inline void loop::poll() {
int n_ev;
task::io t;
/* TODO: add task throughput control */
while (que().pop(t)) { t(*this); }
poller::event occur[maxfd()];
if ((n_ev = p().wait(occur, maxfd(), timeout())) < 0) {
if (p().error_again()) { return; }
TRACE("poller::wait: %d", p().error_no());
return;
}
for (int i = 0; i < n_ev; i++) {
read(occur[i]);
}
}
int
reconnect(void)
{
int sts;
int ctx;
__pmContext *ctxp;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: reconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
sts = pmReconnectContext(ctx);
if (sts >= 0) {
time_t now;
time(&now);
fprintf(stderr, "%s: re-established connection to PMCD on \"%s\" at %s\n",
pmProgname, pmcd_host, ctime(&now));
pmcdfd = ctxp->c_pmcd->pc_fd;
__pmFD_SET(pmcdfd, &fds);
numfds = maxfd() + 1;
}
PM_UNLOCK(ctxp->c_lock);
return sts;
}
void
disconnect(int sts)
{
time_t now;
#if CAN_RECONNECT
int ctx;
__pmContext *ctxp;
#endif
time(&now);
if (sts != 0)
fprintf(stderr, "%s: Error: %s\n", pmProgname, pmErrStr(sts));
fprintf(stderr, "%s: Lost connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
#if CAN_RECONNECT
if (primary) {
fprintf(stderr, "This is fatal for the primary logger.");
exit(1);
}
close(pmcdfd);
__pmFD_CLR(pmcdfd, &fds);
pmcdfd = -1;
numfds = maxfd() + 1;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: disconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
ctxp->c_pmcd->pc_fd = -1;
PM_UNLOCK(ctxp->c_lock);
#else
exit(1);
#endif
}
void
disconnect(int sts)
{
time_t now;
int ctx;
__pmContext *ctxp;
time(&now);
if (sts != 0)
fprintf(stderr, "%s: Error: %s\n", pmProgname, pmErrStr(sts));
fprintf(stderr, "%s: Lost connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
if (pmcdfd != -1) {
close(pmcdfd);
__pmFD_CLR(pmcdfd, &fds);
pmcdfd = -1;
}
numfds = maxfd() + 1;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: disconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
ctxp->c_pmcd->pc_fd = -1;
PM_UNLOCK(ctxp->c_lock);
}
int
reconnect(void)
{
int sts;
int ctx;
time_t now;
__pmContext *ctxp;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: reconnect botch: cannot get context: %s\n",
pmProgname, pmErrStr(ctx));
exit(1);
}
sts = pmReconnectContext(ctx);
if (sts >= 0) {
pmcdfd = ctxp->c_pmcd->pc_fd;
__pmFD_SET(pmcdfd, &fds);
numfds = maxfd() + 1;
}
PM_UNLOCK(ctxp->c_lock);
if (sts < 0)
return sts;
time(&now);
fprintf(stderr, "%s: re-established connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
/*
* Metrics may have changed while PMCD was unreachable, so we
* need to recheck each metric to make sure that its PMID and
* semantics have not changed. We cannot recover if there is
* an incompatible change - must defer to controlling scripts
* or processes (a new-named archive will have to be created,
* from a new pmlogger process, and pmlogrewrite/pmlogextract
* will need to become involved if they need to be merged).
*/
validate_metrics();
/*
* All metrics have been validated, however, this state change
* represents a potential gap in the stream of metrics. So we
* must store a <mark> record at this point.
*/
if ((sts = putmark()) < 0) {
fprintf(stderr, "putmark: %s\n", pmErrStr(sts));
exit(1);
}
return 0;
}
int
main(int argc, char **argv)
{
int c;
int sts;
int sep = __pmPathSeparator();
int use_localtime = 0;
int isdaemon = 0;
char *pmnsfile = PM_NS_DEFAULT;
char *username;
char *logfile = "pmlogger.log";
/* default log (not archive) file name */
char *endnum;
int i;
task_t *tp;
optcost_t ocp;
__pmFdSet readyfds;
char *p;
char *runtime = NULL;
int ctx; /* handle corresponding to ctxp below */
__pmContext *ctxp; /* pmlogger has just this one context */
int niter;
pid_t target_pid = 0;
__pmGetUsername(&username);
/*
* Warning:
* If any of the pmlogger options change, make sure the
* corresponding changes are made to pmnewlog when pmlogger
* options are passed through from the control file
*/
while ((c = pmgetopt_r(argc, argv, &opts)) != EOF) {
switch (c) {
case 'c': /* config file */
if (access(opts.optarg, F_OK) == 0)
configfile = opts.optarg;
else {
/* does not exist as given, try the standard place */
char *sysconf = pmGetConfig("PCP_VAR_DIR");
int sz = strlen(sysconf)+strlen("/config/pmlogger/")+strlen(opts.optarg)+1;
if ((configfile = (char *)malloc(sz)) == NULL)
__pmNoMem("config file name", sz, PM_FATAL_ERR);
snprintf(configfile, sz,
"%s%c" "config%c" "pmlogger%c" "%s",
sysconf, sep, sep, sep, opts.optarg);
if (access(configfile, F_OK) != 0) {
/* still no good, error handling happens below */
free(configfile);
configfile = opts.optarg;
}
}
break;
case 'D': /* debug flag */
sts = __pmParseDebug(opts.optarg);
if (sts < 0) {
pmprintf("%s: unrecognized debug flag specification (%s)\n",
pmProgname, opts.optarg);
opts.errors++;
}
else
pmDebug |= sts;
break;
case 'h': /* hostname for PMCD to contact */
pmcd_host_conn = opts.optarg;
break;
case 'l': /* log file name */
logfile = opts.optarg;
break;
case 'L': /* linger if not primary logger */
linger = 1;
break;
case 'm': /* note for port map file */
note = opts.optarg;
isdaemon = ((strcmp(note, "pmlogger_check") == 0) ||
(strcmp(note, "pmlogger_daily") == 0));
break;
case 'n': /* alternative name space file */
pmnsfile = opts.optarg;
break;
case 'p':
target_pid = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0') {
pmprintf("%s: invalid process identifier (%s)\n",
pmProgname, opts.optarg);
opts.errors++;
} else if (!__pmProcessExists(target_pid)) {
pmprintf("%s: PID error - no such process (%d)\n",
pmProgname, target_pid);
opts.errors++;
}
break;
case 'P': /* this is the primary pmlogger */
primary = 1;
isdaemon = 1;
break;
case 'r': /* report sizes of pmResult records */
rflag = 1;
break;
case 's': /* exit size */
sts = ParseSize(opts.optarg, &exit_samples, &exit_bytes, &exit_time);
if (sts < 0) {
pmprintf("%s: illegal size argument '%s' for exit size\n",
pmProgname, opts.optarg);
opts.errors++;
}
else if (exit_time.tv_sec > 0) {
__pmAFregister(&exit_time, NULL, run_done_callback);
}
break;
case 'T': /* end time */
runtime = opts.optarg;
break;
case 't': /* change default logging interval */
if (pmParseInterval(opts.optarg, &delta, &p) < 0) {
pmprintf("%s: illegal -t argument\n%s", pmProgname, p);
free(p);
opts.errors++;
}
break;
case 'U': /* run as named user */
username = opts.optarg;
isdaemon = 1;
break;
case 'u': /* flush output buffers after each fetch */
/*
* all archive write I/O is unbuffered now, so maintain -u
* for backwards compatibility only
*/
break;
case 'v': /* volume switch after given size */
sts = ParseSize(opts.optarg, &vol_switch_samples, &vol_switch_bytes,
&vol_switch_time);
if (sts < 0) {
pmprintf("%s: illegal size argument '%s' for volume size\n",
pmProgname, opts.optarg);
opts.errors++;
}
else if (vol_switch_time.tv_sec > 0) {
vol_switch_afid = __pmAFregister(&vol_switch_time, NULL,
vol_switch_callback);
}
break;
case 'V':
archive_version = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0' || archive_version != PM_LOG_VERS02) {
pmprintf("%s: -V requires a version number of %d\n",
pmProgname, PM_LOG_VERS02);
opts.errors++;
}
break;
case 'x': /* recording session control fd */
rsc_fd = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0' || rsc_fd < 0) {
pmprintf("%s: -x requires a non-negative numeric argument\n", pmProgname);
opts.errors++;
}
else {
time(&rsc_start);
}
break;
case 'y':
use_localtime = 1;
break;
case '?':
default:
opts.errors++;
break;
}
}
if (primary && pmcd_host != NULL) {
pmprintf(
"%s: -P and -h are mutually exclusive; use -P only when running\n"
"%s on the same (local) host as the PMCD to which it connects.\n",
pmProgname, pmProgname);
opts.errors++;
}
if (!opts.errors && opts.optind != argc - 1) {
pmprintf("%s: insufficient arguments\n", pmProgname);
opts.errors++;
}
if (opts.errors) {
pmUsageMessage(&opts);
exit(1);
}
if (rsc_fd != -1 && note == NULL) {
/* add default note to indicate running with -x */
static char xnote[10];
snprintf(xnote, sizeof(xnote), "-x %d", rsc_fd);
note = xnote;
}
/* if we are running as a daemon, change user early */
if (isdaemon)
__pmSetProcessIdentity(username);
__pmOpenLog("pmlogger", logfile, stderr, &sts);
if (sts != 1) {
fprintf(stderr, "%s: Warning: log file (%s) creation failed\n", pmProgname, logfile);
/* continue on ... writing to stderr */
}
/* base name for archive is here ... */
archBase = argv[opts.optind];
if (pmcd_host_conn == NULL)
pmcd_host_conn = "local:";
/* initialise access control */
if (__pmAccAddOp(PM_OP_LOG_ADV) < 0 ||
__pmAccAddOp(PM_OP_LOG_MAND) < 0 ||
__pmAccAddOp(PM_OP_LOG_ENQ) < 0) {
fprintf(stderr, "%s: access control initialisation failed\n", pmProgname);
exit(1);
}
if (pmnsfile != PM_NS_DEFAULT) {
if ((sts = pmLoadASCIINameSpace(pmnsfile, 1)) < 0) {
fprintf(stderr, "%s: Cannot load namespace from \"%s\": %s\n", pmProgname, pmnsfile, pmErrStr(sts));
exit(1);
}
}
if ((ctx = pmNewContext(PM_CONTEXT_HOST, pmcd_host_conn)) < 0) {
fprintf(stderr, "%s: Cannot connect to PMCD on host \"%s\": %s\n", pmProgname, pmcd_host_conn, pmErrStr(ctx));
exit(1);
}
pmcd_host = (char *)pmGetContextHostName(ctx);
if (strlen(pmcd_host) == 0) {
fprintf(stderr, "%s: pmGetContextHostName(%d) failed\n",
pmProgname, ctx);
exit(1);
}
if (rsc_fd == -1) {
/* no -x, so register client id with pmcd */
__pmSetClientIdArgv(argc, argv);
}
/*
* discover fd for comms channel to PMCD ...
*/
if ((ctxp = __pmHandleToPtr(ctx)) == NULL) {
fprintf(stderr, "%s: botch: __pmHandleToPtr(%d) returns NULL!\n", pmProgname, ctx);
exit(1);
}
pmcdfd = ctxp->c_pmcd->pc_fd;
PM_UNLOCK(ctxp->c_lock);
if (configfile != NULL) {
if ((yyin = fopen(configfile, "r")) == NULL) {
fprintf(stderr, "%s: Cannot open config file \"%s\": %s\n",
pmProgname, configfile, osstrerror());
exit(1);
}
}
else {
/* **ANY** Lex would read from stdin automagically */
configfile = "<stdin>";
}
__pmOptFetchGetParams(&ocp);
ocp.c_scope = 1;
__pmOptFetchPutParams(&ocp);
/* prevent early timer events ... */
__pmAFblock();
if (yyparse() != 0)
exit(1);
if (configfile != NULL)
fclose(yyin);
yyend();
#ifdef PCP_DEBUG
fprintf(stderr, "Config parsed\n");
#endif
fprintf(stderr, "Starting %slogger for host \"%s\" via \"%s\"\n",
primary ? "primary " : "", pmcd_host, pmcd_host_conn);
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_LOG) {
fprintf(stderr, "optFetch Cost Parameters: pmid=%d indom=%d fetch=%d scope=%d\n",
ocp.c_pmid, ocp.c_indom, ocp.c_fetch, ocp.c_scope);
fprintf(stderr, "\nAfter loading config ...\n");
for (tp = tasklist; tp != NULL; tp = tp->t_next) {
if (tp->t_numvalid == 0)
continue;
fprintf(stderr, " state: %sin log, %savail, %s, %s",
PMLC_GET_INLOG(tp->t_state) ? "" : "not ",
PMLC_GET_AVAIL(tp->t_state) ? "" : "un",
PMLC_GET_MAND(tp->t_state) ? "mand" : "adv",
PMLC_GET_ON(tp->t_state) ? "on" : "off");
fprintf(stderr, " delta: %ld usec",
(long)1000 * tp->t_delta.tv_sec + tp->t_delta.tv_usec);
fprintf(stderr, " numpmid: %d\n", tp->t_numpmid);
for (i = 0; i < tp->t_numpmid; i++) {
fprintf(stderr, " %s (%s):\n", pmIDStr(tp->t_pmidlist[i]), tp->t_namelist[i]);
}
__pmOptFetchDump(stderr, tp->t_fetch);
}
}
#endif
if (!primary && tasklist == NULL && !linger) {
fprintf(stderr, "Nothing to log, and not the primary logger instance ... good-bye\n");
exit(1);
}
if ((sts = __pmLogCreate(pmcd_host, archBase, archive_version, &logctl)) < 0) {
fprintf(stderr, "__pmLogCreate: %s\n", pmErrStr(sts));
exit(1);
}
else {
/*
* try and establish $TZ from the remote PMCD ...
* Note the label record has been set up, but not written yet
*/
char *name = "pmcd.timezone";
pmID pmid;
pmResult *resp;
__pmtimevalNow(&epoch);
sts = pmUseContext(ctx);
if (sts >= 0)
sts = pmLookupName(1, &name, &pmid);
if (sts >= 0)
sts = pmFetch(1, &pmid, &resp);
if (sts >= 0) {
if (resp->vset[0]->numval > 0) { /* pmcd.timezone present */
strcpy(logctl.l_label.ill_tz, resp->vset[0]->vlist[0].value.pval->vbuf);
/* prefer to use remote time to avoid clock drift problems */
epoch = resp->timestamp; /* struct assignment */
if (! use_localtime)
pmNewZone(logctl.l_label.ill_tz);
}
#ifdef PCP_DEBUG
else if (pmDebug & DBG_TRACE_LOG) {
fprintf(stderr,
"main: Could not get timezone from host %s\n",
pmcd_host);
}
#endif
pmFreeResult(resp);
}
}
/* do ParseTimeWindow stuff for -T */
if (runtime) {
struct timeval res_end; /* time window end */
struct timeval start;
struct timeval end;
struct timeval last_delta;
char *err_msg; /* parsing error message */
time_t now;
struct timeval now_tv;
time(&now);
now_tv.tv_sec = now;
now_tv.tv_usec = 0;
start = now_tv;
end.tv_sec = INT_MAX;
end.tv_usec = INT_MAX;
sts = __pmParseTime(runtime, &start, &end, &res_end, &err_msg);
if (sts < 0) {
fprintf(stderr, "%s: illegal -T argument\n%s", pmProgname, err_msg);
exit(1);
}
last_delta = res_end;
tsub(&last_delta, &now_tv);
__pmAFregister(&last_delta, NULL, run_done_callback);
last_stamp = res_end;
}
fprintf(stderr, "Archive basename: %s\n", archBase);
#ifndef IS_MINGW
/* detach yourself from the launching process */
if (isdaemon)
setpgid(getpid(), 0);
#endif
/* set up control port */
init_ports();
__pmFD_ZERO(&fds);
for (i = 0; i < CFD_NUM; ++i) {
if (ctlfds[i] >= 0)
__pmFD_SET(ctlfds[i], &fds);
}
#ifndef IS_MINGW
__pmFD_SET(pmcdfd, &fds);
#endif
if (rsc_fd != -1)
__pmFD_SET(rsc_fd, &fds);
numfds = maxfd() + 1;
if ((sts = do_preamble()) < 0)
fprintf(stderr, "Warning: problem writing archive preamble: %s\n",
pmErrStr(sts));
sts = 0; /* default exit status */
parse_done = 1; /* enable callback processing */
__pmAFunblock();
for ( ; ; ) {
int nready;
#ifdef PCP_DEBUG
if ((pmDebug & DBG_TRACE_APPL2) && (pmDebug & DBG_TRACE_DESPERATE)) {
fprintf(stderr, "before __pmSelectRead(%d,...): run_done_alarm=%d vol_switch_alarm=%d log_alarm=%d\n", numfds, run_done_alarm, vol_switch_alarm, log_alarm);
}
#endif
niter = 0;
while (log_alarm && niter++ < 10) {
__pmAFblock();
log_alarm = 0;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: log_alarm\n");
#endif
for (tp = tasklist; tp != NULL; tp = tp->t_next) {
if (tp->t_alarm) {
tp->t_alarm = 0;
do_work(tp);
}
}
__pmAFunblock();
}
if (vol_switch_alarm) {
__pmAFblock();
vol_switch_alarm = 0;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: vol_switch_alarm\n");
#endif
newvolume(VOL_SW_TIME);
__pmAFunblock();
}
if (run_done_alarm) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: run_done_alarm\n");
#endif
run_done(0, NULL);
/*NOTREACHED*/
}
__pmFD_COPY(&readyfds, &fds);
nready = __pmSelectRead(numfds, &readyfds, NULL);
#ifdef PCP_DEBUG
if ((pmDebug & DBG_TRACE_APPL2) && (pmDebug & DBG_TRACE_DESPERATE)) {
fprintf(stderr, "__pmSelectRead(%d,...) done: nready=%d run_done_alarm=%d vol_switch_alarm=%d log_alarm=%d\n", numfds, nready, run_done_alarm, vol_switch_alarm, log_alarm);
}
#endif
__pmAFblock();
if (nready > 0) {
/* handle request on control port */
for (i = 0; i < CFD_NUM; ++i) {
if (ctlfds[i] >= 0 && __pmFD_ISSET(ctlfds[i], &readyfds)) {
if (control_req(ctlfds[i])) {
/* new client has connected */
__pmFD_SET(clientfd, &fds);
if (clientfd >= numfds)
numfds = clientfd + 1;
}
}
}
if (clientfd >= 0 && __pmFD_ISSET(clientfd, &readyfds)) {
/* process request from client, save clientfd in case client
* closes connection, resetting clientfd to -1
*/
int fd = clientfd;
if (client_req()) {
/* client closed connection */
__pmFD_CLR(fd, &fds);
__pmCloseSocket(clientfd);
clientfd = -1;
numfds = maxfd() + 1;
qa_case = 0;
}
}
#ifndef IS_MINGW
if (pmcdfd >= 0 && __pmFD_ISSET(pmcdfd, &readyfds)) {
/*
* do not expect this, given synchronous commumication with the
* pmcd ... either pmcd has terminated, or bogus PDU ... or its
* Win32 and we are operating under the different conditions of
* our AF.c implementation there, which has to deal with a lack
* of signal support on Windows - race condition exists between
* this check and the async event timer callback.
*/
__pmPDU *pb;
__pmPDUHdr *php;
sts = __pmGetPDU(pmcdfd, ANY_SIZE, TIMEOUT_NEVER, &pb);
if (sts <= 0) {
if (sts < 0)
fprintf(stderr, "Error: __pmGetPDU: %s\n", pmErrStr(sts));
disconnect(sts);
}
else {
php = (__pmPDUHdr *)pb;
fprintf(stderr, "Error: Unsolicited %s PDU from PMCD\n",
__pmPDUTypeStr(php->type));
disconnect(PM_ERR_IPC);
}
if (sts > 0)
__pmUnpinPDUBuf(pb);
}
#endif
if (rsc_fd >= 0 && __pmFD_ISSET(rsc_fd, &readyfds)) {
/*
* some action on the recording session control fd
* end-of-file means launcher has quit, otherwise we
* expect one of these commands
* V<number>\n - version
* F<folio>\n - folio name
* P<name>\n - launcher's name
* R\n - launcher can replay
* D\n - detach from launcher
* Q\n - quit pmlogger
*/
char rsc_buf[MAXPATHLEN];
char *rp = rsc_buf;
char myc;
int fake_x = 0;
for (rp = rsc_buf; ; rp++) {
if (read(rsc_fd, &myc, 1) <= 0) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "recording session control: eof\n");
#endif
if (rp != rsc_buf) {
*rp = '\0';
fprintf(stderr, "Error: incomplete recording session control message: \"%s\"\n", rsc_buf);
}
fake_x = 1;
break;
}
if (rp >= &rsc_buf[MAXPATHLEN]) {
fprintf(stderr, "Error: absurd recording session control message: \"%100.100s ...\"\n", rsc_buf);
fake_x = 1;
break;
}
if (myc == '\n') {
*rp = '\0';
break;
}
*rp = myc;
}
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2) {
if (fake_x == 0)
fprintf(stderr, "recording session control: \"%s\"\n", rsc_buf);
}
#endif
if (fake_x)
do_dialog('X');
else if (strcmp(rsc_buf, "Q") == 0 ||
strcmp(rsc_buf, "D") == 0 ||
strcmp(rsc_buf, "?") == 0)
do_dialog(rsc_buf[0]);
else if (rsc_buf[0] == 'F')
folio_name = strdup(&rsc_buf[1]);
else if (rsc_buf[0] == 'P')
rsc_prog = strdup(&rsc_buf[1]);
else if (strcmp(rsc_buf, "R") == 0)
rsc_replay = 1;
else if (rsc_buf[0] == 'V' && rsc_buf[1] == '0') {
/*
* version 0 of the recording session control ...
* this is all we grok at the moment
*/
;
}
else {
fprintf(stderr, "Error: illegal recording session control message: \"%s\"\n", rsc_buf);
do_dialog('X');
}
}
}
else if (vol_switch_flag) {
newvolume(VOL_SW_SIGHUP);
vol_switch_flag = 0;
}
else if (nready < 0 && neterror() != EINTR)
fprintf(stderr, "Error: select: %s\n", netstrerror());
__pmAFunblock();
if (target_pid && !__pmProcessExists(target_pid))
exit(EXIT_SUCCESS);
if (exit_code)
break;
}
exit(exit_code);
}
