a9e6334ed · cryptotaxi247/netdata-1

@@ -394,11 +394,19 @@ int read_apps_groups_conf(const char *name)

394 // data to store for each pid

395 // see: man proc

396

397 +#define PID_LOG_IO 0x00000001

398 +#define PID_LOG_STATM 0x00000002

399 +#define PID_LOG_CMDLINE 0x00000004

400 +#define PID_LOG_FDS 0x00000008

401 +#define PID_LOG_STAT 0x00000010

402 +

403 struct pid_stat {

404 int32_t pid;

405 char comm[MAX_COMPARE_NAME + 1];

406 char cmdline[MAX_CMDLINE + 1];

407

408 + uint32_t log_thrown;

409 +

410 // char state;

411 int32_t ppid;

412 // int32_t pgrp;

@@ -583,7 +591,7 @@ int read_proc_pid_cmdline(struct pid_stat *p) {

591 ssize_t i, bytes = read(fd, p->cmdline, MAX_CMDLINE);

592 close(fd);

593

586 - if(unlikely(bytes <= 0)) goto cleanup;

594 + if(unlikely(bytes < 0)) goto cleanup;

595

596 p->cmdline[bytes] = '\0';

597 for(i = 0; i < bytes ; i++)

@@ -592,7 +600,7 @@ int read_proc_pid_cmdline(struct pid_stat *p) {

600 if(unlikely(debug))

601 fprintf(stderr, "Read file '%s' contents: %s\n", p->cmdline_filename, p->cmdline);

602

595 - return 0;

603 + return 1;

604

605 cleanup:

606 // copy the command to the command line

@@ -603,7 +611,7 @@ cleanup:

611 int read_proc_pid_ownership(struct pid_stat *p) {

612 if(unlikely(!p->stat_filename)) {

613 error("pid %d does not have a stat_filename", p->pid);

606 - return 1;

614 + return 0;

615 }

616

617 // ----------------------------------------

@@ -618,7 +626,7 @@ int read_proc_pid_ownership(struct pid_stat *p) {

626 p->uid = st.st_uid;

627 p->gid = st.st_gid;

628

621 - return 0;

629 + return 1;

630 }

631

632 int read_proc_pid_stat(struct pid_stat *p) {

@@ -749,7 +757,7 @@ int read_proc_pid_stat(struct pid_stat *p) {

757 p->cgtime = 0;

758 }

759

752 - return 0;

760 + return 1;

761

762 cleanup:

763 p->minflt = 0;

@@ -764,7 +772,7 @@ cleanup:

772 p->cgtime = 0;

773 p->num_threads = 0;

774 p->rss = 0;

767 - return 1;

775 + return 0;

776 }

777

778 int read_proc_pid_statm(struct pid_stat *p) {

@@ -792,7 +800,7 @@ int read_proc_pid_statm(struct pid_stat *p) {

800 p->statm_data = strtoull(procfile_lineword(ff, 0, 5), NULL, 10);

801 p->statm_dirty = strtoull(procfile_lineword(ff, 0, 6), NULL, 10);

802

795 - return 0;

803 + return 1;

804

805 cleanup:

806 p->statm_size = 0;

@@ -802,7 +810,7 @@ cleanup:

810 p->statm_lib = 0;

811 p->statm_data = 0;

812 p->statm_dirty = 0;

805 - return 1;

813 + return 0;

814 }

815

816 int read_proc_pid_io(struct pid_stat *p) {

@@ -866,7 +874,7 @@ int read_proc_pid_io(struct pid_stat *p) {

874 p->io_cancelled_write_bytes = 0;

875 }

876

869 - return 0;

877 + return 1;

878

879 cleanup:

880 p->io_logical_bytes_read = 0;

@@ -876,7 +884,7 @@ cleanup:

884 p->io_storage_bytes_read = 0;

885 p->io_storage_bytes_written = 0;

886 p->io_cancelled_write_bytes = 0;

879 - return 1;

887 + return 0;

888 }

889

890 unsigned long long global_utime = 0;

@@ -937,13 +945,13 @@ int read_proc_stat() {

945 global_gtime = 0;

946 }

947

940 - return 0;

948 + return 1;

949

950 cleanup:

951 global_utime = 0;

952 global_stime = 0;

953 global_gtime = 0;

946 - return 1;

954 + return 0;

955 }

956

957

@@ -1248,7 +1256,7 @@ int read_pid_file_descriptors(struct pid_stat *p) {

1256

1257 // else make it positive again, we need it

1258 // of course, the actual file may have changed, but we don't care so much

1251 - // FIXME: we could compare the inode as returned by readdir direct structure

1259 + // FIXME: we could compare the inode as returned by readdir dirent structure

1260 else p->fds[fdid] = -p->fds[fdid];

1261 }

1262 closedir(fds);

@@ -1259,9 +1267,9 @@ int read_pid_file_descriptors(struct pid_stat *p) {

1267 p->fds[c] = 0;

1268 }

1269 }

1262 - else return 1;

1270 + else return 0;

1271

1264 - return 0;

1272 + return 1;

1273 }

1274

1275 // ----------------------------------------------------------------------------

@@ -1571,6 +1579,49 @@ static int compar_pid(const void *pid1, const void *pid2) {

1579 return 1;

1580 }

1581

1582 +static inline int managed_log(struct pid_stat *p, uint32_t log, int status) {

1583 + if(unlikely(!status)) {

1584 + // error("command failed log %u, errno %d", log, errno);

1585 +

1586 + if(unlikely(debug || errno != ENOENT)) {

1587 + if(unlikely(debug || !(p->log_thrown & log))) {

1588 + p->log_thrown |= log;

1589 + switch(log) {

1590 + case PID_LOG_IO:

1591 + error("Cannot process %s/proc/%d/io (command '%s')", host_prefix, p->pid, p->comm);

1592 + break;

1593 +

1594 + case PID_LOG_STATM:

1595 + error("Cannot process %s/proc/%d/statm (command '%s')", host_prefix, p->pid, p->comm);

1596 + break;

1597 +

1598 + case PID_LOG_CMDLINE:

1599 + error("Cannot process %s/proc/%d/cmdline (command '%s')", host_prefix, p->pid, p->comm);

1600 + break;

1601 +

1602 + case PID_LOG_FDS:

1603 + error("Cannot process entries in %s/proc/%d/fd (command '%s')", host_prefix, p->pid, p->comm);

1604 + break;

1605 +

1606 + case PID_LOG_STAT:

1607 + break;

1608 +

1609 + default:

1610 + error("unhandled error for pid %d, command '%s'", p->pid, p->comm);

1611 + break;

1612 + }

1613 + }

1614 + }

1615 + errno = 0;

1616 + }

1617 + else if(unlikely(p->log_thrown & log)) {

1618 + // error("unsetting log %u on pid %d", log, p->pid);

1619 + p->log_thrown &= ~log;

1620 + }

1621 +

1622 + return status;

1623 +}

1624 +

1625 void collect_data_for_pid(pid_t pid) {

1626 if(unlikely(pid <= 0 || pid > pid_max)) {

1627 error("Invalid pid %d read (expected 1 to %d). Ignoring process.", pid, pid_max);

@@ -1586,14 +1637,9 @@ void collect_data_for_pid(pid_t pid) {

1637 // --------------------------------------------------------------------

1638 // /proc/<pid>/stat

1639

1589 - if(unlikely(read_proc_pid_stat(p))) {

1590 - if(errno != ENOENT || debug)

1591 - error("Cannot process %s/proc/%d/stat (command '%s')", host_prefix, pid, p->comm);

1592 - else

1593 - errno = 0;

1640 + if(unlikely(!managed_log(p, PID_LOG_STAT, read_proc_pid_stat(p))))

1641 // there is no reason to proceed if we cannot get its status

1642 return;

1596 - }

1643

1644 read_proc_pid_ownership(p);

1645

@@ -1606,24 +1652,14 @@ void collect_data_for_pid(pid_t pid) {

1652 // --------------------------------------------------------------------

1653 // /proc/<pid>/io

1654

1609 - if(unlikely(read_proc_pid_io(p))) {

1610 - if(errno != ENOENT || debug)

1611 - error("Cannot process %s/proc/%d/io (command '%s')", host_prefix, pid, p->comm);

1612 - else

1613 - errno = 0;

1614 - }

1655 + managed_log(p, PID_LOG_IO, read_proc_pid_io(p));

1656

1657 // --------------------------------------------------------------------

1658 // /proc/<pid>/statm

1659

1619 - if(unlikely(read_proc_pid_statm(p))) {

1620 - if(errno != ENOENT || debug)

1621 - error("Cannot process %s/proc/%d/statm (command '%s')", host_prefix, pid, p->comm);

1622 - else

1623 - errno = 0;

1660 + if(unlikely(!managed_log(p, PID_LOG_STATM, read_proc_pid_statm(p))))

1661 // there is no reason to proceed if we cannot get its memory status

1662 return;

1626 - }

1663

1664 // --------------------------------------------------------------------

1665 // link it

@@ -1632,14 +1668,8 @@ void collect_data_for_pid(pid_t pid) {

1668 // we do this only once, the first time this pid is loaded

1669 if(unlikely(p->new_entry)) {

1670 // /proc/<pid>/cmdline

1635 - if(likely(proc_pid_cmdline_is_needed)) {

1636 - if(unlikely(read_proc_pid_cmdline(p))) {

1637 - if(errno != ENOENT || debug)

1638 - error("Cannot process %s/proc/%d/cmdline (command '%s')", host_prefix, pid, p->comm);

1639 - else

1640 - errno = 0;

1641 - }

1642 - }

1671 + if(likely(proc_pid_cmdline_is_needed))

1672 + managed_log(p, PID_LOG_CMDLINE, read_proc_pid_cmdline(p));

1673

1674 if(unlikely(debug))

1675 fprintf(stderr, "apps.plugin: \tJust added %d (%s)\n", pid, p->comm);

@@ -1675,12 +1705,7 @@ void collect_data_for_pid(pid_t pid) {

1705 // --------------------------------------------------------------------

1706 // /proc/<pid>/fd

1707

1678 - if(unlikely(read_pid_file_descriptors(p))) {

1679 - if(errno != ENOENT || debug)

1680 - error("Cannot process entries in %s/proc/%d/fd (command '%s')", host_prefix, pid, p->comm);

1681 - else

1682 - errno = 0;

1683 - }

1708 + managed_log(p, PID_LOG_FDS, read_pid_file_descriptors(p));

1709

1710 // --------------------------------------------------------------------

1711 // done!

@@ -2752,6 +2777,8 @@ int main(int argc, char **argv)

2777 // set the name for logging

2778 program_name = "apps.plugin";

2779

2780 + info("started on pid %d", getpid());

2781 +

2782 // disable syslog for apps.plugin

2783 error_log_syslog = 0;

2784

@@ -2761,17 +2788,17 @@ int main(int argc, char **argv)

2788

2789 host_prefix = getenv("NETDATA_HOST_PREFIX");

2790 if(host_prefix == NULL) {

2764 - info("NETDATA_HOST_PREFIX is not passed from netdata");

2791 + // info("NETDATA_HOST_PREFIX is not passed from netdata");

2792 host_prefix = "";

2793 }

2767 - else info("Found NETDATA_HOST_PREFIX='%s'", host_prefix);

2794 + // else info("Found NETDATA_HOST_PREFIX='%s'", host_prefix);

2795

2796 config_dir = getenv("NETDATA_CONFIG_DIR");

2797 if(config_dir == NULL) {

2771 - info("NETDATA_CONFIG_DIR is not passed from netdata");

2798 + // info("NETDATA_CONFIG_DIR is not passed from netdata");

2799 config_dir = CONFIG_DIR;

2800 }

2774 - else info("Found NETDATA_CONFIG_DIR='%s'", config_dir);

2801 + // else info("Found NETDATA_CONFIG_DIR='%s'", config_dir);

2802

2803 #ifdef NETDATA_INTERNAL_CHECKS

2804 if(debug_flags != 0) {

apps.plugin logs repeating errors, only once per process; fixes #779