pmgenmap - generate C code to simplify handling of performance metrics
pmgenmap [
infile]
Given one or more lists of metric names in
infile or on standard input,
pmgenmap generates C declarations and
cpp(1) macros suitable for
use across the Performance Metrics Programming Interface (PMAPI) on standard
output.
The declarations produced by
pmgenmap simplify the coding for client
applications using the PMAPI.
The input should consist of one or more lists of metric names of the form
listname {
metricname1 symbolname1
metricname2 symbolname2
...
}
which will generate C and
cpp(1) declarations of the form
char *listname[] = {
#define symbolname1 0
"metricname1",
#define symbolname2 1
"metricname2",
...
};
The array declarations produced are suitable as parameters to
pmLookupName(3) and the
#defined constants may be used to index
the
vsets in the
pmResult structure returned by a
pmFetch(3) call.
Obviously,
listname must conform to the C identifier naming rules, each
symbolname must conform to the
cpp(1) macro naming rules, and
each
metricname is expected to be a valid performance metrics name (see
PMNS(5) for more details).
The input may include
sh-style comment lines, i.e. with a `
#' as
the first non-blank character of a line, and these are translated on output to
either single line or multi-line C comments in the K&R style. For example,
the input:
# leading block of multi-line comments
# initialization group
foo {
a.b.c ONE
d.e.f.g TWO
# embedded block of multi-lines
# comments and boring pad text
xx.yy.zz THREE
}
# trailing single line comment
Produces the output:
/*
* leading block of multi-line comments
* initialization group
*/
char *foo[] = {
#define ONE 0
"a.b.c",
#define TWO 1
"d.e.f.g",
/*
* embedded block of multi-lines
* comments and boring pad text
*/
#define THREE 2
"xx.yy.zz",
};
/* trailing single line comment */
For brevity we have removed the error handling code, and assumed the chosen
metrics do not have multiple values.
The input file
mystats {
kernel.percpu.cpu.idle IDLE
kernel.percpu.cpu.sys SYS
kernel.percpu.cpu.user USER
hinv.ncpu NCPU
}
produces the following C code, suitable for
#include-ing
/*
* Performance Metrics Name Space Map
* Built by pmgenmap from the file
* mystats.metrics
* on Wed Dec 28 19:44:17 EST 1994
*
* Do not edit this file!
*/
char *mystats[] = {
#define IDLE 0
"kernel.percpu.cpu.idle",
#define SYS 1
"kernel.percpu.cpu.sys",
#define USER 2
"kernel.percpu.cpu.user",
#define NCPU 3
"hinv.ncpu",
};
Using the code generated by
pmgenmap, we are now able to easily obtain
metrics from the Performance Metrics Collection Subsystem (PMCS) as follows:
#define MAX_PMID 4
int trip = 0;
int numpmid = sizeof(mystats)/sizeof(mystats[0]);
double duration;
pmResult *resp;
pmResult *prev;
pmID pmidlist[MAX_PMID];
pmNewContext(PM_CONTEXT_HOST, "localhost");
pmLookupName(numpmid, mystats, pmidlist);
pmFetch(numpmid, pmidlist, &resp);
printf("%d CPUs: %d usr %d sys %d idle0,
resp->vset[NCPU]->vlist[0].value.lval,
resp->vset[USER]->vlist[0].value.lval,
resp->vset[SYS]->vlist[0].value.lval,
resp->vset[IDLE]->vlist[0].value.lval);
Some calls to ensure portability have been removed from the code above for the
sake of clarity - the example above should not be used as a template for
programming. In particular, the raw values of the metrics were used when
pmLookupDesc(3) should have been called to determine the semantics of
each metric.
More complete examples that demonstrate the use of
pmgenmap which may be
used as a basis for program development are included in the PCP demos, e.g.
$PCP_DEMOS_DIR/pmclient.
- $PCP_VAR_DIR/pmns/*
- default PMNS specification files
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).
cpp(1),
PMAPI(3),
pmFetch(3),
pmLookupName(3),
pmNewContext(3),
pcp.conf(5),
pcp.env(5) and
PMNS(5).