1PMAPI(3) Library Functions Manual PMAPI(3)
2
6 PMAPI - introduction to the Performance Metrics Application Programming
7 Interface
8
10 #include <pcp/pmapi.h>
11 ... assorted routines ...
13 cc ... -lpcp
15
17 Within the framework of the Performance Co-Pilot (PCP), client applica‐
18 tions are developed using the Performance Metrics Application Program‐
19 ming Interface (PMAPI) that defines a procedural interface with ser‐
20 vices suited to the development of applications with a particular in‐
21 terest in performance metrics.
22 This description presents an overview of the PMAPI and the context in
24 which PMAPI applications are run. The PMAPI is more fully described in
25 the Performance Co-Pilot Programmer's Guide, and the manual pages for
26 the individual PMAPI routines.
27
29 For a description of the Performance Metrics Name Space (PMNS) and as‐
30 sociated terms and concepts, see PCPIntro(1).
31 Not all PMIDs need be represented in the PMNS of every application.
33 For example, an application which monitors disk traffic will likely use
34 a name space which references only the PMIDs for I/O statistics.
35 Applications which use the PMAPI may have independent versions of a
37 PMNS, constructed from an initialization file when the application
38 starts; see pmLoadASCIINameSpace(3), pmLoadNameSpace(3), and PMNS(5).
39 Internally (below the PMAPI) the implementation of the Performance Met‐
41 rics Collection System (PMCS) uses only the PMIDs, and a PMNS provides
42 an external mapping from a hierarchic taxonomy of names to PMIDs that
43 is convenient in the context of a particular system or particular use
44 of the PMAPI. For the applications programmer, the routines pmLookup‐
45 Name(3) and pmNameID(3) translate between names in a PMNS and PMIDs,
46 and vice versa. The PMNS may be traversed using pmGetChildren(3)
47 andpmTraversePMNS. The pmFetchGroup(3) functions combine metric name
48 lookup, fetch, and conversion operations.
49
51 An application using the PMAPI may manipulate several concurrent con‐
52 texts, each associated with a source of performance metrics, e.g.
53 pmcd(1) on some host, or a set of archives of performance metrics as
54 created by pmlogger(1).
55 Contexts are identified by a ``handle'', a small integer value that is
57 returned when the context is created; see pmNewContext(3) and pmDupCon‐
58 text(3). Some PMAPI functions require an explicit ``handle'' to iden‐
59 tify the correct context, but more commonly the PMAPI function is exe‐
60 cuted in the ``current'' context. The current context may be discov‐
61 ered using pmWhichContext(3) and changed using pmUseContext(3).
62 If a PMAPI context has not been explicitly established (or the previous
64 current context has been closed using pmDestroyContext(3)) then the
65 current PMAPI context is undefined.
66 In addition to the source of the performance metrics, the context also
68 includes the instance profile and collection time (both described be‐
69 low) which controls how much information is returned, and when the in‐
70 formation was collected.
71
73 When performance metric values are returned across the PMAPI to a re‐
74 questing application, there may be more than one value for a particular
75 metric. Multiple values, or instances, for a single metric are typi‐
76 cally the result of instrumentation being implemented for each instance
77 of a set of similar components or services in a system, e.g. indepen‐
78 dent counts for each CPU, or each process, or each disk, or each system
79 call type, etc. This multiplicity of values is not enumerated in the
80 name space but rather, when performance metrics are delivered across
81 the PMAPI by pmFetch(3), the format of the result accommodates values
82 for one or more instances, with an instance-value pair encoding the
83 metric value for a particular instance.
84 The instances are identified by an internal identifier assigned by the
86 agent responsible for instantiating the values for the associated per‐
87 formance metric. Each instance identifier has a corresponding external
88 instance identifier name (an ASCII string). The routines
89 pmGetInDom(3), pmLookupInDom(3) and pmNameInDom(3) may be used to enu‐
90 merate all instance identifiers, and to translate between internal and
91 external instance identifiers.
92 All of the instance identifiers for a particular performance metric are
94 collectively known as an instance domain. Multiple performance metrics
95 may share the same instance domain.
96 If only one instance is ever available for a particular performance
98 metric, the instance identifier in the result from pmFetch(3) assumes
99 the special value PM_IN_NULL and may be ignored by the application, and
100 only one instance-value pair appears in the result for that metric.
101 Under these circumstances, the associated instance domain (as returned
102 via pmLookupDesc(3)) is set to PM_INDOM_NULL to indicate that values
103 for this metric are singular.
104 The difficult issue of transient performance metrics (e.g. per-filesys‐
106 tem information, hot-plug replaceable hardware modules, etc.) means
107 that repeated requests for the same PMID may return different numbers
108 of values, and/or some changes in the particular instance identifiers
109 returned. This means applications need to be aware that metric instan‐
110 tiation is guaranteed to be valid at the time of collection only. Sim‐
111 ilar rules apply to the transient semantics of the associated metric
112 values. In general however, it is expected that the bulk of the per‐
113 formance metrics will have instantiation semantics that are fixed over
114 the execution life-time of any PMAPI client.
115
117 The PMAPI supports a wide range of format and type encodings for the
118 values of performance metrics, namely signed and unsigned integers,
119 floating point numbers, 32-bit and 64-bit encodings of all of the
120 above, ASCII strings (C-style, NULL byte terminated), and arbitrary ag‐
121 gregates of binary data.
122 The type field in the pmDesc structure returned by pmLookupDesc(3)
124 identifies the format and type of the values for a particular perfor‐
125 mance metric within a particular PMAPI context.
126 Note that the encoding of values for a particular performance metric
128 may be different for different PMAPI contexts, due to differences in
129 the underlying implementation for different contexts. However it is
130 expected that the vast majority of performance metrics will have con‐
131 sistent value encoding across all versions of all implementations, and
132 hence across all PMAPI contexts.
133 The PMAPI supports routines to automate the handling of the various
135 value formats and types, particularly for the common case where conver‐
136 sion to a canonical format is desired, see pmExtractValue(3) and pm‐
137 PrintValue(3).
138
140 Independent of how the value is encoded, the value for a performance
141 metric is assumed to be drawn from a set of values that can be de‐
142 scribed in terms of their dimensionality and scale by a compact encod‐
143 ing as follows. The dimensionality is defined by a power, or index, in
144 each of 3 orthogonal dimensions, namely Space, Time and Count (or
145 Events, which are dimensionless). For example I/O throughput might be
146 represented as Space/Time, while the running total of system calls is
147 Count, memory allocation is Space and average service time is
148 Time/Count. In each dimension there are a number of common scale val‐
149 ues that may be used to better encode ranges that might otherwise ex‐
150 haust the precision of a 32-bit value. This information is encoded in
151 the pmUnits structure which is embedded in the pmDesc structure re‐
152 turned from pmLookupDesc(3).
153 The routine pmConvScale(3) is provided to convert values in conjunction
155 with the pmUnits structures that defines the dimensionality and scale
156 of the values for a particular performance metric as returned from pm‐
157 Fetch(3), and the desired dimensionality and scale of the value the
158 PMAPI client wishes to manipulate. Alternatively, the pmFetchGroup(3)
159 functions can perform data format and unit conversion operations, spec‐
160 ified by textual descriptions of desired unit / scales.
161
163 The set of instances for performance metrics returned from a pmFetch(3)
164 call may be filtered or restricted using an instance profile. There is
165 one instance profile for each PMAPI context the application creates,
166 and each instance profile may include instances from one or more in‐
167 stance domains.
168 The routines pmAddProfile(3) and pmDelProfile(3) may be used to dynami‐
170 cally adjust the instance profile.
171
173 For each set of values for performance metrics returned via pmFetch(3)
174 there is an associated ``timestamp'' that serves to identify when the
175 performance metric values were collected; for metrics being delivered
176 from a real-time source (i.e. pmcd(1) on some host) this would typi‐
177 cally be not long before they were exported across the PMAPI, and for
178 metrics being delivered from a set of archives, this would be the time
179 when the metrics were written into the archive.
180 There is an issue here of exactly when individual metrics may have been
182 collected, especially given their origin in potentially different Per‐
183 formance Metric Domains, and variability in the metric updating fre‐
184 quency at the lowest level of the Performance Metric Domain. The PMCS
185 opts for the pragmatic approach, in which the PMAPI implementation un‐
186 dertakes to return all of the metrics with values accurate as of the
187 timestamp, to the best of our ability. The belief is that the inaccu‐
188 racy this introduces is small, and the additional burden of accurate
189 individual timestamping for each returned metric value is neither war‐
190 ranted nor practical (from an implementation viewpoint).
191 Of course, in the case of collection of metrics from multiple hosts the
193 PMAPI client must assume the sanity of the timestamps is constrained by
194 the extent to which clock synchronization protocols are implemented
195 across the network.
196 A PMAPI application may call pmSetMode(3) to vary the requested collec‐
198 tion time, e.g. to rescan performance metrics values from the recent
199 past, or to ``fast-forward'' through a set of archives.
200
202 Across the PMAPI, all arguments and results involving a ``list of some‐
203 thing'' are declared to be arrays with an associated argument or func‐
204 tion value to identify the number of elements in the list. This has
205 been done to avoid both the varargs(3) approach and sentinel-terminated
206 lists.
207 Where the size of a result is known at the time of a call, it is the
209 caller's responsibility to allocate (and possibly free) the storage,
210 and the called function will assume the result argument is of an appro‐
211 priate size. Where a result is of variable size and that size cannot
212 be known in advance (e.g. for pmGetChildren(3), pmGetInDom(3), pm‐
213 NameInDom(3), pmNameID(3), pmLookupLabels(3), pmLookupText(3) and pm‐
214 Fetch(3)) the PMAPI implementation uses a range of dynamic allocation
215 schemes in the called routine, with the caller responsible for subse‐
216 quently releasing the storage when no longer required. In some cases
217 this simply involves calls to free(3), but in others (most notably for
218 the result from pmFetch(3)), special routines (e.g. pmFreeResult(3) and
219 pmFreeLabelSets(3)) should be used to release the storage.
220 As a general rule, if the called routine returns an error status then
222 no allocation will have been done, and any pointer to a variable sized
223 result is undefined.
224
226 Where error conditions may arise, the functions that comprise the PMAPI
227 conform to a single, simple error notification scheme, as follows;
228 + the function returns an integer
230 + values >= 0 indicate no error, and perhaps some positive status,
232 e.g. the number of things really processed
233 + values < 0 indicate an error, with a global table of error condi‐
235 tions and error messages
236 The PMAPI routine pmErrStr(3) translates error conditions into error
238 messages. By convention, the small negative values are assumed to be
239 negated versions of the Unix error codes as defined in <errno.h> and
240 the strings returned are as per strerror(3). The larger, negative er‐
241 ror codes are PMAPI error conditions.
242 One error, common to all PMAPI routines that interact with pmcd(1) on
244 some host is PM_ERR_IPC, which indicates the communication link to
245 pmcd(1) has been lost.
246
248 The original design for PCP was based around single-threaded applica‐
249 tions, or more strictly applications in which only one thread was ever
250 expected to call the PCP libraries. This restriction has been relaxed
251 for libpcp to allow the most common PMAPI routines to be safely called
252 from any thread in a multi-threaded application.
253 However the following groups of functions and services in libpcp are
255 still restricted to being called from a single-thread, and this is en‐
256 forced by returning PM_ERR_THREAD when an attempt to call the routines
257 in each group from more than one thread is detected.
258 1. Any use of a PM_CONTEXT_LOCAL context, as the DSO PMDAs that are
260 called directly from libpcp may not be thread-safe.
261
263 Most environment variables are described in PCPIntro(1). In addition,
264 environment variables with the prefix PCP_ are used to parameterize the
265 file and directory names used by PCP. On each installation, the file
266 /etc/pcp.conf contains the local values for these variables. The
267 $PCP_CONF variable may be used to specify an alternative configuration
268 file, as described in pcp.conf(5). Values for these variables may be
269 obtained programmatically using the pmGetConfig(3) function.
270
272 PCPIntro(1), PCPIntro(3), PMDA(3), PMWEBAPI(3), pmGetConfig(3),
273 pcp.conf(5), pcp.env(5) and PMNS(5).
274Performance Co-Pilot PCP PMAPI(3)