PCPIntro (3) - Linux Man Pages
PCPIntro: introduction to the Performance Co-Pilot (PCP) libraries
NAMEPCPIntro - introduction to the Performance Co-Pilot (PCP) libraries
INTRODUCTIONPerformance Co-Pilot (PCP) is a toolkit designed for monitoring and managing system-level performance.
The PCP libraries support the APIs required to create new performance monitoring tools and new agents (or PMDAs) to export performance data. The libpcp library is used in both cases. The libpcp_pmda library is used only for PMDAs.
Individual library routines are documented in their own manual page entries.
Most routines return an integer value; greater than equal to zero for success and less than zero for an error. The error codes have symbolic names defined in <pcp/pmapi.h>. Other negative values are used to encode errors that can be mapped to the traditional errno values defined in <errno.h>, with the value negated. To translate all PCP error codes into useful messages use either pmerr(1) or pmErrStr(3); the latter may also be used to decode the -errno cases.
GENERAL ERRORSThese common errors may occur in various PCP interactions.
Timeout waiting for a response from PMCD
Many interactions between PCP applications involve synchronous message passing, and these are subject to timeout constraints. These errors are most frequently encountered when using network connections with slow data rates or long latencies.
Metric not supported by this version of monitored application
Some performance metrics are unavailable from specific versions of the associated PMDA, or may be unavailable because the underlying instrumentation has changed or is not installed or is not enabled. This error is used in results from pmFetch(3) to indicate these situations.
IPC protocol failure
Generic protocol failure on a pipe or socket connecting two PCP applications, eg. client-pmcd, or client-pmtime, or PMDA-pmcd or pmlc-pmlogger.
Oneline or help text is not available
Set by a PMDA, and passed back to a PCP client, to indicate that the PMDA is unable to supply the requested metric or instance domain help text.
Missing metric value(s)
This error is used for a number of conditions in which the value of a performance metric is inappropriate for the context in which it is being used, eg.
Unknown metric name
Just what the message says.
Unknown or illegal metric identifier
Just what the message says.
Unknown or illegal instance domain identifier
A request nominates an instance domain that is unknown or PM_INDOM_NULL. May occur as a consequence of the instance domain identifier passed by a PCP client to pmGetInDom(3), pmLookupInDom(3), pmNameInDom(3), pmGetInDomArchive(3), pmLookupInDomArchive(3), pmNameInDomArchive(3) or a request passed from pmcd(1) to a PMDA.
IPC channel closed
End of file on a pipe or socket connecting two PCP applications, eg. client-pmcd, or client-pmtime or PMDA-pmcd.
Attempt to use an illegal context
Typically caused by a PCP client that tries to make calls before calling pmNewContext(3) or after calling pmDestroyContext(3).
No permission to perform requested operation
PCP-specific access controls apply to pmcd(1) and pmlogger(1). Platform-specific permission errors are returned as -EPERM.
Impossible value or scale conversion
Some value conversion requests are illegal, eg. bad debug flag symbolic name for -D option, or asking pmExtractValue(3) to translate non-numeric data types to numbers and vice versa.
Truncation in value conversion
Some conversion requests to pmExtractValue(3) cannot be performed based on the metric types and values involved, in this case conversion would result in loss of precision.
Negative value in conversion to unsigned
Some conversion requests to pmExtractValue(3) cannot be performed based on the metric types and values involved, in this case converting a negative value to an unsigned value.
Unknown or illegal metric type
The metric type is held in the metric descriptor and sometimes encoded in the metric values returned from a call to pmFetch(3). Legal values for the metric type are defined by the PM_TYPE_* macros in <pcp/pmapi.h>.
Illegal pmUnits specification
Some conversion requests to pmConvScale(3) cannot be performed due to illegal or incompatible specifications of the source and destination units.
Explicit instance identifier(s) required
Some PMDAs, especially the proc PMDA, will not return ``all instances'' for a pmFetch(3) request due to the cost. The client must explicitly built an instance profile using pmAddProfile(3) and/or pmDelProfile(3) before calling pmFetch(3). See also the -F option to pminfo(1).
Unknown or illegal instance identifier
A request to a PMDA nominates an instance that is unknown. May occur as a consequence of the profile established prior to a pmFetch(3) call, or an explicit instance name or identifier to pmLookupInDom(3) or pmNameInDom(3).
Illegal mode specification
Illegal mode argument to pmSetMode(3).
NULL pmInDom with non-NULL instlist
Bad arguments passed from a PCP client to pmAddProfile(3).
Insufficient elements in list
Parameter passing error by caller specifying a list with less than one element to pmFetch(3), pmLookupName(3) or pmStore(3).
Operation not supported for multi-threaded applications
As documented in PMAPI(3) and elsewhere, some libpcp routines are intended solely for use from single-threaded applications.
Result size exceeded
Indicates a fatal error in the message (or PDU) passing protocol between two PCP applications. This is an internal error, and other than an exotic networking failure, should not occur.
PMCD reset or configuration change
Refer to pmFetch(3) for an alternative mechanism that may be used to notify a PCP client when pmcd(1) has experienced one or more configuration changes since the last request from the client. Usually these changes involve a change to the namespace exported via pmcd and/or changes to the PMDAs under pmcd's control.
QA fault injected
Used only for PCP Quality Assurance (QA) testing.
Functionality not yet implemented
Self explanatory and rarely used.
Generic error, already reported above
Rarely used, this error may be returned when the error condition is a consequent of some earlier returned error and a more precise characterization is not possible.
CLIENT-PMCD ERRORSThese errors may occur in the interactions between a PCP client and pmcd(1) providing real-time performance data.
No PMCD agent for domain of request
A request sent to pmcd(1) requires information from an agent or PMDA that does not exist. Usually this means the namespace being used by the client application contains metric names from a previously installed PMDA.
PMCD connection limit for this host exceeded
The client connection limit for pmcd(1) is controlled by the optional access controls in $PCP_PMCDCONF_PATH. By default there is no limit imposed by the PCP code, and this error would not be seen.
Try again. Information not currently available
Used to notify a PCP client that the PMDA responsible for delivering the information is temporarily unavailable. See also PM_ERR_PMDANOTREADY.
Missing profile - protocol botch
Internal error in the communication between a client application and pmcd(1) - should not occur.
CLIENT-ARCHIVE ERRORSThese errors may occur in the interactions between a PCP client and the library routines that provide historical performance data from PCP archives created by pmlogger(1).
Missing archive file
Each PCP archive consists of multiple physical files as described in pmlogger(1). This error occurs when one of the physical files is missing or cannot be opened for reading.
End of PCP archive log
An attempt is made to read past the end file of the last volume of a set of PCP archives, or past the end of the time window (as specified with a -T option) for a set of PCP archives.
Operation requires context with host source of metrics
Operations involving help text (ie. pmLookupText(3) and pmLookupInDomText(3)) or calls to pmStore(3) require a host context and are not supported for PCP archives.
Corrupted record in a PCP archive log
PCP archives can become corrupted for a variety of reasons, but the most common is premature termination of pmlogger(1) without flushing its output buffers.
Illegal label record at start of a PCP archive log file
Each physical file in a PCP archive should begin with a common label record. This is a special case of PM_ERR_LOGREC errors.
Empty archive log file
An empty physical file can never be part of a valid PCP archive (at least the label record should be present). This is a special case of PM_ERR_LOGREC errors.
Operation requires context with archive source of metrics
A call to one of the archive variant routines, i.e. pmFetchArchive(3), pmGetInDomArchive(3), pmLookupInDomArchive(3) or pmNameInDomArchive(3), when the current context is not associated with a set of PCP archives.
Metric not defined in the PCP archive log
A PCP client has requested information about a metric, and there is no corresponding information in the set of PCP archives. This should not happen for well-behaved PCP clients.
Instance domain identifier not defined in the PCP archive log
A PCP client has requested information about an instance domain for one or more performance metrics, and there is no corresponding information in the set of PCP archives. If the client is using metric descriptors from the set of archives to identify the instance domain, this is less likely to happen.
Instance identifier not defined in the PCP archive log
A PCP client has requested information about a specific instance of a performance metric, and there is no corresponding information in the set of PCP archives. If the client is using instance names from the instance domain in the set of archives (rather than hard-coded instance names) and instance identifiers from the results returned by pmFetch(3) or pmFetchArchive(3) this is less likely to happen.
Archives overlap in time
When using a context associated with a set of archives, the archives in the set may not overlap in time.
Archives differ by host
When using a context associated with a set of archives, the archives in the set must all have been generated on the same host.
Archives differ by time zone
When using a context associated with a set of archives, the archives in the set must all have been generated using the same time zone.
The type of a metric differs among archives
When using a context associated with a set of archives, the type of each metric must be same in all of the archives.
The semantics of a metric differs among archives
When using a context associated with a set of archives, the semantics of each metric must be same in all of the archives.
The instance domain of a metric differs among archives
When using a context associated with a set of archives, the instance domain of each metric must be same in all of the archives.
The units of a metric differs among archives
When using a context associated with a set of archives, the units of each metric must be same in all of the archives.
TIME CONTROL ERRORSThese errors may occur in the interactions between a GUI PCP client and the time control services provided by pmtime(1).
A PCP client application called pmTimeConnect(3) when already connected to a pmtime(1) instance.
A PCP client application called one of the time control routines pmTime*(3) when not currently connected to any pmtime(1) instance.
A non-null port name is required
If a shared pmtime(1) instance is being created the port argument to pmTimeConnect(3) must not be invalid.
NAMESPACE ERRORSThese errors may occur in the processing of PCP namespace operations. A PCP namespace, see pmns(5), provides the external names and the internal identifiers for the available performance metrics.
Metric name is not a leaf in PMNS
The metric name passed to pmLookupName(3) names a non-terminal path in the namespace, i.e. a group of metrics rather than a single metric.
Attempt to reload the PMNS
When using an explicit local namespace, it is illegal to call either of pmLoadNameSpace(3) or pmLoadASCIINameSpace(3) more than once.
Problems parsing PMNS definitions
Only occurs when an ASCII namespace is explicitly loaded.
PMNS not accessible
Only occurs when an ASCII namespace is explicitly loaded.
PMCD-PMDA ERRORSThese error codes are used in the interactions between pmcd(1) and the PMDAs that provide the performance data.
PMDA is not yet ready to respond to requests
Some PMDAs have long initialization or reset times, and will respond to pmcd(1) with this error if they are busy at the moment. This error translates to PM_ERR_AGAIN for the PCP client who made the request to pmcd which caused the initial request to the PMDA. At some later time the PMDA will inform pmcd (see PM_ERR_PMDAREADY) that it is now ready to process requests, and client requests will begin to succeed.
PMDA is now responsive to requests
Used by PMDAs to asynchronously inform pmcd(1) that they are now willing to resume processing requests. See also PM_ERR_PMDANOTREADY.