| PMLOOKUPLABELS(3) | Library Functions Manual | PMLOOKUPLABELS(3) |
pmLookupLabels, pmGetInstancesLabels, pmGetItemLabels, pmGetClusterLabels, pmGetInDomLabels, pmGetDomainLabels, pmGetContextLabels - retrieve labels associated with performance metric values
#include <pcp/pmapi.h>
int pmLookupLabels(pmID pmid, pmLabelSet **labelsets); int pmGetInstancesLabels(pmInDom indom, pmLabelSet **labelsets);
int pmGetItemLabels(pmID pmid, pmLabelSet **labelsets);
int pmGetClusterLabels(pmID pmid, pmLabelSet **labelsets);
int pmGetInDomLabels(pmInDom indom, pmLabelSet **labelsets);
int pmGetDomainLabels(int domain, pmLabelSet **labelsets);
int pmGetContextLabels(pmLabelSet **labelsets); cc ... -lpcp
from pcp import pmapi
labelsets = pmapi.pmContext().pmLookupLabels(pmid) labelsets = pmapi.pmContext().pmGetInstancesLabels(indom)
labelsets = pmapi.pmContext().pmGetItemLabels(pmid)
labelsets = pmapi.pmContext().pmGetClusterLabels(pmid)
labelsets = pmapi.pmContext().pmGetInDomLabels(indom)
labelsets = pmapi.pmContext().pmGetDomainLabels(domain)
labelsets = pmapi.pmContext().pmGetContextLabels()
Labels are name:value pairs associated with performance metric values for the purpose of attaching additional metric metadata to values. This metadata is less structured and exists separately to the metric descriptor available for every PCP metric from pmLookupDesc(3).
Much like the metric descriptor metadata, labels are an integral part of the identity of each metric, and should rarely, if ever, change.
The pmLookupLabels routine is a convenience interface providing retrieval for all labels associated with a single performance metric identifier, pmid, except labels at the instances level. Labels at the instances level must be retrieved separately with a call to pmGetInstancesLabels because different metric instances may have labels with the same label name. The pmLookupLabels function performs no caching of labels internally.
For efficiency in communication and storage within the various components of the PMCS (Performance Metrics Collection System), labels are maintained using a hierarchy. The set of labels associated with any individual metric value consists of the union of labels from each of these sets of labels:
These independent labelsets can be merged using pmMergeLabelSets(3) to form the complete set of all labels associated with a given value. Note that the label sets returned by pmGetInstancesLabels can be traversed but should not be merged because the label names are unlikely to be unique for different instances of the given indom.
Labels are stored and communicated within PCP using JSONB format. This format is a restricted form of JSON suitable for indexing and other operations. In JSONB form, insignificant whitespace is discarded, and the order of label names is not preserved. Within the PMCS a lexicographically sorted key space is always maintained, however. Duplicate label names are not permitted. The label with highest precedence is the only one presented. If duplicate names are presented at the same hierarchy level, only one will be preserved (exactly which one wins is arbitrary, so do not rely on this).
All name:value pair(s) present will be converted to JSONB form and merged with the existing set of labels for the requested entity (context, domain, indom, metric or instance).
The label names are further constrained to the same set of rules defined for PMNS subtree names.
Each component in a label name must begin with an alphabetic character, and be followed by zero or more characters drawn from the alphabetics, the digits and the underscore (``_'') character. For alphabetic characters in a name, upper and lower case are distinguished.
The value of a label offers significantly more freedom, and may be any valid value as defined by the JSON (http://json.org) specification. Redundant whitespace is always removed within the PMCS.
The complete set of labels associated with any metric value is built from several sources and duplicate label names may exist at any point in the source hierarchy. However, when evaluating the label set (merging labels from the different sources) the JSONB concept of only presenting unique labels is used. It is therefore important to define precedence rules in order that a deterministic set of uniquely named labels can be defined.
As a rule of thumb, the labels closest to PMNS leaf nodes and metric values take precedence:
The primary output from pmLookupLabels is returned in the argument labelset as an array, using the following component data structures;
struct {
uint name : 16; /* label name offset in JSONB string */
uint namelen : 8; /* length of name excluding the null */
uint flags : 8; /* information about this label */
uint value : 16; /* offset of the label value */
uint valuelen : 16; /* length of value in bytes */ } pmLabel; struct {
uint inst; /* PM_IN_NULL or the instance ID */
int nlabels; /* count of labels or error code */
char *json; /* JSON formatted labels string */
uint jsonlen : 16; /* JSON string length byte count */
uint padding : 16; /* zero, reserved for future use */
pmLabel *labels; /* indexing into the JSON string */ } pmLabelSet;
The pmLabel provides information about an individual label. This includes the offsets to the start of its name and value in the json string of a pmLabelSet, their respective lengths, and also any informative flags associated with the label (describing where it lies in the hierarchy of labels, and whether it is an intrinsic or extrinsic label).
Building on this, the pmLabelSet provides information about the set of labels associated with an entity (context, domain, indom, metric cluster, item or instance). The entity will be from any one level of the label hierarchy. If at the lowest hierarchy level (which happens to be highest precedence - PM_LABEL_INSTANCES) then the inst field will contain an actual instance identifier instead of PM_IN_NULL.
The nlabels field describes the number of labels (name:value pairs) that can be found in both the accompanying json string (which is JSONB format - no unnecessary whitespace and with no duplicate label names) and the accompanying labels array (which has nlabels elements).
Consider a deployment with global labels (assume $PCP_SYSCONF_DIR is set to its usual location of /etc/pcp) as follows:
$ cat /etc/pcp/labels/*
{
"tier": "production",
"datacenter": "hkg",
"services": ["indexer","database"]
}
Use pminfo to form the merged labelsets for several pmdasample(1) metrics as follows:
$ pminfo -m -f --labels sample.rapid sample.colour sample.mirage
sample.rapid PMID: 30.0.64
value 800000000
labels {"agent":"sample","datacenter":"sydney","hostname":"acme.com","measure":"speed","role":"testing","services":["indexer","database"],"tier":"production","units":"metres per second","unitsystem":"SI"}
sample.colour PMID: 30.0.5
inst [0 or "red"] value 101
inst [1 or "green"] value 202
inst [2 or "blue"] value 303
inst [0 or "red"] labels {"agent":"sample","datacenter":"syd","hostname":"acme.com","model":"RGB","role":"testing","services":["indexer","database"],"tier":"production"}
inst [1 or "green"] labels {"agent":"sample","datacenter":"syd","hostname":"acme.com","model":"RGB","role":"testing","services":["indexer","database"],"tier":"production"}
inst [2 or "blue"] labels {"agent":"sample","datacenter":"syd","hostname":"acme.com","model":"RGB","role":"testing","services":["indexer","database"],"tier":"production"}
sample.mirage PMID: 29.0.37
inst [0 or "m-00"] value 99
inst [0 or "m-00"] labels {"agent":"sample","datacenter":"sydney","hostname":"acme.com","role":"testing","services":["indexer","database"],"tier":"production","transient":false}
Here, pminfo has merged the separate sets of labels returned from pmGetContextLabels (names: datacenter, hostname, services, tier), pmGetDomainLabels (names: role, agent), pmGetInDomLabels (names: model), pmGetItemLabels (names: units, unitsystem) and pmGetInstancesLabels (names: transient) to form the complete set for each of the metrics.
#!/usr/bin/env pmpython import sys from pcp import pmapi import cpmapi as c_api ctx = pmapi.pmContext(c_api.PM_CONTEXT_HOST, "local:") for metric in sys.argv[1:]:
pmid = ctx.pmLookupName(metric)[0]
desc = ctx.pmLookupDescs(pmid)[0]
print("== label sets for %s ==" % metric)
labelSetList = ctx.pmLookupLabels(pmid)
# class pmLabelSet has a __str__ handler
for labelSet in labelSetList:
print("%s" % labelSet)
ctx.pmFreeLabelSets(labelSetList)
if desc.contents.indom != c_api.PM_INDOM_NULL:
print("== instances label sets for %s ==" % metric)
labelSetList = ctx.pmGetInstancesLabels(desc.contents.indom)
for labelSet in labelSetList:
print("%s" % labelSet)
ctx.pmFreeLabelSets(labelSetList)
On success these interfaces all return the number of elements in the labelsets array. associated with performance metrics. The memory associated with labelsets should be released using pmFreeLabelSets(3) when no longer needed.
Only in the case of pmLookupLabels will the resulting labelset be a merged set of labels from all hierarchy levels (except at the instances level, as described above).
For the other routines, except for pmGetInstancesLabels, if no labels exist at all for the requested hierarchy level the return code will be zero and no space will have been allocated.
In the case of pmGetInstancesLabels, which can return multiple elements in its labelsets result (one set of labels for each instance), the nlabels field may be either zero indicating no labels for that instance, or a positive count of labels, or a negative PMAPI error code.
Note that it is mandatory for a call to pmGetInstancesLabels to be preceded by a call to pmGetInDom(3) to ensure the instances have been resolved within the PMDA.
If no result can be obtained, e.g. due to IPC failure using the current PMAPI context then pmGetInstancesLabels will return a negative error code which may be examined using
A successful return from the Python API always provides the labelset in the form of a list, for all labels functions. On error a pmErr exception is raised containing the error code and diagnostic. pmErrStr(3).
pmcd(1), PMAPI(3), pmFetch(3), pmGetInDom(3), pmLookupDesc(3), pmLookupName(3), pmFreeLabelSets(3), pmMergeLabelSets(3) and pmNewContext(3).
| PCP | Performance Co-Pilot |