pmlogctl [-afNV?] [-c classname] [-i ident] [-p policyfile] command [host ...]
pmlogctl may be used to manage non-primary instances of the Performance Co-Pilot (PCP) archiver pmlogger(1). This would be most relevant in a PCP archive logger “farm” where many pmlogger(1) instances would be creating archives of performance data collected from pmcd(1) on many remote hosts.
The primary pmlogger(1) instance is closely linked to the local pmcd(1) process and as a consequence shares the same control infrastructure, namely systemd(1) or the PCP init(1) “rc scripts”. This is why the primary pmlogger(1) instance cannot be managed with pmlogctl.
For brevity in the description below, the term “instance” means a pmlogger(1) instance.
All instances managed by pmlogctl, pmlogger_check(1) and pmlogger_daily(1) abide by the following rules:
- Each instance is fetching performance data from a single pmcd(1) (i.e. one host), but each pmcd(1) may be providing performance data to zero, one or more pmlogger(1) processes running on one or more hosts.
- On the local host, each pmlogger(1) instance must be specified once in a pmlogger_check(1) control file and pmlogger(1) creates archives in a unique directory, named in the 4th parameter of the associated control file entry (see the CONFIGURATION section of pmlogger_check(1)).
- Each instance belongs to exactly one class, optionally named using a $class=... assignment in the associated control file. The special default class is reserved for all instances that do not have an associated $class=... assignment. For reporting purposes (refer to the summary command below), the primary pmlogger(1) instance is automatically assigned to the special primary class.
Each pmlogctl execution manages one or more instances updating the associated control files and then running pmlogger_check(1) to effect the desired change.
The host arguments are usually valid host names. For all commands except create and cond-create (described below) the host arguments may also be egrep(1) regular expressions that match the whole of a valid host name, so the pattern used is actually ^host$. For example foo.* (matches all host names beginning with “foo”) or .*foo (matches all host names ending with “foo”) or .*[fF][oO][oO].* (matches all host names containing “foo” in upper, lower or mixed case).
The combination of a classname from the optional -c option (or default) and the host arguments to each command identifies a target of set instances to which the command operation should be applied.
The -N option runs pmlogctl in “show me” mode where the intent of the command is shown, but no changes are made.
The -i option may be used with the create or cond-create commands to over-ride the instance identity that is specified in the ident section of the class policy file (see the Class Policy File section below). Since the identifier must be unique across all instances and all classes, it only makes sense to use this option when there is a single host argument.
The -V option generally increases the verbosity of the output. A second -V makes pmlogctl very chatty.
Given the tasks that pmlogctl is undertaking it usually must be run a “root”, the exceptions being the status command or when the -N option is specified.
command is one of the following:
Create new instances in the class classname (default by default) for the host hosts. The name of the class is used to identify a class policy file (see the Class Policy File section below) which provides a template for each new control file with %h replaced by host and %i replaced by the instance's (unique) identifier from the ident section of the class policy file or ident from the -i option.
The -p option may be used to identify a class policy file other than the one found in the standard place, i.e. $PCP_ETC_DIR/pcp/pmlogger/class.d/classname.
At least one host must be specified for the create command.
Similar to create but the instance will only be created if one of the conditions in the create section of associated class policy file(s) evaluates to true.
If a -c option is specified, then only the class classname will be considered, otherwise all classes will be considered. In the latter case, if none of the conditions in any of the classes evaluates to true, the the special pmfind class is used (this is the “default” class for the cond-create command when all else fails).
If more than one class evalutes to true then a composite pmlogger(1) configuration file is created and the instance is also assigned to the special pmfind class. The composite configuration file may contain a mix of explicit configuration clauses and the more general conditional configuration clauses as understood by the pmlogconf(1) tool.
Start the target set of instances.
If the target set intentionally contains more than one instance, then the -a option should be used (this is designed to reduce the risk of accidentally starting a potentially large number of pmlogger(1) processes).
If an instance in the target set is already started, no change is made (see the restart command to force new instances to be launched).
Stop the target set of instances.
If the target set intentionally contains more than one instance, then the -a option should be used (this is designed to reduce the risk of accidentally stopping a potentially large number of pmlogger(1) processes).
If an instance in the target set is already stopped, no change is made but a warning is issued.
Stop and then start the target set of instances.
If the target set intentionally contains more than one instance, then the -a option should be used (this is designed to reduce the risk of accidentally stopping and starting a potentially large number of pmlogger(1) processes).
If an instance in the target set is already stopped, a warning is issued before the instance is started.
Report the status of the target set of instances. If there is no -c option specified and no host arguments, i.e. the target set is empty, then all instances will be reported.
Destroy the target set of instances. At the end of this operation the associated pmlogger(1) processes will have been stopped and the relevant control file information removed.
If the target set intentionally contains more than one instance, then the -a option should be used (this is designed to reduce the risk of accidentally terminating a potentially large number of pmlogger(1) processes and removing their pmlogger_check(1) configurations).
The policy file linked to the class of an instance to be destroyed may contain a destroy section that may influence if and how the destroy operation should be performed. This may require a -p option to find the associated class policy file.
The class policy checking can be by-passed using the -f (force) command line option.
Class Policy File
A class policy file contains a number of sections, each section begins with a line that simply starts with the name of the section followed enclosed by “[” and “]”.
Lines beginning with a hash (#) are treated as comments and ignored. Blank or empty lines are also ignored.
The typical location of the policy file for the class foo is $PCP_ETC_DIR/pcp/pmlogger/class.d/foo.
The optional class name section names the class. If it is missing, the name of the policy file (stripped of any directory prefix) is used as the name of the class.
The ident section specifies the template to be used for the instance identifier to be given to each member of the class. This identifier needs to be unique across all instances and all classes, and it needs to be a valid file name in the local filesystem, so would normally contain the class name and the macro %h, e.g. foo-%h. The macro is replaced by the host when each instance is created.
The control section consists of one or more lines of template text that will be used to create the control file for each instance. This must at least include the pmlogger_check(1) control line to specify how to start the associated pmlogger(1) process; this line contains fields separated by white space as follows:
- the hostname, usually the macro %h
- n to indicate this is a non-primary instance
- the “socks” flag, typically n
- the directory in which the pmlogger(1) archives will be created; this needs to be unique and is usually specified using the pmlogger_check(1) macro PCP_ARCHIVE_DIR as the start of the path, followed by the instance identifier, usually the macro %i, e.g. PCP_ARCHIVE_DIR/%i
- additional parameters to pmlogger(1) which probably include at least a -c option to provide a configuration file that describes which metrics should be logged for instances of this class, which may be the same for all instances in this class, or it may include the %i macro to use a different configuration file for each instance. Note that if this configuration file does not exist, it will be created using pmlogconf(1) the first time pmlogger_check(1) is run.
Before the control line there should be a line that defines the version of the control line that follows, i.e.
If this is missing, pmlogctl will assume the version is 1.1 and insert the line when the instance is created.
pmlogctl will also add the class name during creation. e.g.
The create section defines the conditions that must be met before an instance will be created with the cond-create command. The intent is to allow different decisions to be made when a new host running pmcd(1) is discovered, e.g. by pmfind(1).
Each non-blank line in the create section is a condition of the form function(arg), where function is one of the following:
arg is the name of a performance metric and exits evaluates to true if that metric exists in the Performance Metrics Name Space (PMNS) on the remote host
arg is the name of a performance metric in the remote PMNS and values evaluates to true if some instance of that metric has a value on the remote host
arg is a derived metric expression in the format supported by pmRegisterDerived(3), and condition evaluates to true if the value of that expression on the remote host is greater than zero
arg is a regular expression in the style of egrep(1) and hostname evaluates to true if the remote host name matches arg
The destroy section defines the policy to be applied when an instance is destroyed.
The intent is to allow different decisions to be made when discovery service, e.g. pmfind(1), notices that a host associated with an instance is no longer present. But in the current version this is not implemented and the destroy section syntax and semantics is not yet defined.
A sample class policy file is as follows:
# policy file for the foo class [class] foo [ident] foo-%h [control] $version=1.1 %h n n PCP_ARCHIVE_DIR/%i -c foo-metrics.config [create] # matches all hosts hostname(.*) [destroy] # still to be defined
Most error or warning messages are self-explanatory.
Environment variables with the prefix PCP_ are used to parameterize the file and directory names used by PCP. On each installation, the file /etc/pcp.conf contains the local values for these variables. The $PCP_CONF variable may be used to specify an alternative configuration file, as described in pcp.conf(5).
egrep(1), init(1), PCPIntro(1), pmcd(1), pmlc(1), pmlogconf(1), pmlogger(1), pmlogger_check(1), pmlogger_daily(1), systemd(1), PMAPI(3), pmDerivedRegister(3) and pcp.conf(5).