1PMDATRACE(3) Library Functions Manual PMDATRACE(3)
2
6 pmtracebegin, pmtraceend, pmtraceabort, pmtracepoint, pmtraceobs,
7 pmtracecounter, pmtracestate, pmtraceerrstr - application-level perfor‐
8 mance instrumentation services
9
11 #include <pcp/trace.h>
12 int pmtracebegin(const char *tag);
14 int pmtraceend(const char *tag);
15 int pmtraceabort(const char *tag);
16 int pmtracepoint(const char *tag);
17 int pmtraceobs(const char *tag, double value);
18 int pmtracecounter(const char *tag, double value);
19 char *pmtraceerrstr(int code);
20 int pmtracestate(int flags);
21 cc ... -lpcp_trace
23
25 character*(*) tag
26 integer code
27 integer flags
28 integer state
29 character*(*) estr
30 real*8 value
31 code = pmtracebegin(tag)
33 code = pmtraceend(tag)
34 code = pmtraceabort(tag)
35 code = pmtracepoint(tag)
36 code = pmtraceobs(tag, value)
37 code = pmtracecounter(tag, value)
38 pmtraceerrstr(code, estr)
39 state = pmtracestate(flags)
40 f77 ... -lpcp_trace or f90 ... -lpcp_trace
42
44 import sgi.pcp.trace;
45 int trace.pmtracebegin(String tag);
47 int trace.pmtraceend(String tag);
48 int trace.pmtraceabort(String tag);
49 int trace.pmtracepoint(String tag);
50 int trace.pmtraceobs(String tag, double value);
51 int trace.pmtracecounter(String tag, double value);
52 String trace.pmtraceerrstr(int code);
53 int trace.pmtracestate(int flags);
54
57 The pcp_trace library functions provide a mechanism for identifying
58 sections of a program as transactions or events for use by the trace
59 Performance Metrics Domain Agent (refer to pmdatrace(1) and PMDA(3)).
60 The monitoring of transactions using the Performance Co-Pilot (PCP) in‐
62 frastructure is initiated through a call to pmtracebegin. Time will be
63 recorded from the end of each pmtracebegin call to the start of the
64 following call to pmtraceend, where the same tag identifier is used in
65 both calls. Following from this, no visible recording will occur until
66 at least one call to pmtraceend is made referencing a tag previously
67 used in a call to pmtracebegin.
68 A transaction which is currently in progress can be cancelled by call‐
70 ing pmtraceabort. No transaction data gathered for that particular
71 transaction will be exported, although data from previous and subse‐
72 quent successful transactions with that tag name are still exported.
73 This is most useful when an error condition arises during transaction
74 processing and the transaction does not run to completion.
75 The tag argument to pmtracebegin, pmtraceend and pmtraceabort is used
77 to uniquely identify each transaction within the pcp_trace library and
78 later by the trace PMDA as the instance domain identifiers for the
79 transaction performance metrics which it exports. These routines are
80 most useful when used around blocks of code which are likely to be exe‐
81 cuted a number of times over some relatively long time period (in a
82 daemon process, for example).
83 pmtracebegin has two distinct roles - firstly as the initiator of a new
85 transaction, and secondly as a mechanism for setting a new start time.
86 Similarly, pmtraceend is used to register a new tag and its initial
87 state with the trace PMDA, or alternatively to update the statistics
88 which the PMDA currently associates with the given tag.
89 A second form of program instrumentation can be obtained from pmtrace‐
91 point. This is a simpler form of monitoring as it exports only the
92 number of times that a particular point in a program has been passed.
93 This differs to the transaction monitoring offered by pmtracebegin and
94 pmtraceend, which exports a running count of successful transaction
95 completions as well as statistics on the time interval between the
96 start and end points of each transaction. This function is most useful
97 when start and end points are not well defined. Examples of this would
98 be when the code branches in such a way that a transaction cannot be
99 clearly identified, or when processing does not follow a transactional
100 model, or the desired instrumentation is akin to event rates rather
101 than event service times.
102 The pmtraceobs and pmtracecounter functions have similar semantics to
104 pmtracepoint, but also allow an arbitrary numeric value to be passed to
105 the trace PMDA. The most recent value for each tag is then immediately
106 available from the PMDA. The only difference between pmtraceobs and
107 pmtracecounter is that the value exported via pmtracecounter is assumed
108 to be a monotonically increasing counter value (e.g. the number of
109 bytes read from a socket), whereas the value exported via pmtraceobs
110 can be any value at all.
111 pmtracestate allows the application to set state flags which are hon‐
113 oured by subsequent calls to the pcp_trace library routines. There are
114 currently two types of flag - debugging flags and the asynchronous pro‐
115 tocol flag. A single call may specify a number of flags together, com‐
116 bined using a (bitwise) logical OR operation, and overrides the previ‐
117 ous state setting.
118 The debugging flags to pmtracestate cause pcp_trace to print diagnostic
120 messages on the standard output stream at important processing points.
121 The default protocol used between the trace PMDA and individual
122 pcp_trace client applications is a synchronous protocol, which allows
123 for dropped connections to be reestablished at a later stage should
124 this become possible. An asynchronous protocol is also available which
125 does not provide the reconnection capability, but which does away with
126 much of the overhead inherent in synchronous communication. This be‐
127 haviour can be toggled using the pmtracestate call, but must be called
128 before other calls to the library. This differs to the debugging state
129 behaviour, which can be altered at any time. pmtracestate returns the
130 previous state (setting prior to being called).
131 The following table describes each of the pmtracestate flags - examples
133 of the use of these flags in each supported language are given in the
134 demo applications (refer to the ``FILES'' section below).
135 ┌────────────┬───────────────────────────────────────────────┐
137 │State Flags │ Semantics │
138 ├────────────┼───────────────────────────────────────────────┤
139 │0 NONE │ Synchronous PDUs and no diagnostics (default) │
140 │1 API │ Shows processing just below the API (debug) │
141 │2 COMMS │ Shows network-related activity (debug) │
142 │4 PDU │ Shows app<->PMDA IPC traffic (debug) │
143 │8 PDUBUF │ Shows internal IPC buffer management (debug) │
144 │16 NOAGENT │ No PMDA communications at all (debug) │
145 │32 ASYNC │ Use the asynchronous PDU protocol (control) │
146 └────────────┴───────────────────────────────────────────────┘
147 Should any of the pcp_trace library functions return a negative value,
148 an error has occured. This can be diagnosed further using the
149 pmtraceerrstr routine, which takes the negative return value as its
150 code argument, and in the C-callable interface returns a pointer to the
151 associated error message. This points into a static error table, and
152 should therefore not be passed to free(3). The Fortran-callable inter‐
153 face has a slightly different syntax, requiring the destination charac‐
154 ter array to be passed in as the second argument. The Java-callable
155 interface returns a UTF-8 string, created using the JNI (Java Native
156 Interface) routine NewStringUTF.
157
159 The pcp_trace routines communicate with the trace PMDA via a socket
160 connection, which by default uses TCP/IP port number 4323. This can be
161 over-ridden by setting PCP_TRACE_PORT to a different port number when
162 the application is started. The host where the trace PMDA is running
163 is by default the localhost, but this can be changed using
164 PCP_TRACE_HOST. When attempting to connect to a remote trace PMDA,
165 after some specified time interval has elapsed, the connection attempt
166 will be aborted and an error status will be returned. The default
167 timeout interval is 3 seconds, and this can be modified by setting
168 PCP_TRACE_TIMEOUT in the environment to a real number of seconds for
169 the desired timeout. This is most useful in cases where the remote
170 host is at the end of a slow network, requiring longer latencies to
171 establish the connection correctly.
172
174 The pcp_trace Java class interface has been developed and verified
175 using version 1.1 of the Java Native Interface (JNI) specification.
176
178 $PCP_DEMOS_DIR/trace/*.c
179 Sample C programs and source for pmtrace(1). Use make(1) to
180 build these programs.
181 $PCP_DEMOS_DIR/trace/fapp1.f
183 Sample Fortran program. Call `make fortran77' or `make for‐
184 tran90' to build this program.
185 $PCP_DEMOS_DIR/trace/japp1.java
187 Sample Java program. `make java' builds the java class file.
188 /usr/java/classes/sgi/pcp/trace.java
190 Java trace class definition.
191
193 Environment variables with the prefix PCP_ are used to parameterize the
194 file and directory names used by PCP. On each installation, the file
195 /etc/pcp.conf contains the local values for these variables. The
196 $PCP_CONF variable may be used to specify an alternative configuration
197 file, as described in pcp.conf(4). Values for these variables may be
198 obtained programatically using the pmGetConfig(3) function.
199
201 file:$PCP_DOC_DIR/Tutorial/trace.html, pcp.man.tutorial, Provided the,
202 make(1), pmcd(1), pmdatrace(1), pmprobe(1), pmtrace(1), Relevant infor‐
203 mation is also available from the on-line PCP Tutorial, subsystem from
204 the PCP images has been installed, access the URL and from your web
205 browser.
206
208 A negative return value from a pcp_trace function indicates that an
209 error has occured - if this is the case, the return value can be passed
210 to pmtraceerrstr to obtain the associated error message.
211 Success is indicated by a return value of zero.
213 pmtracestate also returns an integer representing the state flags which
215 were set prior to the call.
216
218 Applications that use gethostbyname(3N) should exercise caution because
219 the static fields in struct hostent may not be preserved across some
220 pcp_trace calls. In particular, pmtracebegin, pmtraceend, pmtrace‐
221 point, pmtracecounter, and pmtraceobs may all call gethostbyname(3N)
222 internally.
223Performance Co-Pilot SGI PMDATRACE(3)