pmlogger(1) create archive log for performance metrics


pmlogger [-c configfile] [-h host] [-K spec] [-l logfile] [-L] [-m note] [-n pmnsfile] [-o] [-p pid] [-P] [-r] [-s endsize] [-t interval] [-T endtime] [-u] [-U username] [-v volsize] [-V version] [-x fd] [-y] archive


pmlogger creates the archive logs of performance metric values that may be ``played back'' by other Performance Co-Pilot (see PCPIntro(1)) tools. These logs form the basis of the VCR paradigm and retrospective performance analysis services common to the PCP toolkit.

The mandatory argument archive is the base name for the physical files that constitute an archive log.

The -V option specifies the version for the archive that is generated. By default a version 2 archive is generated, and the only value currently supported for version is 2.

Unless directed to another host by the -h option or when directly using PMDAs via the -o option, pmlogger will contact the Performance Metrics Collector Daemon (PMCD) on the local host and use that as the source of the metric values to be logged.

To support the required flexibility and control over what is logged and when, pmlogger maintains an independent two level logging state for each instance of each performance metric. At the first (mandatory) level, logging is allowed to be on (with an associated interval between samples), or off or maybe. In the latter case, the second (advisory) level logging is allowed to be on (with an associated interval between samples), or off.

The mandatory level allows universal specification that some metrics must be logged, or must not be logged. The default state for all instances of all metrics when pmlogger starts is mandatory maybe and advisory off.

Use pmlc(1) to interrogate and change the logging state once pmlogger is running.

If a metric's state is mandatory (on or off) and a request is made to change it to mandatory maybe, the new state is mandatory maybe and advisory off. If a metric's state is already advisory (on or off) and a request is made to change it to mandatory maybe, the current state is retained.

It is not possible for pmlogger to log specific instances of a metric and all instances of the same metric concurrently. If specific instances are being logged and a request to log all instances is made, then all instances of the metric will be logged according to the new request, superseding any prior logging request for the metric. A request to log all instances of a metric will supersede any previous request to log all instances. A request to log specific instances of a metric when all instances are already being logged is refused. To do this one must turn off logging for all instances of the metric first. In each case, the validity of the request is checked first; for example a request to change a metric's logging state to advisory on when it is currently mandatory off is never permitted (it is necessary to change the state to mandatory maybe first).

Optionally, each system running pmcd(1) may also be configured to run a ``primary'' pmlogger instance. This pmlogger instance is launched by $PCP_RC_DIR/pmlogger, and is affected by the files $PCP_SYSCONF_DIR/pmlogger/control, $PCP_SYSCONF_DIR/pmlogger/control.d (use chkconfig(8) or similar platform-specific commands to activate or disable the primary pmlogger instance), $PCP_SYSCONFIG_DIR/pmlogger (environment variable settings for the primary pmlogger) $PCP_SYSCONF_DIR/pmlogger/pmlogger.options (command line options passed to the primary pmlogger) and $PCP_VAR_DIR/config/pmlogger/config.default (the default initial configuration file for the primary pmlogger).

The primary pmlogger instance is identified by the -P option. There may be at most one ``primary'' pmlogger instance on each system. The primary pmlogger instance (if any) must be running on the same host as the pmcd(1) to which it connects (if any), so the -h and -P options are mutually exclusive.

Logging of some metrics is possible even in the absence of a local pmcd(1), using the "local context" mode of operation. This is activated using the -o option, and causes pmlogger to make use of local DSO PMDAs instead of communicating with pmcd(1). When operating using a local context, the -K option may be used to control the DSO PMDAs that should be made accessible. The spec argument conforms to the syntax described in __pmSpecLocalPMDA(3). More than one -K option may be used.

When launched as a non-primary instance, pmlogger will exit immediately if the configuration file causes no metric logging to be scheduled. The -L option overrides this behavior, and causes a non-primary pmlogger instance to ``linger'', presumably pending some future dynamic re-configuration and state change via pmlc(1). pmlogger will also linger without the -L option being used if all the metrics to be logged are logged as once only metrics. When the once only metrics have been logged, a warning message will be generated stating that the event queue is empty and no more events will be scheduled.

By default all diagnostics and errors from pmlogger are written to the file pmlogger.log in the directory where pmlogger is launched. The -l option may be used to override the default behavior. If the log file cannot be created or is not writable, output is written to standard error instead.

If specified, the -s option instructs pmlogger to terminate after a certain size in records, bytes or time units has been accumulated. If endsize is an integer then endsize records will be written to the log. If endsize is an integer suffixed by b or bytes then endsize bytes of the archive data will be written out (note, however, that archive log record boundaries will not be broken and so this limit may be slightly surpassed). Other viable file size units include: K, Kb, Kbyte, Kilobyte for kilobytes and M, Mb, Mbyte, Megabyte for megabytes and G, Gb, Gbyte, Gigabyte for gigabytes. These units may be optionally suffixed by an s and may be of mixed case. Alternatively endsize may be an integer or a floating point number suffixed using a time unit as described in PCPIntro(1) for the interval argument (to the standard PCP -t command line option).

Some examples of different formats:
-s 100 
-s 100bytes
-s 100K 
-s 100Mb
-s 10Gbyte
-s 10mins
-s 1.5hours
The default is for pmlogger to run forever.

The -r option causes the size of the physical record(s) for each group of metrics and the expected contribution of the group to the size of the PCP archive for one full day of collection to be reported in the log file. This information is reported the first time each group is successfully written to the archive.

The -U option specifies the user account under which to run pmlogger. The default is the current user account for interactive use. When run as a daemon, the unprivileged "pcp" account is used in current versions of PCP, but in older versions the superuser account ("root") was used by default.

The log file is potentially a multi-volume data set, and the -v option causes pmlogger to start a new volume after a certain size in records, bytes, or time units has been accumulated for the current volume. The format of this size specification is identical to that of the -s option (see above). The default is for pmlogger to create a single volume log. Additional volume switches can also be forced asynchronously by either using pmlc(1) or sending pmlogger a SIGHUP signal (see below). Note, if a scheduled volume switch is in operation due to the -v option, then its counters will be reset after an asynchronous switch.

Independent of any -v option, each volume of an archive is limited to no more than 2^31 bytes, so pmlogger will automatically create a new volume for the archive before this limit is reached.

Normally pmlogger operates on the distributed Performance Metrics Name Space (PMNS), however if the -n option is specified an alternative local PMNS is loaded from the file pmnsfile.

Under normal circumstances, pmlogger will run forever (except for a -s option or a termination signal). The -T option may be used to limit the execution time using the format of time as prescribed by PCPIntro(1). The time is interpreted within the time zone of the PMCD server, unless the -y option is given, within which case the time zone at this logger host is used.

Some examples of different formats:
-T 10mins 
-T '@ 11:30' 
From this it can be seen that -T 10mins and -s 10mins perform identical actions.

Alternatively, pmlogger runtime may be limited to the lifetime of another process by using the -p or --PID option to nominate the PID of the process of interest. In this case the pmlogger will exit when the other process no longer exists.

When pmlogger receives a SIGHUP signal, the current volume of the log is closed, and a new volume is opened. This mechanism (or the alternative mechanism via pmlc(1)) may be used to manage the growth of the log files - once a log volume is closed, that file may be archived without ill-effect on the continued operation of pmlogger. See also the -v option above.

Historically the buffers for the current log may be flushed to disk using the flush command of pmlc(1), or by sending pmlogger a SIGUSR1 signal or by using the -u option. The current version of pmlogger and the libpcp routines that underpin pmlogger unconditionally use unbuffered writes and a single fwrite(3) for each logical record written, and so ``flushing'' does not force any additional data to be written to the file system. The -u option, the SIGUSR1 handling and the pmlc(1) flush command are retained for backwards compatibility.

When launched with the -x option, pmlogger will accept asynchronous control requests on the file descriptor fd. This option is only expected to be used internally by PCP applications that support ``live record mode''.

The -m option allows the string note to be appended to the map file for this instance of pmlogger in the $PCP_TMP_DIR/pmlogger directory. This is currently used internally to document the file descriptor (fd) when the -x option is used, or to indicate that this pmlogger instance was started under the control of pmlogger_check(1).


The configuration file may be specified with the -c option. If it is not, configuration specifications are read from standard input.

If configfile does not exist, then a search is made in the directory $PCP_VAR_DIR/config/pmlogger for a file of the same name, and if found that file is used, e.g. if config.mumble does not exist in the current directory and the file $PCP_VAR_DIR/config/pmlogger/config.mumble does exist, then -c config.mumble and -c $PCP_VAR_DIR/config/pmlogger/config.mumble are equivalent.

The syntax for the configuration file is as follows.

Words are separated by white space (space, tab or newline).
The symbol ``#'' (hash) introduces a comment, and all text up to the next newline is ignored.
Keywords (shown in bold below) must appear literally (i.e. in lower case).
Each specification begins with the optional keyword log, followed by one of the states mandatory on, mandatory off, mandatory maybe, advisory on or advisory off.
For the on states, a logging interval must follow using the syntax ``once'', or ``default'', or ``every N timeunits'', or simply ``N timeunits'' - N is an unsigned integer, and timeunits is one of the keywords msec, millisecond, sec, second, min, minute, hour or the plural form of one of the above.

Internal limitations require the interval to be smaller than (approximately) 74 hours. An interval value of zero is a synonym for once. An interval of default means to use the default logging interval of 60 seconds; this default value may be changed to interval with the -t command line option.

The interval argument follows the syntax described in PCPIntro(1), and in the simplest form may be an unsigned integer (the implied units in this case are seconds).
Following the state and possible interval specifications comes a ``{'', followed by a list of one or more metric specifications and a closing ``}''. The list is white space (or comma) separated. If there is only one metric specification in the list, the braces are optional.
A metric specification consists of a metric name optionally followed by a set of instance names. The metric name follows the standard PCP naming conventions, see pmns(5), and if the metric name is a non-leaf node in the PMNS (see pmns(5)), then pmlogger will recursively descend the PMNS and apply the logging specification to all descendent metric names that are leaf nodes in the PMNS. The set of instance names is a ``['', followed by a list of one or more space (or comma) separated names, numbers or strings, and a closing ``]''. Elements in the list that are numbers are assumed to be internal instance identifiers, other elements are assumed to be external instance identifiers - see pmGetInDom(3) for more information.

If no instances are given, then the logging specification is applied to all instances of the associated metric.

There may be an arbitrary number of logging specifications.
Following all of the logging specifications, there may be an optional access control section, introduced by the literal token [access]. Thereafter come access control rules that allow or disallow operations from particular hosts or groups of hosts.

The operations may be used to interrogate or control a running pmlogger using pmlc(1) and fall into the following classes:

interrogate the status of pmlogger and the metrics it is logging
Change advisory logging.
Change mandatory logging.
All of the above.

Access control rules are of the form ``allow hostlist : operationlist ;'' and ``disallow hostlist : operationlist ;''.

The hostlist follows the syntax and semantics for the access control mechanisms used by PMCD and are fully documented in pmcd(1). An operationslist is a comma separated list of the operations advisory, mandatory, enquire and all.

A missing [access] section allows all access and is equivalent to allow * : all;.

The configuration (either from standard input or configfile) is initially scanned by pmcpp(1) with the options -rs and -I $PCP_VAR_DIR/config/pmlogger. This extends the configuration file syntax with include file processing (%include), a common location to search for include files ($PCP_VAR_DIR/config/pmlogger), macro definitions (%define), macro expansion (%name and %{name}) and conditional inclusion of lines (%ifdef name ... %else ... %endif and %ifndef name ... %else ... %endif).


For each PCP utility, there is a sample pmlogger configuration file that could be used to create an archive log suitable for replaying with that tool (i.e. includes all of the performance metrics used by the tool). For a tool named foo this configuration file is located in $PCP_VAR_DIR/config/pmlogger/

The following is a simple default configuration file for a primary pmlogger instance, and demonstrates most of the capabilities of the configuration specification language.

log mandatory on once { hinv.ncpu hinv.ndisk }
log mandatory on every 10 minutes {
    disk.all.write [ "et0" ]
    network.interface.out.packets [ "et0" ]
    nfs.server.reqs [ "lookup" "getattr" "read" "write" ]
log advisory on every 30 minutes {
%include "macros.default"
%ifdef %disk_detail
log mandatory on %disk_detail_freq {
disallow * : all except enquire;
allow localhost : mandatory, advisory;


metadata (metric descriptions, instance domains, etc.) for the archive log
initial volume of metrics values (subsequent volumes have suffixes 1
, 2, ...)
temporal index to support rapid random access to the other files in the archive log
pmlogger maintains the files in this directory as the map between the process id of the pmlogger instance and the IPC port that may be used to control each pmlogger instance (as used by pmlc(1))
default configuration file for the primary logger instance launched from $PCP_RC_DIR/pmlogger
assorted configuration files suitable for creating logs that may be subsequently replayed with the PCP visualization and monitoring tools
Default directory for PCP archive files for performance metric values collected from the host hostname.
additional environment variables that will be set when the primary pmlogger instance executes. Only settings of the form "PMLOGGER_VARIABLE=value" will be honoured.
(or $PCP_LOG_DIR/pmlogger/hostname/pmlogger.log when started automatically by either $PCP_RC_DIR/pmlogger or one of the pmlogger(1) monitoring scripts such as pmlogger_check(1))
all messages and diagnostics are directed here


Normally pmlogger creates a socket to receive control messages from pmlc(1) on the first available TCP/IP port numbered 4330 or higher. The environment variable PMLOGGER_PORT may be used to specify an alternative starting port number.

If set to the value 1, the PMLOGGER_LOCAL environment variable will cause pmlogger to run in a localhost-only mode of operation, where it binds only to the loopback interface.

The PMLOGGER_MAXPENDING variable can be set to indicate the maximum length to which the queue of pending pmlc connections may grow.


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).


The archive logs are sufficiently precious that pmlogger will not truncate an existing physical file. A message of the form
__pmLogNewFile: "foo.index" already exists, not over-written
__pmLogCreate: File exists indicates this situation has arisen. You must explicitly remove the files and launch pmlogger again.

There may be at most one primary pmlogger instance per monitored host; attempting to bend this rule produces the error:
pmlogger: there is already a primary pmlogger running

Various other messages relating to the creation and/or deletion of files in $PCP_TMP_DIR/pmlogger suggest a permission problem on this directory, or some feral files have appeared therein.