a character string, that an application can set by program, using the WLM API.
The classification is done whenever an attribute changes by comparing the value of these process attributes against lists of possible values given in the class assignment rules file (called rules). The comparison determines which rule is a match for the current value of the process attributes.
A class assignment rule is a text string that includes the following fields, separated by one or more spaces:
| Item | Description |
|---|---|
| Name | Must contain the name of a class which is defined in the class file corresponding to the level of the rules file (superclass or subclass). Class names can contain only uppercase and lowercase letters, numbers, and underscores and can be up to 16 characters in length. No assignment rule can be specified for the system defined classes Unclassified, Unmanaged and Shared. |
| Reserved | Reserved for future extension. Its value must be a hyphen (-), and it must be present. |
| User | Can contain either a hyphen (-) or at least one valid user name (as defined in the /etc/passwd file). The list is composed of one or more names, separated by a comma (,). An exclamation mark (!) can be used before a name to exclude a given user from the class. Patterns can be specified to match a set of user names, using full Korn shell pattern-matching syntax. If there are no valid user names, the rule is ignored. |
| Group | Can contain either a hyphen (-) or at least one valid group name (as defined in the /etc/group file). The list is composed of one or more names, separated by a comma (,). An exclamation mark (!) can be used before a name to exclude a given group from the class. Patterns can be specified to match a set of group names using full Korn shell pattern matching syntax. If there are no valid group names, the rule is ignored. |
| Application | Can contain either a hyphen (-) or a list of application path names.
This is the path name of the applications (programs) executed by processes
included in the class. The application names will be either full path names
or Korn shell patterns that match path names. The list is composed of one
or more path names, separated by a comma (,). An exclamation mark (!) can
be used before a name to exclude a given application. At least one application in the list must be found at load time or the rule is ignored. Rules that are initially ignored for this reason might become effective later on if a file system is mounted that contains one or more applications in the list. |
| Type | Can contain either a hyphen (-) or a list of process attributes. The
possible values for these attributes are:
The value of the type field can be a combination of one or more of the above attributes separated by a plus (+). The 32bit and 64bit values are mutually exclusive. |
| Tag | May contain either a hyphen (-) or a list of application tags. An application tag is a string of up to 30 alphanumeric characters. The list is composed of one or more application tag values separated by commas. |
The User, Group, Application, and Tag attributes can be an attribute value grouping.
When a process is created (fork), it remains in the same class as its parent. Reclassification happens when the new process issues a system call which can modify one of the attributes of the process used for classification; for example, exec, setuid (and related calls), setgid (and related calls), setpri and plock.
To classify the process, WLM examines the top-level rules file for the active configuration to determine which superclass the process belongs. For each rule in the file, WLM checks the current values for the process attributes against the values and lists of values specified in the rule. Rules are checked in the order that they appear in the file. When a match is found, the process is assigned to the superclass named in the first field of the rule. Then the rules file for the superclass is examined in the same way to determine to which subclass the process should be assigned.
For a process to match one of the rules, each of its attributes must match the corresponding field in the rule. The following is a list of the criteria used to determine whether the value of an attribute matches the values in the field of the rules file:
- If the field in the rules file has a value of hyphen (-), then any value of the corresponding process attribute is a match.
- For all the attributes except type, if the value of the process attribute matches one of the values in the list in the rules file that is not excluded (prefaced by a "!"), then a match has occurred.
- For the type attribute, if one of the values in the rule is comprised of two or more values separated by a plus (+), then a process is a match only if its characteristics match all the values.
At both superclass and subclass levels, WLM goes through the rules in the order in which they appear in the rules file, and classifies the process in the class corresponding to the first rule for which the process is a match. The order of the rules in the rules file is therefore extremely important. Use caution when you create or modify the rules file.