@cryptotaxi247 / netdata-1 / commits / 7b7b03d9a

cgroups CPU usage is per core; fixes #906

Costa Tsaousis (ktsaou) committed Sep 7, 2016 at 10:04 UTC 7b7b03d9aa337d65728dea613b20e00df873af17

5 files changed +120 -101

src/apps_plugin.c

+20 -75

@@ -13,8 +13,6 @@

13 // etc.

14 #define RATES_DETAIL 10000ULL

16 -int processors = 1;

17 -pid_t pid_max = 32768;

16 int debug = 0;

18 int update_every = 1;

@@ -22,7 +20,6 @@ unsigned long long global_iterations_counter = 1;

20 unsigned long long file_counter = 0;

21 int proc_pid_cmdline_is_needed = 0;

22 int include_exited_childs = 1;

25 -char *host_prefix = "";

23 char *config_dir = CONFIG_DIR;

25 pid_t *all_pids_sortlist = NULL;

@@ -41,58 +38,6 @@ void netdata_cleanup_and_exit(int ret) {

38 }

44 -// ----------------------------------------------------------------------------

45 -// system functions

46 -// to retrieve settings of the system

47 -

48 -long get_system_cpus(void) {

49 - procfile *ff = NULL;

50 -

51 - int processors = 0;

52 -

53 - char filename[FILENAME_MAX + 1];

54 - snprintfz(filename, FILENAME_MAX, "%s/proc/stat", host_prefix);

55 -

56 - ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);

57 - if(!ff) return 1;

58 -

59 - ff = procfile_readall(ff);

60 - if(!ff)

61 - return 1;

62 -

63 - unsigned int i;

64 - for(i = 0; i < procfile_lines(ff); i++) {

65 - if(!procfile_linewords(ff, i)) continue;

66 -

67 - if(strncmp(procfile_lineword(ff, i, 0), "cpu", 3) == 0) processors++;

68 - }

69 - processors--;

70 - if(processors < 1) processors = 1;

71 -

72 - procfile_close(ff);

73 - return processors;

74 -}

75 -

76 -pid_t get_system_pid_max(void) {

77 - procfile *ff = NULL;

78 - pid_t mpid = 32768;

79 -

80 - char filename[FILENAME_MAX + 1];

81 - snprintfz(filename, FILENAME_MAX, "%s/proc/sys/kernel/pid_max", host_prefix);

82 - ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);

83 - if(!ff) return mpid;

84 -

85 - ff = procfile_readall(ff);

86 - if(!ff)

87 - return mpid;

88 -

89 - mpid = (pid_t)atoi(procfile_lineword(ff, 0, 0));

90 - if(!mpid) mpid = 32768;

91 -

92 - procfile_close(ff);

93 - return mpid;

94 -}

95 -

41 // ----------------------------------------------------------------------------

42 // target

43 // target is the structure that process data are aggregated

@@ -576,7 +521,7 @@ int read_proc_pid_cmdline(struct pid_stat *p) {

521

522 if(unlikely(!p->cmdline_filename)) {

523 char filename[FILENAME_MAX + 1];

579 - snprintfz(filename, FILENAME_MAX, "%s/proc/%d/cmdline", host_prefix, p->pid);

524 + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/cmdline", global_host_prefix, p->pid);

525 p->cmdline_filename = strdupz(filename);

526 }

527

@@ -629,7 +574,7 @@ int read_proc_pid_stat(struct pid_stat *p) {

574

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

576 char filename[FILENAME_MAX + 1];

632 - snprintfz(filename, FILENAME_MAX, "%s/proc/%d/stat", host_prefix, p->pid);

577 + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/stat", global_host_prefix, p->pid);

578 p->stat_filename = strdupz(filename);

579 }

580

@@ -737,7 +682,7 @@ int read_proc_pid_stat(struct pid_stat *p) {

682 }

683

684 if(unlikely(debug || (p->target && p->target->debug)))

740 - fprintf(stderr, "apps.plugin: READ PROC/PID/STAT: %s/proc/%d/stat, process: '%s' on target '%s' (dt=%llu) VALUES: utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu, threads=%d\n", host_prefix, p->pid, p->comm, (p->target)?p->target->name:"UNSET", p->stat_collected_usec - p->last_stat_collected_usec, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt, p->num_threads);

685 + fprintf(stderr, "apps.plugin: READ PROC/PID/STAT: %s/proc/%d/stat, process: '%s' on target '%s' (dt=%llu) VALUES: utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu, threads=%d\n", global_host_prefix, p->pid, p->comm, (p->target)?p->target->name:"UNSET", p->stat_collected_usec - p->last_stat_collected_usec, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt, p->num_threads);

686

687 if(unlikely(global_iterations_counter == 1)) {

688 p->minflt = 0;

@@ -775,7 +720,7 @@ int read_proc_pid_statm(struct pid_stat *p) {

720

721 if(unlikely(!p->statm_filename)) {

722 char filename[FILENAME_MAX + 1];

778 - snprintfz(filename, FILENAME_MAX, "%s/proc/%d/statm", host_prefix, p->pid);

723 + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/statm", global_host_prefix, p->pid);

724 p->statm_filename = strdupz(filename);

725 }

726

@@ -813,7 +758,7 @@ int read_proc_pid_io(struct pid_stat *p) {

758

759 if(unlikely(!p->io_filename)) {

760 char filename[FILENAME_MAX + 1];

816 - snprintfz(filename, FILENAME_MAX, "%s/proc/%d/io", host_prefix, p->pid);

761 + snprintfz(filename, FILENAME_MAX, "%s/proc/%d/io", global_host_prefix, p->pid);

762 p->io_filename = strdupz(filename);

763 }

764

@@ -892,7 +837,7 @@ int read_proc_stat() {

837 static unsigned long long utime_raw = 0, stime_raw = 0, gtime_raw = 0, gntime_raw = 0, ntime_raw = 0, collected_usec = 0, last_collected_usec = 0;

838

839 if(unlikely(!ff)) {

895 - snprintfz(filename, FILENAME_MAX, "%s/proc/stat", host_prefix);

840 + snprintfz(filename, FILENAME_MAX, "%s/proc/stat", global_host_prefix);

841 ff = procfile_open(filename, " \t:", PROCFILE_FLAG_DEFAULT);

842 if(unlikely(!ff)) goto cleanup;

843 }

@@ -1194,7 +1139,7 @@ int file_descriptor_find_or_add(const char *name)

1139 int read_pid_file_descriptors(struct pid_stat *p) {

1140 char dirname[FILENAME_MAX+1];

1141

1197 - snprintfz(dirname, FILENAME_MAX, "%s/proc/%d/fd", host_prefix, p->pid);

1142 + snprintfz(dirname, FILENAME_MAX, "%s/proc/%d/fd", global_host_prefix, p->pid);

1143 DIR *fds = opendir(dirname);

1144 if(fds) {

1145 int c;

@@ -1232,7 +1177,7 @@ int read_pid_file_descriptors(struct pid_stat *p) {

1177 if(p->fds[fdid] == 0) {

1178 // we don't know this fd, get it

1179

1235 - sprintf(fdname, "%s/proc/%d/fd/%s", host_prefix, p->pid, de->d_name);

1180 + sprintf(fdname, "%s/proc/%d/fd/%s", global_host_prefix, p->pid, de->d_name);

1181 ssize_t l = readlink(fdname, linkname, FILENAME_MAX);

1182 if(l == -1) {

1183 if(debug || (p->target && p->target->debug)) {

@@ -1583,19 +1528,19 @@ static inline int managed_log(struct pid_stat *p, uint32_t log, int status) {

1528 p->log_thrown |= log;

1529 switch(log) {

1530 case PID_LOG_IO:

1586 - error("Cannot process %s/proc/%d/io (command '%s')", host_prefix, p->pid, p->comm);

1531 + error("Cannot process %s/proc/%d/io (command '%s')", global_host_prefix, p->pid, p->comm);

1532 break;

1533

1534 case PID_LOG_STATM:

1590 - error("Cannot process %s/proc/%d/statm (command '%s')", host_prefix, p->pid, p->comm);

1535 + error("Cannot process %s/proc/%d/statm (command '%s')", global_host_prefix, p->pid, p->comm);

1536 break;

1537

1538 case PID_LOG_CMDLINE:

1594 - error("Cannot process %s/proc/%d/cmdline (command '%s')", host_prefix, p->pid, p->comm);

1539 + error("Cannot process %s/proc/%d/cmdline (command '%s')", global_host_prefix, p->pid, p->comm);

1540 break;

1541

1542 case PID_LOG_FDS:

1598 - error("Cannot process entries in %s/proc/%d/fd (command '%s')", host_prefix, p->pid, p->comm);

1543 + error("Cannot process entries in %s/proc/%d/fd (command '%s')", global_host_prefix, p->pid, p->comm);

1544 break;

1545

1546 case PID_LOG_STAT:

@@ -1751,7 +1696,7 @@ int collect_data_for_all_processes_from_proc(void) {

1696

1697 char dirname[FILENAME_MAX + 1];

1698

1754 - snprintfz(dirname, FILENAME_MAX, "%s/proc", host_prefix);

1699 + snprintfz(dirname, FILENAME_MAX, "%s/proc", global_host_prefix);

1700 DIR *dir = opendir(dirname);

1701 if(!dir) return 0;

1702

@@ -2803,12 +2748,12 @@ int main(int argc, char **argv)

2748 error_log_errors_per_period = 100;

2749 error_log_throttle_period = 3600;

2750

2806 - host_prefix = getenv("NETDATA_HOST_PREFIX");

2807 - if(host_prefix == NULL) {

2751 + global_host_prefix = getenv("NETDATA_HOST_PREFIX");

2752 + if(global_host_prefix == NULL) {

2753 // info("NETDATA_HOST_PREFIX is not passed from netdata");

2809 - host_prefix = "";

2754 + global_host_prefix = "";

2755 }

2811 - // else info("Found NETDATA_HOST_PREFIX='%s'", host_prefix);

2756 + // else info("Found NETDATA_HOST_PREFIX='%s'", global_host_prefix);

2757

2758 config_dir = getenv("NETDATA_CONFIG_DIR");

2759 if(config_dir == NULL) {

@@ -2830,9 +2775,9 @@ int main(int argc, char **argv)

2775

2776 time_t started_t = time(NULL);

2777 time_t current_t;

2833 - get_HZ();

2834 - pid_max = get_system_pid_max();

2835 - processors = get_system_cpus();

2778 + get_system_HZ();

2779 + get_system_pid_max();

2780 + get_system_cpus();

2781

2782 parse_args(argc, argv);

2783

src/common.c

+83 -17

@@ -860,23 +860,6 @@ int fd_is_valid(int fd) {

860 return fcntl(fd, F_GETFD) != -1 || errno != EBADF;

861 }

862

863 -/*

864 - ***************************************************************************

865 - * Get number of clock ticks per second.

866 - ***************************************************************************

867 - */

868 -unsigned int hz;

869 -

870 -void get_HZ(void) {

871 - long ticks;

872 -

873 - if ((ticks = sysconf(_SC_CLK_TCK)) == -1) {

874 - perror("sysconf");

875 - }

876 -

877 - hz = (unsigned int) ticks;

878 -}

879 -

863 pid_t gettid(void) {

864 return (pid_t)syscall(SYS_gettid);

865 }

@@ -933,3 +916,86 @@ int snprintfz(char *dst, size_t n, const char *fmt, ...) {

916

917 return ret;

918 }

919 +

920 +// ----------------------------------------------------------------------------

921 +// system functions

922 +// to retrieve settings of the system

923 +

924 +int processors = 1;

925 +long get_system_cpus(void) {

926 + procfile *ff = NULL;

927 +

928 + processors = 1;

929 +

930 + char filename[FILENAME_MAX + 1];

931 + snprintfz(filename, FILENAME_MAX, "%s/proc/stat", global_host_prefix);

932 +

933 + ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);

934 + if(!ff) {

935 + error("Cannot open file '%s'. Assuming system has %d processors.", filename, processors);

936 + return processors;

937 + }

938 +

939 + ff = procfile_readall(ff);

940 + if(!ff) {

941 + error("Cannot open file '%s'. Assuming system has %d processors.", filename, processors);

942 + return processors;

943 + }

944 +

945 + processors = 0;

946 + unsigned int i;

947 + for(i = 0; i < procfile_lines(ff); i++) {

948 + if(!procfile_linewords(ff, i)) continue;

949 +

950 + if(strncmp(procfile_lineword(ff, i, 0), "cpu", 3) == 0) processors++;

951 + }

952 + processors--;

953 + if(processors < 1) processors = 1;

954 +

955 + procfile_close(ff);

956 +

957 + info("System has %d processors.", processors);

958 + return processors;

959 +}

960 +

961 +pid_t pid_max = 32768;

962 +pid_t get_system_pid_max(void) {

963 + procfile *ff = NULL;

964 +

965 + char filename[FILENAME_MAX + 1];

966 + snprintfz(filename, FILENAME_MAX, "%s/proc/sys/kernel/pid_max", global_host_prefix);

967 + ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);

968 + if(!ff) {

969 + error("Cannot open file '%s'. Assuming system supports %d pids.", filename, pid_max);

970 + return pid_max;

971 + }

972 +

973 + ff = procfile_readall(ff);

974 + if(!ff) {

975 + error("Cannot read file '%s'. Assuming system supports %d pids.", filename, pid_max);

976 + return pid_max;

977 + }

978 +

979 + pid_max = (pid_t)atoi(procfile_lineword(ff, 0, 0));

980 + if(!pid_max) {

981 + procfile_close(ff);

982 + pid_max = 32768;

983 + error("Cannot parse file '%s'. Assuming system supports %d pids.", filename, pid_max);

984 + return pid_max;

985 + }

986 +

987 + procfile_close(ff);

988 + info("System supports %d pids.", pid_max);

989 + return pid_max;

990 +}

991 +

992 +unsigned int hz;

993 +void get_system_HZ(void) {

994 + long ticks;

995 +

996 + if ((ticks = sysconf(_SC_CLK_TCK)) == -1) {

997 + perror("sysconf");

998 + }

999 +

1000 + hz = (unsigned int) ticks;

1001 +}

src/common.h

+11 -4

@@ -153,10 +153,6 @@ extern int fd_is_valid(int fd);

153 extern char *global_host_prefix;

154 extern int enable_ksm;

155

156 -/* Number of ticks per second */

157 -extern unsigned int hz;

158 -extern void get_HZ(void);

159 -

156 extern pid_t gettid(void);

157

158 extern unsigned long long time_usec(void);

@@ -164,6 +160,17 @@ extern int sleep_usec(unsigned long long usec);

160

161 extern char *fgets_trim_len(char *buf, size_t buf_size, FILE *fp, size_t *len);

162

163 +extern int processors;

164 +extern long get_system_cpus(void);

165 +

166 +extern pid_t pid_max;

167 +extern pid_t get_system_pid_max(void);

168 +

169 +/* Number of ticks per second */

170 +extern unsigned int hz;

171 +extern void get_system_HZ(void);

172 +

173 +

174 /* fix for alpine linux */

175 #ifndef RUSAGE_THREAD

176 #ifdef RUSAGE_CHILDREN

src/main.c

+4 -3

@@ -283,9 +283,6 @@ int main(int argc, char **argv)

283 size_t wanted_stacksize = 0, stacksize = 0;

284 pthread_attr_t attr;

285

286 - // global initialization

287 - get_HZ();

288 -

286 // set the name for logging

287 program_name = "netdata";

288

@@ -472,6 +469,10 @@ int main(int argc, char **argv)

469 global_host_prefix = config_get("global", "host access prefix", "");

470 setenv("NETDATA_HOST_PREFIX", global_host_prefix, 1);

471

472 + get_system_HZ();

473 + get_system_cpus();

474 + get_system_pid_max();

475 +

476 // --------------------------------------------------------------------

477

478 stdout_filename = config_get("global", "debug log", LOG_DIR "/debug.log");

src/sys_fs_cgroup.c

+2 -2

@@ -1077,7 +1077,7 @@ void update_cgroup_charts(int update_every) {

1077 if(cg->cpuacct_stat.updated) {

1078 st = rrdset_find_bytype(type, "cpu");

1079 if(!st) {

1080 - snprintfz(title, CHART_TITLE_MAX, "CPU Usage for cgroup %s", cg->chart_title);

1080 + snprintfz(title, CHART_TITLE_MAX, "CPU Usage (%d%% = %d core%s) for cgroup %s", (processors * 100), processors, (processors>1)?"s":"", cg->chart_title);

1081 st = rrdset_create(type, "cpu", NULL, "cpu", "cgroup.cpu", title, "%", 40000, update_every, RRDSET_TYPE_STACKED);

1082

1083 rrddim_add(st, "user", NULL, 100, hz, RRDDIM_INCREMENTAL);

@@ -1096,7 +1096,7 @@ void update_cgroup_charts(int update_every) {

1096

1097 st = rrdset_find_bytype(type, "cpu_per_core");

1098 if(!st) {

1099 - snprintfz(title, CHART_TITLE_MAX, "CPU Usage Per Core for cgroup %s", cg->chart_title);

1099 + snprintfz(title, CHART_TITLE_MAX, "CPU Usage (%d%% = %d core%s) Per Core for cgroup %s", (processors * 100), processors, (processors>1)?"s":"", cg->chart_title);

1100 st = rrdset_create(type, "cpu_per_core", NULL, "cpu", "cgroup.cpu_per_core", title, "%", 40100, update_every, RRDSET_TYPE_STACKED);

1101

1102 for(i = 0; i < cg->cpuacct_usage.cpus ;i++) {