summaryrefslogtreecommitdiffstats
path: root/src/sys_fs_cgroup.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/sys_fs_cgroup.c')
-rw-r--r--src/sys_fs_cgroup.c2191
1 files changed, 1129 insertions, 1062 deletions
diff --git a/src/sys_fs_cgroup.c b/src/sys_fs_cgroup.c
index 9f3d3f0f..892f737c 100644
--- a/src/sys_fs_cgroup.c
+++ b/src/sys_fs_cgroup.c
@@ -1,22 +1,4 @@
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <inttypes.h>
-#include <sys/types.h>
-#include <dirent.h>
-#include <string.h>
-
#include "common.h"
-#include "appconfig.h"
-#include "procfile.h"
-#include "log.h"
-#include "rrd.h"
-#include "main.h"
-#include "popen.h"
-#include "proc_self_mountinfo.h"
// ----------------------------------------------------------------------------
// cgroup globals
@@ -24,164 +6,187 @@
static int cgroup_enable_cpuacct_stat = CONFIG_ONDEMAND_ONDEMAND;
static int cgroup_enable_cpuacct_usage = CONFIG_ONDEMAND_ONDEMAND;
static int cgroup_enable_memory = CONFIG_ONDEMAND_ONDEMAND;
+static int cgroup_enable_devices = CONFIG_ONDEMAND_ONDEMAND;
static int cgroup_enable_blkio = CONFIG_ONDEMAND_ONDEMAND;
static int cgroup_enable_new_cgroups_detected_at_runtime = 1;
static int cgroup_check_for_new_every = 10;
static char *cgroup_cpuacct_base = NULL;
static char *cgroup_blkio_base = NULL;
static char *cgroup_memory_base = NULL;
+static char *cgroup_devices_base = NULL;
static int cgroup_root_count = 0;
static int cgroup_root_max = 500;
static int cgroup_max_depth = 0;
void read_cgroup_plugin_configuration() {
- cgroup_check_for_new_every = config_get_number("plugin:cgroups", "check for new cgroups every", cgroup_check_for_new_every);
-
- cgroup_enable_cpuacct_stat = config_get_boolean_ondemand("plugin:cgroups", "enable cpuacct stat", cgroup_enable_cpuacct_stat);
- cgroup_enable_cpuacct_usage = config_get_boolean_ondemand("plugin:cgroups", "enable cpuacct usage", cgroup_enable_cpuacct_usage);
- cgroup_enable_memory = config_get_boolean_ondemand("plugin:cgroups", "enable memory", cgroup_enable_memory);
- cgroup_enable_blkio = config_get_boolean_ondemand("plugin:cgroups", "enable blkio", cgroup_enable_blkio);
-
- char filename[FILENAME_MAX + 1], *s;
- struct mountinfo *mi, *root = mountinfo_read();
-
- mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "cpuacct");
- if(!mi) mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "cpuacct");
- if(!mi) s = "/sys/fs/cgroup/cpuacct";
- else s = mi->mount_point;
- snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
- cgroup_cpuacct_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/cpuacct", filename);
-
- mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "blkio");
- if(!mi) mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "blkio");
- if(!mi) s = "/sys/fs/cgroup/blkio";
- else s = mi->mount_point;
- snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
- cgroup_blkio_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/blkio", filename);
-
- mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "memory");
- if(!mi) mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "memory");
- if(!mi) s = "/sys/fs/cgroup/memory";
- else s = mi->mount_point;
- snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
- cgroup_memory_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/memory", filename);
-
- cgroup_root_max = config_get_number("plugin:cgroups", "max cgroups to allow", cgroup_root_max);
- cgroup_max_depth = config_get_number("plugin:cgroups", "max cgroups depth to monitor", cgroup_max_depth);
-
- cgroup_enable_new_cgroups_detected_at_runtime = config_get_boolean("plugin:cgroups", "enable new cgroups detected at run time", cgroup_enable_new_cgroups_detected_at_runtime);
-
- mountinfo_free(root);
+ cgroup_check_for_new_every = config_get_number("plugin:cgroups", "check for new cgroups every", cgroup_check_for_new_every);
+
+ cgroup_enable_cpuacct_stat = config_get_boolean_ondemand("plugin:cgroups", "enable cpuacct stat", cgroup_enable_cpuacct_stat);
+ cgroup_enable_cpuacct_usage = config_get_boolean_ondemand("plugin:cgroups", "enable cpuacct usage", cgroup_enable_cpuacct_usage);
+ cgroup_enable_memory = config_get_boolean_ondemand("plugin:cgroups", "enable memory", cgroup_enable_memory);
+ cgroup_enable_blkio = config_get_boolean_ondemand("plugin:cgroups", "enable blkio", cgroup_enable_blkio);
+
+ char filename[FILENAME_MAX + 1], *s;
+ struct mountinfo *mi, *root = mountinfo_read();
+
+ mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "cpuacct");
+ if(!mi) mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "cpuacct");
+ if(!mi) {
+ error("Cannot find cgroup cpuacct mountinfo. Assuming default: /sys/fs/cgroup/cpuacct");
+ s = "/sys/fs/cgroup/cpuacct";
+ }
+ else s = mi->mount_point;
+ snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
+ cgroup_cpuacct_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/cpuacct", filename);
+
+ mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "blkio");
+ if(!mi) mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "blkio");
+ if(!mi) {
+ error("Cannot find cgroup blkio mountinfo. Assuming default: /sys/fs/cgroup/blkio");
+ s = "/sys/fs/cgroup/blkio";
+ }
+ else s = mi->mount_point;
+ snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
+ cgroup_blkio_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/blkio", filename);
+
+ mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "memory");
+ if(!mi) mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "memory");
+ if(!mi) {
+ error("Cannot find cgroup memory mountinfo. Assuming default: /sys/fs/cgroup/memory");
+ s = "/sys/fs/cgroup/memory";
+ }
+ else s = mi->mount_point;
+ snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
+ cgroup_memory_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/memory", filename);
+
+ mi = mountinfo_find_by_filesystem_super_option(root, "cgroup", "devices");
+ if(!mi) mi = mountinfo_find_by_filesystem_mount_source(root, "cgroup", "devices");
+ if(!mi) {
+ error("Cannot find cgroup devices mountinfo. Assuming default: /sys/fs/cgroup/devices");
+ s = "/sys/fs/cgroup/devices";
+ }
+ else s = mi->mount_point;
+ snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, s);
+ cgroup_devices_base = config_get("plugin:cgroups", "path to /sys/fs/cgroup/devices", filename);
+
+ cgroup_root_max = config_get_number("plugin:cgroups", "max cgroups to allow", cgroup_root_max);
+ cgroup_max_depth = config_get_number("plugin:cgroups", "max cgroups depth to monitor", cgroup_max_depth);
+
+ cgroup_enable_new_cgroups_detected_at_runtime = config_get_boolean("plugin:cgroups", "enable new cgroups detected at run time", cgroup_enable_new_cgroups_detected_at_runtime);
+
+ mountinfo_free(root);
}
// ----------------------------------------------------------------------------
// cgroup objects
struct blkio {
- int updated;
+ int updated;
- char *filename;
+ char *filename;
- unsigned long long Read;
- unsigned long long Write;
+ unsigned long long Read;
+ unsigned long long Write;
/*
- unsigned long long Sync;
- unsigned long long Async;
- unsigned long long Total;
+ unsigned long long Sync;
+ unsigned long long Async;
+ unsigned long long Total;
*/
};
// https://www.kernel.org/doc/Documentation/cgroup-v1/memory.txt
struct memory {
- int updated;
-
- char *filename;
-
- int has_dirty_swap;
-
- unsigned long long cache;
- unsigned long long rss;
- unsigned long long rss_huge;
- unsigned long long mapped_file;
- unsigned long long writeback;
- unsigned long long dirty;
- unsigned long long swap;
- unsigned long long pgpgin;
- unsigned long long pgpgout;
- unsigned long long pgfault;
- unsigned long long pgmajfault;
+ int updated;
+
+ char *filename;
+
+ int has_dirty_swap;
+
+ unsigned long long cache;
+ unsigned long long rss;
+ unsigned long long rss_huge;
+ unsigned long long mapped_file;
+ unsigned long long writeback;
+ unsigned long long dirty;
+ unsigned long long swap;
+ unsigned long long pgpgin;
+ unsigned long long pgpgout;
+ unsigned long long pgfault;
+ unsigned long long pgmajfault;
/*
- unsigned long long inactive_anon;
- unsigned long long active_anon;
- unsigned long long inactive_file;
- unsigned long long active_file;
- unsigned long long unevictable;
- unsigned long long hierarchical_memory_limit;
- unsigned long long total_cache;
- unsigned long long total_rss;
- unsigned long long total_rss_huge;
- unsigned long long total_mapped_file;
- unsigned long long total_writeback;
- unsigned long long total_dirty;
- unsigned long long total_swap;
- unsigned long long total_pgpgin;
- unsigned long long total_pgpgout;
- unsigned long long total_pgfault;
- unsigned long long total_pgmajfault;
- unsigned long long total_inactive_anon;
- unsigned long long total_active_anon;
- unsigned long long total_inactive_file;
- unsigned long long total_active_file;
- unsigned long long total_unevictable;
+ unsigned long long inactive_anon;
+ unsigned long long active_anon;
+ unsigned long long inactive_file;
+ unsigned long long active_file;
+ unsigned long long unevictable;
+ unsigned long long hierarchical_memory_limit;
+ unsigned long long total_cache;
+ unsigned long long total_rss;
+ unsigned long long total_rss_huge;
+ unsigned long long total_mapped_file;
+ unsigned long long total_writeback;
+ unsigned long long total_dirty;
+ unsigned long long total_swap;
+ unsigned long long total_pgpgin;
+ unsigned long long total_pgpgout;
+ unsigned long long total_pgfault;
+ unsigned long long total_pgmajfault;
+ unsigned long long total_inactive_anon;
+ unsigned long long total_active_anon;
+ unsigned long long total_inactive_file;
+ unsigned long long total_active_file;
+ unsigned long long total_unevictable;
*/
};
// https://www.kernel.org/doc/Documentation/cgroup-v1/cpuacct.txt
struct cpuacct_stat {
- int updated;
+ int updated;
- char *filename;
+ char *filename;
- unsigned long long user;
- unsigned long long system;
+ unsigned long long user;
+ unsigned long long system;
};
// https://www.kernel.org/doc/Documentation/cgroup-v1/cpuacct.txt
struct cpuacct_usage {
- int updated;
+ int updated;
- char *filename;
+ char *filename;
- unsigned int cpus;
- unsigned long long *cpu_percpu;
+ unsigned int cpus;
+ unsigned long long *cpu_percpu;
};
struct cgroup {
- int available; // found in the filesystem
- int enabled; // enabled in the config
+ int available; // found in the filesystem
+ int enabled; // enabled in the config
+
+ char *id;
+ uint32_t hash;
- char *id;
- uint32_t hash;
+ char *chart_id;
+ uint32_t hash_chart;
- char *chart_id;
- char *chart_title;
+ char *chart_title;
- struct cpuacct_stat cpuacct_stat;
- struct cpuacct_usage cpuacct_usage;
+ struct cpuacct_stat cpuacct_stat;
+ struct cpuacct_usage cpuacct_usage;
- struct memory memory;
+ struct memory memory;
- struct blkio io_service_bytes; // bytes
- struct blkio io_serviced; // operations
+ struct blkio io_service_bytes; // bytes
+ struct blkio io_serviced; // operations
- struct blkio throttle_io_service_bytes; // bytes
- struct blkio throttle_io_serviced; // operations
+ struct blkio throttle_io_service_bytes; // bytes
+ struct blkio throttle_io_serviced; // operations
- struct blkio io_merged; // operations
- struct blkio io_queued; // operations
+ struct blkio io_merged; // operations
+ struct blkio io_queued; // operations
- struct cgroup *next;
+ struct cgroup *next;
} *cgroup_root = NULL;
@@ -189,393 +194,389 @@ struct cgroup {
// read values from /sys
void cgroup_read_cpuacct_stat(struct cpuacct_stat *cp) {
- static procfile *ff = NULL;
+ static procfile *ff = NULL;
- static uint32_t user_hash = 0;
- static uint32_t system_hash = 0;
+ static uint32_t user_hash = 0;
+ static uint32_t system_hash = 0;
- if(unlikely(user_hash == 0)) {
- user_hash = simple_hash("user");
- system_hash = simple_hash("system");
- }
+ if(unlikely(user_hash == 0)) {
+ user_hash = simple_hash("user");
+ system_hash = simple_hash("system");
+ }
- cp->updated = 0;
- if(cp->filename) {
- ff = procfile_reopen(ff, cp->filename, NULL, PROCFILE_FLAG_DEFAULT);
- if(!ff) return;
+ cp->updated = 0;
+ if(cp->filename) {
+ ff = procfile_reopen(ff, cp->filename, NULL, PROCFILE_FLAG_DEFAULT);
+ if(!ff) return;
- ff = procfile_readall(ff);
- if(!ff) return;
+ ff = procfile_readall(ff);
+ if(!ff) return;
- unsigned long i, lines = procfile_lines(ff);
+ unsigned long i, lines = procfile_lines(ff);
- if(lines < 1) {
- error("File '%s' should have 1+ lines.", cp->filename);
- return;
- }
+ if(lines < 1) {
+ error("File '%s' should have 1+ lines.", cp->filename);
+ return;
+ }
- for(i = 0; i < lines ; i++) {
- char *s = procfile_lineword(ff, i, 0);
- uint32_t hash = simple_hash(s);
+ for(i = 0; i < lines ; i++) {
+ char *s = procfile_lineword(ff, i, 0);
+ uint32_t hash = simple_hash(s);
- if(hash == user_hash && !strcmp(s, "user"))
- cp->user = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ if(hash == user_hash && !strcmp(s, "user"))
+ cp->user = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == system_hash && !strcmp(s, "system"))
- cp->system = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- }
+ else if(hash == system_hash && !strcmp(s, "system"))
+ cp->system = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ }
- cp->updated = 1;
+ cp->updated = 1;
- // fprintf(stderr, "READ '%s': user: %llu, system: %llu\n", cp->filename, cp->user, cp->system);
- }
+ // fprintf(stderr, "READ '%s': user: %llu, system: %llu\n", cp->filename, cp->user, cp->system);
+ }
}
void cgroup_read_cpuacct_usage(struct cpuacct_usage *ca) {
- static procfile *ff = NULL;
-
- ca->updated = 0;
- if(ca->filename) {
- ff = procfile_reopen(ff, ca->filename, NULL, PROCFILE_FLAG_DEFAULT);
- if(!ff) return;
-
- ff = procfile_readall(ff);
- if(!ff) return;
-
- if(procfile_lines(ff) < 1) {
- error("File '%s' should have 1+ lines but has %d.", ca->filename, procfile_lines(ff));
- return;
- }
-
- unsigned long i = procfile_linewords(ff, 0);
- if(i <= 0) return;
-
- // we may have 1 more CPU reported
- while(i > 0) {
- char *s = procfile_lineword(ff, 0, i - 1);
- if(!*s) i--;
- else break;
- }
-
- if(i != ca->cpus) {
- free(ca->cpu_percpu);
-
- ca->cpu_percpu = malloc(sizeof(unsigned long long) * i);
- if(!ca->cpu_percpu)
- fatal("Cannot allocate memory (%z bytes)", sizeof(unsigned long long) * i);
-
- ca->cpus = i;
- }
-
- for(i = 0; i < ca->cpus ;i++) {
- ca->cpu_percpu[i] = strtoull(procfile_lineword(ff, 0, i), NULL, 10);
- // fprintf(stderr, "READ '%s': cpu%d/%d: %llu ('%s')\n", ca->filename, i, ca->cpus, ca->cpu_percpu[i], procfile_lineword(ff, 0, i));
- }
-
- ca->updated = 1;
- }
+ static procfile *ff = NULL;
+
+ ca->updated = 0;
+ if(ca->filename) {
+ ff = procfile_reopen(ff, ca->filename, NULL, PROCFILE_FLAG_DEFAULT);
+ if(!ff) return;
+
+ ff = procfile_readall(ff);
+ if(!ff) return;
+
+ if(procfile_lines(ff) < 1) {
+ error("File '%s' should have 1+ lines but has %u.", ca->filename, procfile_lines(ff));
+ return;
+ }
+
+ unsigned long i = procfile_linewords(ff, 0);
+ if(i <= 0) return;
+
+ // we may have 1 more CPU reported
+ while(i > 0) {
+ char *s = procfile_lineword(ff, 0, i - 1);
+ if(!*s) i--;
+ else break;
+ }
+
+ if(i != ca->cpus) {
+ freez(ca->cpu_percpu);
+ ca->cpu_percpu = mallocz(sizeof(unsigned long long) * i);
+ ca->cpus = (unsigned int)i;
+ }
+
+ for(i = 0; i < ca->cpus ;i++) {
+ ca->cpu_percpu[i] = strtoull(procfile_lineword(ff, 0, i), NULL, 10);
+ // fprintf(stderr, "READ '%s': cpu%d/%d: %llu ('%s')\n", ca->filename, i, ca->cpus, ca->cpu_percpu[i], procfile_lineword(ff, 0, i));
+ }
+
+ ca->updated = 1;
+ }
}
void cgroup_read_blkio(struct blkio *io) {
- static procfile *ff = NULL;
+ static procfile *ff = NULL;
- static uint32_t Read_hash = 0;
- static uint32_t Write_hash = 0;
+ static uint32_t Read_hash = 0;
+ static uint32_t Write_hash = 0;
/*
- static uint32_t Sync_hash = 0;
- static uint32_t Async_hash = 0;
- static uint32_t Total_hash = 0;
+ static uint32_t Sync_hash = 0;
+ static uint32_t Async_hash = 0;
+ static uint32_t Total_hash = 0;
*/
- if(unlikely(Read_hash == 0)) {
- Read_hash = simple_hash("Read");
- Write_hash = simple_hash("Write");
+ if(unlikely(Read_hash == 0)) {
+ Read_hash = simple_hash("Read");
+ Write_hash = simple_hash("Write");
/*
- Sync_hash = simple_hash("Sync");
- Async_hash = simple_hash("Async");
- Total_hash = simple_hash("Total");
+ Sync_hash = simple_hash("Sync");
+ Async_hash = simple_hash("Async");
+ Total_hash = simple_hash("Total");
*/
- }
+ }
- io->updated = 0;
- if(io->filename) {
- ff = procfile_reopen(ff, io->filename, NULL, PROCFILE_FLAG_DEFAULT);
- if(!ff) return;
+ io->updated = 0;
+ if(io->filename) {
+ ff = procfile_reopen(ff, io->filename, NULL, PROCFILE_FLAG_DEFAULT);
+ if(!ff) return;
- ff = procfile_readall(ff);
- if(!ff) return;
+ ff = procfile_readall(ff);
+ if(!ff) return;
- unsigned long i, lines = procfile_lines(ff);
+ unsigned long i, lines = procfile_lines(ff);
- if(lines < 1) {
- error("File '%s' should have 1+ lines.", io->filename);
- return;
- }
+ if(lines < 1) {
+ error("File '%s' should have 1+ lines.", io->filename);
+ return;
+ }
- io->Read = 0;
- io->Write = 0;
+ io->Read = 0;
+ io->Write = 0;
/*
- io->Sync = 0;
- io->Async = 0;
- io->Total = 0;
+ io->Sync = 0;
+ io->Async = 0;
+ io->Total = 0;
*/
- for(i = 0; i < lines ; i++) {
- char *s = procfile_lineword(ff, i, 1);
- uint32_t hash = simple_hash(s);
+ for(i = 0; i < lines ; i++) {
+ char *s = procfile_lineword(ff, i, 1);
+ uint32_t hash = simple_hash(s);
- if(hash == Read_hash && !strcmp(s, "Read"))
- io->Read += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
+ if(hash == Read_hash && !strcmp(s, "Read"))
+ io->Read += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
- else if(hash == Write_hash && !strcmp(s, "Write"))
- io->Write += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
+ else if(hash == Write_hash && !strcmp(s, "Write"))
+ io->Write += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
/*
- else if(hash == Sync_hash && !strcmp(s, "Sync"))
- io->Sync += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
+ else if(hash == Sync_hash && !strcmp(s, "Sync"))
+ io->Sync += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
- else if(hash == Async_hash && !strcmp(s, "Async"))
- io->Async += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
+ else if(hash == Async_hash && !strcmp(s, "Async"))
+ io->Async += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
- else if(hash == Total_hash && !strcmp(s, "Total"))
- io->Total += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
+ else if(hash == Total_hash && !strcmp(s, "Total"))
+ io->Total += strtoull(procfile_lineword(ff, i, 2), NULL, 10);
*/
- }
+ }
- io->updated = 1;
- // fprintf(stderr, "READ '%s': Read: %llu, Write: %llu, Sync: %llu, Async: %llu, Total: %llu\n", io->filename, io->Read, io->Write, io->Sync, io->Async, io->Total);
- }
+ io->updated = 1;
+ // fprintf(stderr, "READ '%s': Read: %llu, Write: %llu, Sync: %llu, Async: %llu, Total: %llu\n", io->filename, io->Read, io->Write, io->Sync, io->Async, io->Total);
+ }
}
void cgroup_read_memory(struct memory *mem) {
- static procfile *ff = NULL;
-
- static uint32_t cache_hash = 0;
- static uint32_t rss_hash = 0;
- static uint32_t rss_huge_hash = 0;
- static uint32_t mapped_file_hash = 0;
- static uint32_t writeback_hash = 0;
- static uint32_t dirty_hash = 0;
- static uint32_t swap_hash = 0;
- static uint32_t pgpgin_hash = 0;
- static uint32_t pgpgout_hash = 0;
- static uint32_t pgfault_hash = 0;
- static uint32_t pgmajfault_hash = 0;
+ static procfile *ff = NULL;
+
+ static uint32_t cache_hash = 0;
+ static uint32_t rss_hash = 0;
+ static uint32_t rss_huge_hash = 0;
+ static uint32_t mapped_file_hash = 0;
+ static uint32_t writeback_hash = 0;
+ static uint32_t dirty_hash = 0;
+ static uint32_t swap_hash = 0;
+ static uint32_t pgpgin_hash = 0;
+ static uint32_t pgpgout_hash = 0;
+ static uint32_t pgfault_hash = 0;
+ static uint32_t pgmajfault_hash = 0;
/*
- static uint32_t inactive_anon_hash = 0;
- static uint32_t active_anon_hash = 0;
- static uint32_t inactive_file_hash = 0;
- static uint32_t active_file_hash = 0;
- static uint32_t unevictable_hash = 0;
- static uint32_t hierarchical_memory_limit_hash = 0;
- static uint32_t total_cache_hash = 0;
- static uint32_t total_rss_hash = 0;
- static uint32_t total_rss_huge_hash = 0;
- static uint32_t total_mapped_file_hash = 0;
- static uint32_t total_writeback_hash = 0;
- static uint32_t total_dirty_hash = 0;
- static uint32_t total_swap_hash = 0;
- static uint32_t total_pgpgin_hash = 0;
- static uint32_t total_pgpgout_hash = 0;
- static uint32_t total_pgfault_hash = 0;
- static uint32_t total_pgmajfault_hash = 0;
- static uint32_t total_inactive_anon_hash = 0;
- static uint32_t total_active_anon_hash = 0;
- static uint32_t total_inactive_file_hash = 0;
- static uint32_t total_active_file_hash = 0;
- static uint32_t total_unevictable_hash = 0;
+ static uint32_t inactive_anon_hash = 0;
+ static uint32_t active_anon_hash = 0;
+ static uint32_t inactive_file_hash = 0;
+ static uint32_t active_file_hash = 0;
+ static uint32_t unevictable_hash = 0;
+ static uint32_t hierarchical_memory_limit_hash = 0;
+ static uint32_t total_cache_hash = 0;
+ static uint32_t total_rss_hash = 0;
+ static uint32_t total_rss_huge_hash = 0;
+ static uint32_t total_mapped_file_hash = 0;
+ static uint32_t total_writeback_hash = 0;
+ static uint32_t total_dirty_hash = 0;
+ static uint32_t total_swap_hash = 0;
+ static uint32_t total_pgpgin_hash = 0;
+ static uint32_t total_pgpgout_hash = 0;
+ static uint32_t total_pgfault_hash = 0;
+ static uint32_t total_pgmajfault_hash = 0;
+ static uint32_t total_inactive_anon_hash = 0;
+ static uint32_t total_active_anon_hash = 0;
+ static uint32_t total_inactive_file_hash = 0;
+ static uint32_t total_active_file_hash = 0;
+ static uint32_t total_unevictable_hash = 0;
*/
- if(unlikely(cache_hash == 0)) {
- cache_hash = simple_hash("cache");
- rss_hash = simple_hash("rss");
- rss_huge_hash = simple_hash("rss_huge");
- mapped_file_hash = simple_hash("mapped_file");
- writeback_hash = simple_hash("writeback");
- dirty_hash = simple_hash("dirty");
- swap_hash = simple_hash("swap");
- pgpgin_hash = simple_hash("pgpgin");
- pgpgout_hash = simple_hash("pgpgout");
- pgfault_hash = simple_hash("pgfault");
- pgmajfault_hash = simple_hash("pgmajfault");
+ if(unlikely(cache_hash == 0)) {
+ cache_hash = simple_hash("cache");
+ rss_hash = simple_hash("rss");
+ rss_huge_hash = simple_hash("rss_huge");
+ mapped_file_hash = simple_hash("mapped_file");
+ writeback_hash = simple_hash("writeback");
+ dirty_hash = simple_hash("dirty");
+ swap_hash = simple_hash("swap");
+ pgpgin_hash = simple_hash("pgpgin");
+ pgpgout_hash = simple_hash("pgpgout");
+ pgfault_hash = simple_hash("pgfault");
+ pgmajfault_hash = simple_hash("pgmajfault");
/*
- inactive_anon_hash = simple_hash("inactive_anon");
- active_anon_hash = simple_hash("active_anon");
- inactive_file_hash = simple_hash("inactive_file");
- active_file_hash = simple_hash("active_file");
- unevictable_hash = simple_hash("unevictable");
- hierarchical_memory_limit_hash = simple_hash("hierarchical_memory_limit");
- total_cache_hash = simple_hash("total_cache");
- total_rss_hash = simple_hash("total_rss");
- total_rss_huge_hash = simple_hash("total_rss_huge");
- total_mapped_file_hash = simple_hash("total_mapped_file");
- total_writeback_hash = simple_hash("total_writeback");
- total_dirty_hash = simple_hash("total_dirty");
- total_swap_hash = simple_hash("total_swap");
- total_pgpgin_hash = simple_hash("total_pgpgin");
- total_pgpgout_hash = simple_hash("total_pgpgout");
- total_pgfault_hash = simple_hash("total_pgfault");
- total_pgmajfault_hash = simple_hash("total_pgmajfault");
- total_inactive_anon_hash = simple_hash("total_inactive_anon");
- total_active_anon_hash = simple_hash("total_active_anon");
- total_inactive_file_hash = simple_hash("total_inactive_file");
- total_active_file_hash = simple_hash("total_active_file");
- total_unevictable_hash = simple_hash("total_unevictable");
+ inactive_anon_hash = simple_hash("inactive_anon");
+ active_anon_hash = simple_hash("active_anon");
+ inactive_file_hash = simple_hash("inactive_file");
+ active_file_hash = simple_hash("active_file");
+ unevictable_hash = simple_hash("unevictable");
+ hierarchical_memory_limit_hash = simple_hash("hierarchical_memory_limit");
+ total_cache_hash = simple_hash("total_cache");
+ total_rss_hash = simple_hash("total_rss");
+ total_rss_huge_hash = simple_hash("total_rss_huge");
+ total_mapped_file_hash = simple_hash("total_mapped_file");
+ total_writeback_hash = simple_hash("total_writeback");
+ total_dirty_hash = simple_hash("total_dirty");
+ total_swap_hash = simple_hash("total_swap");
+ total_pgpgin_hash = simple_hash("total_pgpgin");
+ total_pgpgout_hash = simple_hash("total_pgpgout");
+ total_pgfault_hash = simple_hash("total_pgfault");
+ total_pgmajfault_hash = simple_hash("total_pgmajfault");
+ total_inactive_anon_hash = simple_hash("total_inactive_anon");
+ total_active_anon_hash = simple_hash("total_active_anon");
+ total_inactive_file_hash = simple_hash("total_inactive_file");
+ total_active_file_hash = simple_hash("total_active_file");
+ total_unevictable_hash = simple_hash("total_unevictable");
*/
- }
+ }
- mem->updated = 0;
- if(mem->filename) {
- ff = procfile_reopen(ff, mem->filename, NULL, PROCFILE_FLAG_DEFAULT);
- if(!ff) return;
+ mem->updated = 0;
+ if(mem->filename) {
+ ff = procfile_reopen(ff, mem->filename, NULL, PROCFILE_FLAG_DEFAULT);
+ if(!ff) return;
- ff = procfile_readall(ff);
- if(!ff) return;
+ ff = procfile_readall(ff);
+ if(!ff) return;
- unsigned long i, lines = procfile_lines(ff);
+ unsigned long i, lines = procfile_lines(ff);
- if(lines < 1) {
- error("File '%s' should have 1+ lines.", mem->filename);
- return;
- }
+ if(lines < 1) {
+ error("File '%s' should have 1+ lines.", mem->filename);
+ return;
+ }
- for(i = 0; i < lines ; i++) {
- char *s = procfile_lineword(ff, i, 0);
- uint32_t hash = simple_hash(s);
+ for(i = 0; i < lines ; i++) {
+ char *s = procfile_lineword(ff, i, 0);
+ uint32_t hash = simple_hash(s);
- if(hash == cache_hash && !strcmp(s, "cache"))
- mem->cache = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ if(hash == cache_hash && !strcmp(s, "cache"))
+ mem->cache = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == rss_hash && !strcmp(s, "rss"))
- mem->rss = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == rss_hash && !strcmp(s, "rss"))
+ mem->rss = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == rss_huge_hash && !strcmp(s, "rss_huge"))
- mem->rss_huge = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == rss_huge_hash && !strcmp(s, "rss_huge"))
+ mem->rss_huge = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == mapped_file_hash && !strcmp(s, "mapped_file"))
- mem->mapped_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == mapped_file_hash && !strcmp(s, "mapped_file"))
+ mem->mapped_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == writeback_hash && !strcmp(s, "writeback"))
- mem->writeback = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == writeback_hash && !strcmp(s, "writeback"))
+ mem->writeback = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == dirty_hash && !strcmp(s, "dirty")) {
- mem->dirty = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- mem->has_dirty_swap = 1;
- }
+ else if(hash == dirty_hash && !strcmp(s, "dirty")) {
+ mem->dirty = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ mem->has_dirty_swap = 1;
+ }
- else if(hash == swap_hash && !strcmp(s, "swap")) {
- mem->swap = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- mem->has_dirty_swap = 1;
- }
+ else if(hash == swap_hash && !strcmp(s, "swap")) {
+ mem->swap = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ mem->has_dirty_swap = 1;
+ }
- else if(hash == pgpgin_hash && !strcmp(s, "pgpgin"))
- mem->pgpgin = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == pgpgin_hash && !strcmp(s, "pgpgin"))
+ mem->pgpgin = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == pgpgout_hash && !strcmp(s, "pgpgout"))
- mem->pgpgout = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == pgpgout_hash && !strcmp(s, "pgpgout"))
+ mem->pgpgout = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == pgfault_hash && !strcmp(s, "pgfault"))
- mem->pgfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == pgfault_hash && !strcmp(s, "pgfault"))
+ mem->pgfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == pgmajfault_hash && !strcmp(s, "pgmajfault"))
- mem->pgmajfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == pgmajfault_hash && !strcmp(s, "pgmajfault"))
+ mem->pgmajfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
/*
- else if(hash == inactive_anon_hash && !strcmp(s, "inactive_anon"))
- mem->inactive_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == inactive_anon_hash && !strcmp(s, "inactive_anon"))
+ mem->inactive_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == active_anon_hash && !strcmp(s, "active_anon"))
- mem->active_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == active_anon_hash && !strcmp(s, "active_anon"))
+ mem->active_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == inactive_file_hash && !strcmp(s, "inactive_file"))
- mem->inactive_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == inactive_file_hash && !strcmp(s, "inactive_file"))
+ mem->inactive_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == active_file_hash && !strcmp(s, "active_file"))
- mem->active_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == active_file_hash && !strcmp(s, "active_file"))
+ mem->active_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == unevictable_hash && !strcmp(s, "unevictable"))
- mem->unevictable = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == unevictable_hash && !strcmp(s, "unevictable"))
+ mem->unevictable = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == hierarchical_memory_limit_hash && !strcmp(s, "hierarchical_memory_limit"))
- mem->hierarchical_memory_limit = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == hierarchical_memory_limit_hash && !strcmp(s, "hierarchical_memory_limit"))
+ mem->hierarchical_memory_limit = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_cache_hash && !strcmp(s, "total_cache"))
- mem->total_cache = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_cache_hash && !strcmp(s, "total_cache"))
+ mem->total_cache = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_rss_hash && !strcmp(s, "total_rss"))
- mem->total_rss = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_rss_hash && !strcmp(s, "total_rss"))
+ mem->total_rss = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_rss_huge_hash && !strcmp(s, "total_rss_huge"))
- mem->total_rss_huge = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_rss_huge_hash && !strcmp(s, "total_rss_huge"))
+ mem->total_rss_huge = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_mapped_file_hash && !strcmp(s, "total_mapped_file"))
- mem->total_mapped_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_mapped_file_hash && !strcmp(s, "total_mapped_file"))
+ mem->total_mapped_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_writeback_hash && !strcmp(s, "total_writeback"))
- mem->total_writeback = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_writeback_hash && !strcmp(s, "total_writeback"))
+ mem->total_writeback = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_dirty_hash && !strcmp(s, "total_dirty"))
- mem->total_dirty = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_dirty_hash && !strcmp(s, "total_dirty"))
+ mem->total_dirty = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_swap_hash && !strcmp(s, "total_swap"))
- mem->total_swap = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_swap_hash && !strcmp(s, "total_swap"))
+ mem->total_swap = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_pgpgin_hash && !strcmp(s, "total_pgpgin"))
- mem->total_pgpgin = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_pgpgin_hash && !strcmp(s, "total_pgpgin"))
+ mem->total_pgpgin = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_pgpgout_hash && !strcmp(s, "total_pgpgout"))
- mem->total_pgpgout = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_pgpgout_hash && !strcmp(s, "total_pgpgout"))
+ mem->total_pgpgout = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_pgfault_hash && !strcmp(s, "total_pgfault"))
- mem->total_pgfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_pgfault_hash && !strcmp(s, "total_pgfault"))
+ mem->total_pgfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_pgmajfault_hash && !strcmp(s, "total_pgmajfault"))
- mem->total_pgmajfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_pgmajfault_hash && !strcmp(s, "total_pgmajfault"))
+ mem->total_pgmajfault = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_inactive_anon_hash && !strcmp(s, "total_inactive_anon"))
- mem->total_inactive_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_inactive_anon_hash && !strcmp(s, "total_inactive_anon"))
+ mem->total_inactive_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_active_anon_hash && !strcmp(s, "total_active_anon"))
- mem->total_active_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_active_anon_hash && !strcmp(s, "total_active_anon"))
+ mem->total_active_anon = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_inactive_file_hash && !strcmp(s, "total_inactive_file"))
- mem->total_inactive_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_inactive_file_hash && !strcmp(s, "total_inactive_file"))
+ mem->total_inactive_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_active_file_hash && !strcmp(s, "total_active_file"))
- mem->total_active_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_active_file_hash && !strcmp(s, "total_active_file"))
+ mem->total_active_file = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
- else if(hash == total_unevictable_hash && !strcmp(s, "total_unevictable"))
- mem->total_unevictable = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
+ else if(hash == total_unevictable_hash && !strcmp(s, "total_unevictable"))
+ mem->total_unevictable = strtoull(procfile_lineword(ff, i, 1), NULL, 10);
*/
- }
+ }
- // fprintf(stderr, "READ: '%s', cache: %llu, rss: %llu, rss_huge: %llu, mapped_file: %llu, writeback: %llu, dirty: %llu, swap: %llu, pgpgin: %llu, pgpgout: %llu, pgfault: %llu, pgmajfault: %llu, inactive_anon: %llu, active_anon: %llu, inactive_file: %llu, active_file: %llu, unevictable: %llu, hierarchical_memory_limit: %llu, total_cache: %llu, total_rss: %llu, total_rss_huge: %llu, total_mapped_file: %llu, total_writeback: %llu, total_dirty: %llu, total_swap: %llu, total_pgpgin: %llu, total_pgpgout: %llu, total_pgfault: %llu, total_pgmajfault: %llu, total_inactive_anon: %llu, total_active_anon: %llu, total_inactive_file: %llu, total_active_file: %llu, total_unevictable: %llu\n", mem->filename, mem->cache, mem->rss, mem->rss_huge, mem->mapped_file, mem->writeback, mem->dirty, mem->swap, mem->pgpgin, mem->pgpgout, mem->pgfault, mem->pgmajfault, mem->inactive_anon, mem->active_anon, mem->inactive_file, mem->active_file, mem->unevictable, mem->hierarchical_memory_limit, mem->total_cache, mem->total_rss, mem->total_rss_huge, mem->total_mapped_file, mem->total_writeback, mem->total_dirty, mem->total_swap, mem->total_pgpgin, mem->total_pgpgout, mem->total_pgfault, mem->total_pgmajfault, mem->total_inactive_anon, mem->total_active_anon, mem->total_inactive_file, mem->total_active_file, mem->total_unevictable);
+ // fprintf(stderr, "READ: '%s', cache: %llu, rss: %llu, rss_huge: %llu, mapped_file: %llu, writeback: %llu, dirty: %llu, swap: %llu, pgpgin: %llu, pgpgout: %llu, pgfault: %llu, pgmajfault: %llu, inactive_anon: %llu, active_anon: %llu, inactive_file: %llu, active_file: %llu, unevictable: %llu, hierarchical_memory_limit: %llu, total_cache: %llu, total_rss: %llu, total_rss_huge: %llu, total_mapped_file: %llu, total_writeback: %llu, total_dirty: %llu, total_swap: %llu, total_pgpgin: %llu, total_pgpgout: %llu, total_pgfault: %llu, total_pgmajfault: %llu, total_inactive_anon: %llu, total_active_anon: %llu, total_inactive_file: %llu, total_active_file: %llu, total_unevictable: %llu\n", mem->filename, mem->cache, mem->rss, mem->rss_huge, mem->mapped_file, mem->writeback, mem->dirty, mem->swap, mem->pgpgin, mem->pgpgout, mem->pgfault, mem->pgmajfault, mem->inactive_anon, mem->active_anon, mem->inactive_file, mem->active_file, mem->unevictable, mem->hierarchical_memory_limit, mem->total_cache, mem->total_rss, mem->total_rss_huge, mem->total_mapped_file, mem->total_writeback, mem->total_dirty, mem->total_swap, mem->total_pgpgin, mem->total_pgpgout, mem->total_pgfault, mem->total_pgmajfault, mem->total_inactive_anon, mem->total_active_anon, mem->total_inactive_file, mem->total_active_file, mem->total_unevictable);
- mem->updated = 1;
- }
+ mem->updated = 1;
+ }
}
void cgroup_read(struct cgroup *cg) {
- debug(D_CGROUP, "reading metrics for cgroups '%s'", cg->id);
-
- cgroup_read_cpuacct_stat(&cg->cpuacct_stat);
- cgroup_read_cpuacct_usage(&cg->cpuacct_usage);
- cgroup_read_memory(&cg->memory);
- cgroup_read_blkio(&cg->io_service_bytes);
- cgroup_read_blkio(&cg->io_serviced);
- cgroup_read_blkio(&cg->throttle_io_service_bytes);
- cgroup_read_blkio(&cg->throttle_io_serviced);
- cgroup_read_blkio(&cg->io_merged);
- cgroup_read_blkio(&cg->io_queued);
+ debug(D_CGROUP, "reading metrics for cgroups '%s'", cg->id);
+
+ cgroup_read_cpuacct_stat(&cg->cpuacct_stat);
+ cgroup_read_cpuacct_usage(&cg->cpuacct_usage);
+ cgroup_read_memory(&cg->memory);
+ cgroup_read_blkio(&cg->io_service_bytes);
+ cgroup_read_blkio(&cg->io_serviced);
+ cgroup_read_blkio(&cg->throttle_io_service_bytes);
+ cgroup_read_blkio(&cg->throttle_io_serviced);
+ cgroup_read_blkio(&cg->io_merged);
+ cgroup_read_blkio(&cg->io_queued);
}
void read_all_cgroups(struct cgroup *root) {
- debug(D_CGROUP, "reading metrics for all cgroups");
+ debug(D_CGROUP, "reading metrics for all cgroups");
- struct cgroup *cg;
+ struct cgroup *cg;
- for(cg = root; cg ; cg = cg->next)
- if(cg->enabled && cg->available)
- cgroup_read(cg);
+ for(cg = root; cg ; cg = cg->next)
+ if(cg->enabled && cg->available)
+ cgroup_read(cg);
}
// ----------------------------------------------------------------------------
@@ -584,177 +585,189 @@ void read_all_cgroups(struct cgroup *root) {
#define CGROUP_CHARTID_LINE_MAX 1024
void cgroup_get_chart_id(struct cgroup *cg) {
- debug(D_CGROUP, "getting the name of cgroup '%s'", cg->id);
-
- pid_t cgroup_pid;
- char buffer[CGROUP_CHARTID_LINE_MAX + 1];
-
- snprintfz(buffer, CGROUP_CHARTID_LINE_MAX, "exec %s '%s'",
- config_get("plugin:cgroups", "script to get cgroup names", PLUGINS_DIR "/cgroup-name.sh"), cg->chart_id);
-
- debug(D_CGROUP, "executing command '%s' for cgroup '%s'", buffer, cg->id);
- FILE *fp = mypopen(buffer, &cgroup_pid);
- if(!fp) {
- error("CGROUP: Cannot popen(\"%s\", \"r\").", buffer);
- return;
- }
- debug(D_CGROUP, "reading from command '%s' for cgroup '%s'", buffer, cg->id);
- char *s = fgets(buffer, CGROUP_CHARTID_LINE_MAX, fp);
- debug(D_CGROUP, "closing command for cgroup '%s'", cg->id);
- mypclose(fp, cgroup_pid);
- debug(D_CGROUP, "closed command for cgroup '%s'", cg->id);
-
- if(s && *s && *s != '\n') {
- debug(D_CGROUP, "cgroup '%s' should be renamed to '%s'", cg->id, s);
-
- trim(s);
-
- free(cg->chart_title);
- cg->chart_title = strdup(s);
- if(!cg->chart_title)
- fatal("CGROUP: Cannot allocate memory for chart name of cgroup '%s' chart name: '%s'", cg->id, s);
-
- netdata_fix_chart_name(cg->chart_title);
-
- free(cg->chart_id);
- cg->chart_id = strdup(s);
- if(!cg->chart_id)
- fatal("CGROUP: Cannot allocate memory for chart id of cgroup '%s' chart id: '%s'", cg->id, s);
-
- netdata_fix_chart_id(cg->chart_id);
-
- debug(D_CGROUP, "cgroup '%s' renamed to '%s' (title: '%s')", cg->id, cg->chart_id, cg->chart_title);
- }
- else debug(D_CGROUP, "cgroup '%s' is not to be renamed (will be shown as '%s')", cg->id, cg->chart_id);
+ debug(D_CGROUP, "getting the name of cgroup '%s'", cg->id);
+
+ pid_t cgroup_pid;
+ char buffer[CGROUP_CHARTID_LINE_MAX + 1];
+
+ snprintfz(buffer, CGROUP_CHARTID_LINE_MAX, "exec %s '%s'",
+ config_get("plugin:cgroups", "script to get cgroup names", PLUGINS_DIR "/cgroup-name.sh"), cg->chart_id);
+
+ debug(D_CGROUP, "executing command '%s' for cgroup '%s'", buffer, cg->id);
+ FILE *fp = mypopen(buffer, &cgroup_pid);
+ if(!fp) {
+ error("CGROUP: Cannot popen(\"%s\", \"r\").", buffer);
+ return;
+ }
+ debug(D_CGROUP, "reading from command '%s' for cgroup '%s'", buffer, cg->id);
+ char *s = fgets(buffer, CGROUP_CHARTID_LINE_MAX, fp);
+ debug(D_CGROUP, "closing command for cgroup '%s'", cg->id);
+ mypclose(fp, cgroup_pid);
+ debug(D_CGROUP, "closed command for cgroup '%s'", cg->id);
+
+ if(s && *s && *s != '\n') {
+ debug(D_CGROUP, "cgroup '%s' should be renamed to '%s'", cg->id, s);
+
+ trim(s);
+
+ freez(cg->chart_title);
+ cg->chart_title = strdupz(s);
+ netdata_fix_chart_name(cg->chart_title);
+
+ freez(cg->chart_id);
+ cg->chart_id = strdupz(s);
+ netdata_fix_chart_id(cg->chart_id);
+ cg->hash_chart = simple_hash(cg->chart_id);
+
+ debug(D_CGROUP, "cgroup '%s' renamed to '%s' (title: '%s')", cg->id, cg->chart_id, cg->chart_title);
+ }
+ else debug(D_CGROUP, "cgroup '%s' is not to be renamed (will be shown as '%s')", cg->id, cg->chart_id);
}
struct cgroup *cgroup_add(const char *id) {
- debug(D_CGROUP, "adding cgroup '%s'", id);
-
- if(cgroup_root_count >= cgroup_root_max) {
- info("Maximum number of cgroups reached (%d). Not adding cgroup '%s'", cgroup_root_count, id);
- return NULL;
- }
-
- int def = cgroup_enable_new_cgroups_detected_at_runtime;
- const char *chart_id = id;
- if(!*chart_id) {
- chart_id = "/";
-
- // disable by default the root cgroup
- def = 0;
- debug(D_CGROUP, "cgroup '%s' is the root container (by default %s)", id, (def)?"enabled":"disabled");
- }
- else {
- if(*chart_id == '/') chart_id++;
-
- size_t len = strlen(chart_id);
-
- // disable by default the parent cgroup
- // for known cgroup managers
- if(!strcmp(chart_id, "lxc") ||
- !strcmp(chart_id, "docker") ||
- !strcmp(chart_id, "libvirt") ||
- !strcmp(chart_id, "qemu") ||
- !strcmp(chart_id, "systemd") ||
- !strcmp(chart_id, "system.slice") ||
- !strcmp(chart_id, "machine.slice") ||
- !strcmp(chart_id, "user") ||
- !strcmp(chart_id, "system") ||
- !strcmp(chart_id, "machine") ||
- // starts with them
- (len > 6 && !strncmp(chart_id, "user/", 6)) ||
- (len > 11 && !strncmp(chart_id, "user.slice/", 11)) ||
- // ends with them
- (len > 5 && !strncmp(&chart_id[len - 5], ".user", 5)) ||
- (len > 5 && !strncmp(&chart_id[len - 5], ".swap", 5)) ||
- (len > 6 && !strncmp(&chart_id[len - 6], ".slice", 6)) ||
- (len > 6 && !strncmp(&chart_id[len - 6], ".mount", 6)) ||
- (len > 8 && !strncmp(&chart_id[len - 8], ".session", 8)) ||
- (len > 8 && !strncmp(&chart_id[len - 8], ".service", 8)) ||
- (len > 10 && !strncmp(&chart_id[len - 10], ".partition", 10))
- ) {
- def = 0;
- debug(D_CGROUP, "cgroup '%s' is %s (by default)", id, (def)?"enabled":"disabled");
- }
- }
-
- struct cgroup *cg = calloc(1, sizeof(struct cgroup));
- if(!cg) fatal("Cannot allocate memory for cgroup '%s'", id);
-
- debug(D_CGROUP, "adding cgroup '%s'", id);
-
- cg->id = strdup(id);
- if(!cg->id) fatal("Cannot allocate memory for cgroup '%s'", id);
-
- cg->hash = simple_hash(cg->id);
-
- cg->chart_id = strdup(chart_id);
- if(!cg->chart_id) fatal("Cannot allocate memory for cgroup '%s'", id);
-
- cg->chart_title = strdup(chart_id);
- if(!cg->chart_title) fatal("Cannot allocate memory for cgroup '%s'", id);
-
- if(!cgroup_root)
- cgroup_root = cg;
- else {
- // append it
- struct cgroup *e;
- for(e = cgroup_root; e->next ;e = e->next) ;
- e->next = cg;
- }
-
- cgroup_root_count++;
-
- // fix the name by calling the external script
- cgroup_get_chart_id(cg);
-
- char option[FILENAME_MAX + 1];
- snprintfz(option, FILENAME_MAX, "enable cgroup %s", cg->chart_title);
- cg->enabled = config_get_boolean("plugin:cgroups", option, def);
-
- debug(D_CGROUP, "Added cgroup '%s' with chart id '%s' and title '%s' as %s (default was %s)", cg->id, cg->chart_id, cg->chart_title, (cg->enabled)?"enabled":"disabled", (def)?"enabled":"disabled");
-
- return cg;
+ debug(D_CGROUP, "adding cgroup '%s'", id);
+
+ if(cgroup_root_count >= cgroup_root_max) {
+ info("Maximum number of cgroups reached (%d). Not adding cgroup '%s'", cgroup_root_count, id);
+ return NULL;
+ }
+
+ int def = cgroup_enable_new_cgroups_detected_at_runtime;
+ const char *chart_id = id;
+ if(!*chart_id) {
+ chart_id = "/";
+
+ // disable by default the root cgroup
+ def = 0;
+ debug(D_CGROUP, "cgroup '%s' is the root container (by default %s)", id, (def)?"enabled":"disabled");
+ }
+ else {
+ if(*chart_id == '/') chart_id++;
+
+ size_t len = strlen(chart_id);
+
+ // disable by default the parent cgroup
+ // for known cgroup managers
+ if(!strcmp(chart_id, "lxc") ||
+ !strcmp(chart_id, "docker") ||
+ !strcmp(chart_id, "libvirt") ||
+ !strcmp(chart_id, "qemu") ||
+ !strcmp(chart_id, "systemd") ||
+ !strcmp(chart_id, "system.slice") ||
+ !strcmp(chart_id, "machine.slice") ||
+ !strcmp(chart_id, "init.scope") ||
+ !strcmp(chart_id, "user") ||
+ !strcmp(chart_id, "system") ||
+ !strcmp(chart_id, "machine") ||
+ // starts with them
+ (len > 6 && !strncmp(chart_id, "user/", 6)) ||
+ (len > 11 && !strncmp(chart_id, "user.slice/", 11)) ||
+ // ends with them
+ (len > 5 && !strncmp(&chart_id[len - 5], ".user", 5)) ||
+ (len > 5 && !strncmp(&chart_id[len - 5], ".swap", 5)) ||
+ (len > 6 && !strncmp(&chart_id[len - 6], ".slice", 6)) ||
+ (len > 6 && !strncmp(&chart_id[len - 6], ".mount", 6)) ||
+ (len > 8 && !strncmp(&chart_id[len - 8], ".session", 8)) ||
+ (len > 8 && !strncmp(&chart_id[len - 8], ".service", 8)) ||
+ (len > 10 && !strncmp(&chart_id[len - 10], ".partition", 10))
+ ) {
+ def = 0;
+ debug(D_CGROUP, "cgroup '%s' is %s (by default)", id, (def)?"enabled":"disabled");
+ }
+ }
+
+ struct cgroup *cg = callocz(1, sizeof(struct cgroup));
+
+ cg->id = strdupz(id);
+ cg->hash = simple_hash(cg->id);
+
+ cg->chart_id = strdupz(chart_id);
+ netdata_fix_chart_id(cg->chart_id);
+ cg->hash_chart = simple_hash(cg->chart_id);
+
+ cg->chart_title = strdupz(chart_id);
+
+ if(!cgroup_root)
+ cgroup_root = cg;
+ else {
+ // append it
+ struct cgroup *e;
+ for(e = cgroup_root; e->next ;e = e->next) ;
+ e->next = cg;
+ }
+
+ cgroup_root_count++;
+
+ // fix the name by calling the external script
+ cgroup_get_chart_id(cg);
+
+ debug(D_CGROUP, "adding cgroup '%s' with chart id '%s'", id, chart_id);
+
+ char option[FILENAME_MAX + 1];
+ snprintfz(option, FILENAME_MAX, "enable cgroup %s", cg->chart_title);
+ cg->enabled = config_get_boolean("plugin:cgroups", option, def);
+
+ if(cg->enabled) {
+ struct cgroup *t;
+ for (t = cgroup_root; t; t = t->next) {
+ if (t != cg && t->enabled && t->hash_chart == cg->hash_chart && !strcmp(t->chart_id, cg->chart_id)) {
+ if (!strncmp(t->chart_id, "/system.slice/", 14) && !strncmp(cg->chart_id, "/init.scope/system.slice/", 25)) {
+ error("Control group with chart id '%s' already exists with id '%s' and is enabled. Swapping them by enabling cgroup with id '%s' and disabling cgroup with id '%s'.",
+ cg->chart_id, t->id, cg->id, t->id);
+ t->enabled = 0;
+ } else {
+ error("Control group with chart id '%s' already exists with id '%s' and is enabled. Disabling cgroup with id '%s'.",
+ cg->chart_id, t->id, cg->id);
+ cg->enabled = 0;
+ }
+
+ break;
+ }
+ }
+ }
+
+ debug(D_CGROUP, "Added cgroup '%s' with chart id '%s' and title '%s' as %s (default was %s)", cg->id, cg->chart_id, cg->chart_title, (cg->enabled)?"enabled":"disabled", (def)?"enabled":"disabled");
+
+ return cg;
}
void cgroup_free(struct cgroup *cg) {
- debug(D_CGROUP, "Removing cgroup '%s' with chart id '%s' (was %s and %s)", cg->id, cg->chart_id, (cg->enabled)?"enabled":"disabled", (cg->available)?"available":"not available");
-
- free(cg->cpuacct_usage.cpu_percpu);
-
- free(cg->cpuacct_stat.filename);
- free(cg->cpuacct_usage.filename);
- free(cg->memory.filename);
- free(cg->io_service_bytes.filename);
- free(cg->io_serviced.filename);
- free(cg->throttle_io_service_bytes.filename);
- free(cg->throttle_io_serviced.filename);
- free(cg->io_merged.filename);
- free(cg->io_queued.filename);
-
- free(cg->id);
- free(cg->chart_id);
- free(cg->chart_title);
- free(cg);
-
- cgroup_root_count--;
+ debug(D_CGROUP, "Removing cgroup '%s' with chart id '%s' (was %s and %s)", cg->id, cg->chart_id, (cg->enabled)?"enabled":"disabled", (cg->available)?"available":"not available");
+
+ freez(cg->cpuacct_usage.cpu_percpu);
+
+ freez(cg->cpuacct_stat.filename);
+ freez(cg->cpuacct_usage.filename);
+ freez(cg->memory.filename);
+ freez(cg->io_service_bytes.filename);
+ freez(cg->io_serviced.filename);
+ freez(cg->throttle_io_service_bytes.filename);
+ freez(cg->throttle_io_serviced.filename);
+ freez(cg->io_merged.filename);
+ freez(cg->io_queued.filename);
+
+ freez(cg->id);
+ freez(cg->chart_id);
+ freez(cg->chart_title);
+ freez(cg);
+
+ cgroup_root_count--;
}
// find if a given cgroup exists
struct cgroup *cgroup_find(const char *id) {
- debug(D_CGROUP, "searching for cgroup '%s'", id);
+ debug(D_CGROUP, "searching for cgroup '%s'", id);
- uint32_t hash = simple_hash(id);
+ uint32_t hash = simple_hash(id);
- struct cgroup *cg;
- for(cg = cgroup_root; cg ; cg = cg->next) {
- if(hash == cg->hash && strcmp(id, cg->id) == 0)
- break;
- }
+ struct cgroup *cg;
+ for(cg = cgroup_root; cg ; cg = cg->next) {
+ if(hash == cg->hash && strcmp(id, cg->id) == 0)
+ break;
+ }
- debug(D_CGROUP, "cgroup_find('%s') %s", id, (cg)?"found":"not found");
- return cg;
+ debug(D_CGROUP, "cgroup_find('%s') %s", id, (cg)?"found":"not found");
+ return cg;
}
// ----------------------------------------------------------------------------
@@ -762,198 +775,252 @@ struct cgroup *cgroup_find(const char *id) {
// callback for find_file_in_subdirs()
void found_subdir_in_dir(const char *dir) {
- debug(D_CGROUP, "examining cgroup dir '%s'", dir);
-
- struct cgroup *cg = cgroup_find(dir);
- if(!cg) {
- if(*dir && cgroup_max_depth > 0) {
- int depth = 0;
- const char *s;
-
- for(s = dir; *s ;s++)
- if(unlikely(*s == '/'))
- depth++;
-
- if(depth > cgroup_max_depth) {
- info("cgroup '%s' is too deep (%d, while max is %d)", dir, depth, cgroup_max_depth);
- return;
- }
- }
- debug(D_CGROUP, "will add dir '%s' as cgroup", dir);
- cg = cgroup_add(dir);
- }
-
- if(cg) cg->available = 1;
+ debug(D_CGROUP, "examining cgroup dir '%s'", dir);
+
+ struct cgroup *cg = cgroup_find(dir);
+ if(!cg) {
+ if(*dir && cgroup_max_depth > 0) {
+ int depth = 0;
+ const char *s;
+
+ for(s = dir; *s ;s++)
+ if(unlikely(*s == '/'))
+ depth++;
+
+ if(depth > cgroup_max_depth) {
+ info("cgroup '%s' is too deep (%d, while max is %d)", dir, depth, cgroup_max_depth);
+ return;
+ }
+ }
+ debug(D_CGROUP, "will add dir '%s' as cgroup", dir);
+ cg = cgroup_add(dir);
+ }
+
+ if(cg) cg->available = 1;
}
-void find_dir_in_subdirs(const char *base, const char *this, void (*callback)(const char *)) {
- int enabled = -1;
- if(!this) this = base;
- size_t dirlen = strlen(this), baselen = strlen(base);
- const char *relative_path = &this[baselen];
-
- DIR *dir = opendir(this);
- if(!dir) return;
-
- callback(relative_path);
-
- struct dirent *de = NULL;
- while((de = readdir(dir))) {
- if(de->d_type == DT_DIR
- && (
- (de->d_name[0] == '.' && de->d_name[1] == '\0')
- || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
- ))
- continue;
-
- debug(D_CGROUP, "examining '%s/%s'", this, de->d_name);
-
- if(de->d_type == DT_DIR) {
- if(enabled == -1) {
- const char *r = relative_path;
- if(*r == '\0') r = "/";
- else if (*r == '/') r++;
-
- // we check for this option here
- // so that the config will not have settings
- // for leaf directories
- char option[FILENAME_MAX + 1];
- snprintfz(option, FILENAME_MAX, "search for cgroups under %s", r);
- option[FILENAME_MAX] = '\0';
- enabled = config_get_boolean("plugin:cgroups", option, 1);
- }
-
- if(enabled) {
- char *s = malloc(dirlen + strlen(de->d_name) + 2);
- if(s) {
- strcpy(s, this);
- strcat(s, "/");
- strcat(s, de->d_name);
- find_dir_in_subdirs(base, s, callback);
- free(s);
- }
- }
- }
- }
-
- closedir(dir);
+int find_dir_in_subdirs(const char *base, const char *this, void (*callback)(const char *)) {
+ debug(D_CGROUP, "searching for directories in '%s'", base);
+
+ int ret = -1;
+ int enabled = -1;
+ if(!this) this = base;
+ size_t dirlen = strlen(this), baselen = strlen(base);
+ const char *relative_path = &this[baselen];
+
+ DIR *dir = opendir(this);
+ if(!dir) {
+ error("Cannot read cgroups directory '%s'", base);
+ return ret;
+ }
+ ret = 1;
+
+ callback(relative_path);
+
+ struct dirent *de = NULL;
+ while((de = readdir(dir))) {
+ if(de->d_type == DT_DIR
+ && (
+ (de->d_name[0] == '.' && de->d_name[1] == '\0')
+ || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
+ ))
+ continue;
+
+ debug(D_CGROUP, "examining '%s/%s'", this, de->d_name);
+
+ if(de->d_type == DT_DIR) {
+ if(enabled == -1) {
+ const char *r = relative_path;
+ if(*r == '\0') r = "/";
+ else if (*r == '/') r++;
+
+ // we check for this option here
+ // so that the config will not have settings
+ // for leaf directories
+ char option[FILENAME_MAX + 1];
+ snprintfz(option, FILENAME_MAX, "search for cgroups under %s", r);
+ option[FILENAME_MAX] = '\0';
+ enabled = config_get_boolean("plugin:cgroups", option, 1);
+ }
+
+ if(enabled) {
+ char *s = mallocz(dirlen + strlen(de->d_name) + 2);
+ strcpy(s, this);
+ strcat(s, "/");
+ strcat(s, de->d_name);
+ int ret2 = find_dir_in_subdirs(base, s, callback);
+ if(ret2 > 0) ret += ret2;
+ freez(s);
+ }
+ }
+ }
+
+ closedir(dir);
+ return ret;
}
void mark_all_cgroups_as_not_available() {
- debug(D_CGROUP, "marking all cgroups as not available");
+ debug(D_CGROUP, "marking all cgroups as not available");
- struct cgroup *cg;
+ struct cgroup *cg;
- // mark all as not available
- for(cg = cgroup_root; cg ; cg = cg->next)
- cg->available = 0;
+ // mark all as not available
+ for(cg = cgroup_root; cg ; cg = cg->next)
+ cg->available = 0;
}
void cleanup_all_cgroups() {
- struct cgroup *cg = cgroup_root, *last = NULL;
-
- for(; cg ;) {
- if(!cg->available) {
-
- if(!last)
- cgroup_root = cg->next;
- else
- last->next = cg->next;
-
- cgroup_free(cg);
-
- if(!last)
- cg = cgroup_root;
- else
- cg = last->next;
- }
- else {
- last = cg;
- cg = cg->next;
- }
- }
+ struct cgroup *cg = cgroup_root, *last = NULL;
+
+ for(; cg ;) {
+ if(!cg->available) {
+
+ if(!last)
+ cgroup_root = cg->next;
+ else
+ last->next = cg->next;
+
+ cgroup_free(cg);
+
+ if(!last)
+ cg = cgroup_root;
+ else
+ cg = last->next;
+ }
+ else {
+ last = cg;
+ cg = cg->next;
+ }
+ }
}
void find_all_cgroups() {
- debug(D_CGROUP, "searching for cgroups");
-
- mark_all_cgroups_as_not_available();
-
- if(cgroup_enable_cpuacct_stat || cgroup_enable_cpuacct_usage)
- find_dir_in_subdirs(cgroup_cpuacct_base, NULL, found_subdir_in_dir);
-
- if(cgroup_enable_blkio)
- find_dir_in_subdirs(cgroup_blkio_base, NULL, found_subdir_in_dir);
-
- if(cgroup_enable_memory)
- find_dir_in_subdirs(cgroup_memory_base, NULL, found_subdir_in_dir);
-
- // remove any non-existing cgroups
- cleanup_all_cgroups();
-
- struct cgroup *cg;
- for(cg = cgroup_root; cg ; cg = cg->next) {
- // fprintf(stderr, " >>> CGROUP '%s' (%u - %s) with name '%s'\n", cg->id, cg->hash, cg->available?"available":"stopped", cg->name);
-
- if(unlikely(!cg->available))
- continue;
-
- debug(D_CGROUP, "checking paths for cgroup '%s'", cg->id);
-
- // check for newly added cgroups
- // and update the filenames they read
- char filename[FILENAME_MAX + 1];
- if(cgroup_enable_cpuacct_stat && !cg->cpuacct_stat.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/cpuacct.stat", cgroup_cpuacct_base, cg->id);
- cg->cpuacct_stat.filename = strdup(filename);
- debug(D_CGROUP, "cpuacct.stat filename for cgroup '%s': '%s'", cg->id, cg->cpuacct_stat.filename);
- }
- if(cgroup_enable_cpuacct_usage && !cg->cpuacct_usage.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/cpuacct.usage_percpu", cgroup_cpuacct_base, cg->id);
- cg->cpuacct_usage.filename = strdup(filename);
- debug(D_CGROUP, "cpuacct.usage_percpu filename for cgroup '%s': '%s'", cg->id, cg->cpuacct_usage.filename);
- }
- if(cgroup_enable_memory && !cg->memory.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/memory.stat", cgroup_memory_base, cg->id);
- cg->memory.filename = strdup(filename);
- debug(D_CGROUP, "memory.stat filename for cgroup '%s': '%s'", cg->id, cg->memory.filename);
- }
- if(cgroup_enable_blkio) {
- if(!cg->io_service_bytes.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_service_bytes", cgroup_blkio_base, cg->id);
- cg->io_service_bytes.filename = strdup(filename);
- debug(D_CGROUP, "io_service_bytes filename for cgroup '%s': '%s'", cg->id, cg->io_service_bytes.filename);
- }
- if(!cg->io_serviced.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_serviced", cgroup_blkio_base, cg->id);
- cg->io_serviced.filename = strdup(filename);
- debug(D_CGROUP, "io_serviced filename for cgroup '%s': '%s'", cg->id, cg->io_serviced.filename);
- }
- if(!cg->throttle_io_service_bytes.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.throttle.io_service_bytes", cgroup_blkio_base, cg->id);
- cg->throttle_io_service_bytes.filename = strdup(filename);
- debug(D_CGROUP, "throttle_io_service_bytes filename for cgroup '%s': '%s'", cg->id, cg->throttle_io_service_bytes.filename);
- }
- if(!cg->throttle_io_serviced.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.throttle.io_serviced", cgroup_blkio_base, cg->id);
- cg->throttle_io_serviced.filename = strdup(filename);
- debug(D_CGROUP, "throttle_io_serviced filename for cgroup '%s': '%s'", cg->id, cg->throttle_io_serviced.filename);
- }
- if(!cg->io_merged.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_merged", cgroup_blkio_base, cg->id);
- cg->io_merged.filename = strdup(filename);
- debug(D_CGROUP, "io_merged filename for cgroup '%s': '%s'", cg->id, cg->io_merged.filename);
- }
- if(!cg->io_queued.filename) {
- snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_queued", cgroup_blkio_base, cg->id);
- cg->io_queued.filename = strdup(filename);
- debug(D_CGROUP, "io_queued filename for cgroup '%s': '%s'", cg->id, cg->io_queued.filename);
- }
- }
- }
-
- debug(D_CGROUP, "done searching for cgroups");
- return;
+ debug(D_CGROUP, "searching for cgroups");
+
+ mark_all_cgroups_as_not_available();
+
+ if(cgroup_enable_cpuacct_stat || cgroup_enable_cpuacct_usage) {
+ if (find_dir_in_subdirs(cgroup_cpuacct_base, NULL, found_subdir_in_dir) == -1) {
+ cgroup_enable_cpuacct_stat = cgroup_enable_cpuacct_usage = 0;
+ error("disabled cgroup cpu statistics.");
+ }
+ }
+
+ if(cgroup_enable_blkio) {
+ if (find_dir_in_subdirs(cgroup_blkio_base, NULL, found_subdir_in_dir) == -1) {
+ cgroup_enable_blkio = 0;
+ error("disabled cgroup blkio statistics.");
+ }
+ }
+
+ if(cgroup_enable_memory) {
+ if(find_dir_in_subdirs(cgroup_memory_base, NULL, found_subdir_in_dir) == -1) {
+ cgroup_enable_memory = 0;
+ error("disabled cgroup memory statistics.");
+ }
+ }
+
+ if(cgroup_enable_devices) {
+ if(find_dir_in_subdirs(cgroup_devices_base, NULL, found_subdir_in_dir) == -1) {
+ cgroup_enable_devices = 0;
+ error("disabled cgroup devices statistics.");
+ }
+ }
+
+ // remove any non-existing cgroups
+ cleanup_all_cgroups();
+
+ struct cgroup *cg;
+ struct stat buf;
+ for(cg = cgroup_root; cg ; cg = cg->next) {
+ // fprintf(stderr, " >>> CGROUP '%s' (%u - %s) with name '%s'\n", cg->id, cg->hash, cg->available?"available":"stopped", cg->name);
+
+ if(unlikely(!cg->available))
+ continue;
+
+ debug(D_CGROUP, "checking paths for cgroup '%s'", cg->id);
+
+ // check for newly added cgroups
+ // and update the filenames they read
+ char filename[FILENAME_MAX + 1];
+ if(cgroup_enable_cpuacct_stat && !cg->cpuacct_stat.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/cpuacct.stat", cgroup_cpuacct_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->cpuacct_stat.filename = strdupz(filename);
+ debug(D_CGROUP, "cpuacct.stat filename for cgroup '%s': '%s'", cg->id, cg->cpuacct_stat.filename);
+ }
+ else debug(D_CGROUP, "cpuacct.stat file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(cgroup_enable_cpuacct_usage && !cg->cpuacct_usage.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/cpuacct.usage_percpu", cgroup_cpuacct_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->cpuacct_usage.filename = strdupz(filename);
+ debug(D_CGROUP, "cpuacct.usage_percpu filename for cgroup '%s': '%s'", cg->id, cg->cpuacct_usage.filename);
+ }
+ else debug(D_CGROUP, "cpuacct.usage_percpu file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(cgroup_enable_memory && !cg->memory.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/memory.stat", cgroup_memory_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->memory.filename = strdupz(filename);
+ debug(D_CGROUP, "memory.stat filename for cgroup '%s': '%s'", cg->id, cg->memory.filename);
+ }
+ else debug(D_CGROUP, "memory.stat file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(cgroup_enable_blkio) {
+ if(!cg->io_service_bytes.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_service_bytes", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->io_service_bytes.filename = strdupz(filename);
+ debug(D_CGROUP, "io_service_bytes filename for cgroup '%s': '%s'", cg->id, cg->io_service_bytes.filename);
+ }
+ else debug(D_CGROUP, "io_service_bytes file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(!cg->io_serviced.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_serviced", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->io_serviced.filename = strdupz(filename);
+ debug(D_CGROUP, "io_serviced filename for cgroup '%s': '%s'", cg->id, cg->io_serviced.filename);
+ }
+ else debug(D_CGROUP, "io_serviced file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(!cg->throttle_io_service_bytes.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.throttle.io_service_bytes", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->throttle_io_service_bytes.filename = strdupz(filename);
+ debug(D_CGROUP, "throttle_io_service_bytes filename for cgroup '%s': '%s'", cg->id, cg->throttle_io_service_bytes.filename);
+ }
+ else debug(D_CGROUP, "throttle_io_service_bytes file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(!cg->throttle_io_serviced.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.throttle.io_serviced", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->throttle_io_serviced.filename = strdupz(filename);
+ debug(D_CGROUP, "throttle_io_serviced filename for cgroup '%s': '%s'", cg->id, cg->throttle_io_serviced.filename);
+ }
+ else debug(D_CGROUP, "throttle_io_serviced file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(!cg->io_merged.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_merged", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->io_merged.filename = strdupz(filename);
+ debug(D_CGROUP, "io_merged filename for cgroup '%s': '%s'", cg->id, cg->io_merged.filename);
+ }
+ else debug(D_CGROUP, "io_merged file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ if(!cg->io_queued.filename) {
+ snprintfz(filename, FILENAME_MAX, "%s%s/blkio.io_queued", cgroup_blkio_base, cg->id);
+ if(stat(filename, &buf) != -1) {
+ cg->io_queued.filename = strdupz(filename);
+ debug(D_CGROUP, "io_queued filename for cgroup '%s': '%s'", cg->id, cg->io_queued.filename);
+ }
+ else debug(D_CGROUP, "io_queued file for cgroup '%s': '%s' does not exist.", cg->id, filename);
+ }
+ }
+ }
+
+ debug(D_CGROUP, "done searching for cgroups");
+ return;
}
// ----------------------------------------------------------------------------
@@ -962,349 +1029,349 @@ void find_all_cgroups() {
#define CHART_TITLE_MAX 300
void update_cgroup_charts(int update_every) {
- debug(D_CGROUP, "updating cgroups charts");
-
- char type[RRD_ID_LENGTH_MAX + 1];
- char title[CHART_TITLE_MAX + 1];
-
- struct cgroup *cg;
- RRDSET *st;
-
- for(cg = cgroup_root; cg ; cg = cg->next) {
- if(!cg->available || !cg->enabled)
- continue;
-
- if(cg->id[0] == '\0')
- strcpy(type, "cgroup_root");
- else if(cg->id[0] == '/')
- snprintfz(type, RRD_ID_LENGTH_MAX, "cgroup_%s", cg->chart_id);
- else
- snprintfz(type, RRD_ID_LENGTH_MAX, "cgroup_%s", cg->chart_id);
-
- netdata_fix_chart_id(type);
-
- if(cg->cpuacct_stat.updated) {
- st = rrdset_find_bytype(type, "cpu");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "CPU Usage for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "cpu", NULL, "cpu", "cgroup.cpu", title, "%", 40000, update_every, RRDSET_TYPE_STACKED);
-
- rrddim_add(st, "user", NULL, 100, hz, RRDDIM_INCREMENTAL);
- rrddim_add(st, "system", NULL, 100, hz, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "user", cg->cpuacct_stat.user);
- rrddim_set(st, "system", cg->cpuacct_stat.system);
- rrdset_done(st);
- }
-
- if(cg->cpuacct_usage.updated) {
- char id[RRD_ID_LENGTH_MAX + 1];
- unsigned int i;
-
- st = rrdset_find_bytype(type, "cpu_per_core");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "CPU Usage Per Core for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "cpu_per_core", NULL, "cpu", "cgroup.cpu_per_core", title, "%", 40100, update_every, RRDSET_TYPE_STACKED);
-
- for(i = 0; i < cg->cpuacct_usage.cpus ;i++) {
- snprintfz(id, CHART_TITLE_MAX, "cpu%d", i);
- rrddim_add(st, id, NULL, 100, 1000000, RRDDIM_INCREMENTAL);
- }
- }
- else rrdset_next(st);
-
- for(i = 0; i < cg->cpuacct_usage.cpus ;i++) {
- snprintfz(id, CHART_TITLE_MAX, "cpu%d", i);
- rrddim_set(st, id, cg->cpuacct_usage.cpu_percpu[i]);
- }
- rrdset_done(st);
- }
-
- if(cg->memory.updated) {
- if(cg->memory.cache + cg->memory.rss + cg->memory.rss_huge + cg->memory.mapped_file > 0) {
- st = rrdset_find_bytype(type, "mem");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Memory Usage for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "mem", NULL, "mem", "cgroup.mem", title, "MB", 40200, update_every,
- RRDSET_TYPE_STACKED);
-
- rrddim_add(st, "cache", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- rrddim_add(st, "rss", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- if(cg->memory.has_dirty_swap)
- rrddim_add(st, "swap", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- rrddim_add(st, "rss_huge", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- rrddim_add(st, "mapped_file", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "cache", cg->memory.cache);
- rrddim_set(st, "rss", cg->memory.rss);
- if(cg->memory.has_dirty_swap)
- rrddim_set(st, "swap", cg->memory.swap);
- rrddim_set(st, "rss_huge", cg->memory.rss_huge);
- rrddim_set(st, "mapped_file", cg->memory.mapped_file);
- rrdset_done(st);
- }
-
- st = rrdset_find_bytype(type, "writeback");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Writeback Memory for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "writeback", NULL, "mem", "cgroup.writeback", title, "MB", 40300,
- update_every, RRDSET_TYPE_AREA);
-
- if(cg->memory.has_dirty_swap)
- rrddim_add(st, "dirty", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- rrddim_add(st, "writeback", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
- }
- else rrdset_next(st);
-
- if(cg->memory.has_dirty_swap)
- rrddim_set(st, "dirty", cg->memory.dirty);
- rrddim_set(st, "writeback", cg->memory.writeback);
- rrdset_done(st);
-
- if(cg->memory.pgpgin + cg->memory.pgpgout > 0) {
- st = rrdset_find_bytype(type, "mem_activity");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Memory Activity for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "mem_activity", NULL, "mem", "cgroup.mem_activity", title, "MB/s",
- 40400, update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "pgpgin", "in", sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
- rrddim_add(st, "pgpgout", "out", -sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "pgpgin", cg->memory.pgpgin);
- rrddim_set(st, "pgpgout", cg->memory.pgpgout);
- rrdset_done(st);
- }
-
- if(cg->memory.pgfault + cg->memory.pgmajfault > 0) {
- st = rrdset_find_bytype(type, "pgfaults");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Memory Page Faults for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "pgfaults", NULL, "mem", "cgroup.pgfaults", title, "MB/s", 40500,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "pgfault", NULL, sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
- rrddim_add(st, "pgmajfault", "swap", -sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "pgfault", cg->memory.pgfault);
- rrddim_set(st, "pgmajfault", cg->memory.pgmajfault);
- rrdset_done(st);
- }
- }
-
- if(cg->io_service_bytes.updated && cg->io_service_bytes.Read + cg->io_service_bytes.Write > 0) {
- st = rrdset_find_bytype(type, "io");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "I/O Bandwidth (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "io", NULL, "disk", "cgroup.io", title, "KB/s", 41200,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
- rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->io_service_bytes.Read);
- rrddim_set(st, "write", cg->io_service_bytes.Write);
- rrdset_done(st);
- }
-
- if(cg->io_serviced.updated && cg->io_serviced.Read + cg->io_serviced.Write > 0) {
- st = rrdset_find_bytype(type, "serviced_ops");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Serviced I/O Operations (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "serviced_ops", NULL, "disk", "cgroup.serviced_ops", title, "operations/s", 41200,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1, RRDDIM_INCREMENTAL);
- rrddim_add(st, "write", NULL, -1, 1, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->io_serviced.Read);
- rrddim_set(st, "write", cg->io_serviced.Write);
- rrdset_done(st);
- }
-
- if(cg->throttle_io_service_bytes.updated && cg->throttle_io_service_bytes.Read + cg->throttle_io_service_bytes.Write > 0) {
- st = rrdset_find_bytype(type, "io");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Throttle I/O Bandwidth (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "io", NULL, "disk", "cgroup.io", title, "KB/s", 41200,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
- rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->throttle_io_service_bytes.Read);
- rrddim_set(st, "write", cg->throttle_io_service_bytes.Write);
- rrdset_done(st);
- }
-
-
- if(cg->throttle_io_serviced.updated && cg->throttle_io_serviced.Read + cg->throttle_io_serviced.Write > 0) {
- st = rrdset_find_bytype(type, "throttle_serviced_ops");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Throttle Serviced I/O Operations (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "throttle_serviced_ops", NULL, "disk", "cgroup.throttle_serviced_ops", title, "operations/s", 41200,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1, RRDDIM_INCREMENTAL);
- rrddim_add(st, "write", NULL, -1, 1, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->throttle_io_serviced.Read);
- rrddim_set(st, "write", cg->throttle_io_serviced.Write);
- rrdset_done(st);
- }
-
- if(cg->io_queued.updated) {
- st = rrdset_find_bytype(type, "queued_ops");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Queued I/O Operations (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "queued_ops", NULL, "disk", "cgroup.queued_ops", title, "operations", 42000,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1, RRDDIM_ABSOLUTE);
- rrddim_add(st, "write", NULL, -1, 1, RRDDIM_ABSOLUTE);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->io_queued.Read);
- rrddim_set(st, "write", cg->io_queued.Write);
- rrdset_done(st);
- }
-
- if(cg->io_merged.updated && cg->io_merged.Read + cg->io_merged.Write > 0) {
- st = rrdset_find_bytype(type, "merged_ops");
- if(!st) {
- snprintfz(title, CHART_TITLE_MAX, "Merged I/O Operations (all disks) for cgroup %s", cg->chart_title);
- st = rrdset_create(type, "merged_ops", NULL, "disk", "cgroup.merged_ops", title, "operations/s", 42100,
- update_every, RRDSET_TYPE_LINE);
-
- rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
- rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(st);
-
- rrddim_set(st, "read", cg->io_merged.Read);
- rrddim_set(st, "write", cg->io_merged.Write);
- rrdset_done(st);
- }
- }
-
- debug(D_CGROUP, "done updating cgroups charts");
+ debug(D_CGROUP, "updating cgroups charts");
+
+ char type[RRD_ID_LENGTH_MAX + 1];
+ char title[CHART_TITLE_MAX + 1];
+
+ struct cgroup *cg;
+ RRDSET *st;
+
+ for(cg = cgroup_root; cg ; cg = cg->next) {
+ if(!cg->available || !cg->enabled)
+ continue;
+
+ if(cg->id[0] == '\0')
+ strcpy(type, "cgroup_root");
+ else if(cg->id[0] == '/')
+ snprintfz(type, RRD_ID_LENGTH_MAX, "cgroup_%s", cg->chart_id);
+ else
+ snprintfz(type, RRD_ID_LENGTH_MAX, "cgroup_%s", cg->chart_id);
+
+ netdata_fix_chart_id(type);
+
+ if(cg->cpuacct_stat.updated) {
+ st = rrdset_find_bytype(type, "cpu");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "CPU Usage for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "cpu", NULL, "cpu", "cgroup.cpu", title, "%", 40000, update_every, RRDSET_TYPE_STACKED);
+
+ rrddim_add(st, "user", NULL, 100, hz, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "system", NULL, 100, hz, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "user", cg->cpuacct_stat.user);
+ rrddim_set(st, "system", cg->cpuacct_stat.system);
+ rrdset_done(st);
+ }
+
+ if(cg->cpuacct_usage.updated) {
+ char id[RRD_ID_LENGTH_MAX + 1];
+ unsigned int i;
+
+ st = rrdset_find_bytype(type, "cpu_per_core");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "CPU Usage Per Core for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "cpu_per_core", NULL, "cpu", "cgroup.cpu_per_core", title, "%", 40100, update_every, RRDSET_TYPE_STACKED);
+
+ for(i = 0; i < cg->cpuacct_usage.cpus ;i++) {
+ snprintfz(id, CHART_TITLE_MAX, "cpu%u", i);
+ rrddim_add(st, id, NULL, 100, 1000000000, RRDDIM_INCREMENTAL);
+ }
+ }
+ else rrdset_next(st);
+
+ for(i = 0; i < cg->cpuacct_usage.cpus ;i++) {
+ snprintfz(id, CHART_TITLE_MAX, "cpu%u", i);
+ rrddim_set(st, id, cg->cpuacct_usage.cpu_percpu[i]);
+ }
+ rrdset_done(st);
+ }
+
+ if(cg->memory.updated) {
+ if(cg->memory.cache + cg->memory.rss + cg->memory.rss_huge + cg->memory.mapped_file > 0) {
+ st = rrdset_find_bytype(type, "mem");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Memory Usage for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "mem", NULL, "mem", "cgroup.mem", title, "MB", 40200, update_every,
+ RRDSET_TYPE_STACKED);
+
+ rrddim_add(st, "cache", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ rrddim_add(st, "rss", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ if(cg->memory.has_dirty_swap)
+ rrddim_add(st, "swap", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ rrddim_add(st, "rss_huge", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ rrddim_add(st, "mapped_file", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "cache", cg->memory.cache);
+ rrddim_set(st, "rss", cg->memory.rss);
+ if(cg->memory.has_dirty_swap)
+ rrddim_set(st, "swap", cg->memory.swap);
+ rrddim_set(st, "rss_huge", cg->memory.rss_huge);
+ rrddim_set(st, "mapped_file", cg->memory.mapped_file);
+ rrdset_done(st);
+ }
+
+ st = rrdset_find_bytype(type, "writeback");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Writeback Memory for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "writeback", NULL, "mem", "cgroup.writeback", title, "MB", 40300,
+ update_every, RRDSET_TYPE_AREA);
+
+ if(cg->memory.has_dirty_swap)
+ rrddim_add(st, "dirty", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ rrddim_add(st, "writeback", NULL, 1, 1024 * 1024, RRDDIM_ABSOLUTE);
+ }
+ else rrdset_next(st);
+
+ if(cg->memory.has_dirty_swap)
+ rrddim_set(st, "dirty", cg->memory.dirty);
+ rrddim_set(st, "writeback", cg->memory.writeback);
+ rrdset_done(st);
+
+ if(cg->memory.pgpgin + cg->memory.pgpgout > 0) {
+ st = rrdset_find_bytype(type, "mem_activity");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Memory Activity for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "mem_activity", NULL, "mem", "cgroup.mem_activity", title, "MB/s",
+ 40400, update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "pgpgin", "in", sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "pgpgout", "out", -sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "pgpgin", cg->memory.pgpgin);
+ rrddim_set(st, "pgpgout", cg->memory.pgpgout);
+ rrdset_done(st);
+ }
+
+ if(cg->memory.pgfault + cg->memory.pgmajfault > 0) {
+ st = rrdset_find_bytype(type, "pgfaults");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Memory Page Faults for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "pgfaults", NULL, "mem", "cgroup.pgfaults", title, "MB/s", 40500,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "pgfault", NULL, sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "pgmajfault", "swap", -sysconf(_SC_PAGESIZE), 1024 * 1024, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "pgfault", cg->memory.pgfault);
+ rrddim_set(st, "pgmajfault", cg->memory.pgmajfault);
+ rrdset_done(st);
+ }
+ }
+
+ if(cg->io_service_bytes.updated && cg->io_service_bytes.Read + cg->io_service_bytes.Write > 0) {
+ st = rrdset_find_bytype(type, "io");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "I/O Bandwidth (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "io", NULL, "disk", "cgroup.io", title, "KB/s", 41200,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->io_service_bytes.Read);
+ rrddim_set(st, "write", cg->io_service_bytes.Write);
+ rrdset_done(st);
+ }
+
+ if(cg->io_serviced.updated && cg->io_serviced.Read + cg->io_serviced.Write > 0) {
+ st = rrdset_find_bytype(type, "serviced_ops");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Serviced I/O Operations (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "serviced_ops", NULL, "disk", "cgroup.serviced_ops", title, "operations/s", 41200,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "write", NULL, -1, 1, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->io_serviced.Read);
+ rrddim_set(st, "write", cg->io_serviced.Write);
+ rrdset_done(st);
+ }
+
+ if(cg->throttle_io_service_bytes.updated && cg->throttle_io_service_bytes.Read + cg->throttle_io_service_bytes.Write > 0) {
+ st = rrdset_find_bytype(type, "io");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Throttle I/O Bandwidth (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "io", NULL, "disk", "cgroup.io", title, "KB/s", 41200,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->throttle_io_service_bytes.Read);
+ rrddim_set(st, "write", cg->throttle_io_service_bytes.Write);
+ rrdset_done(st);
+ }
+
+
+ if(cg->throttle_io_serviced.updated && cg->throttle_io_serviced.Read + cg->throttle_io_serviced.Write > 0) {
+ st = rrdset_find_bytype(type, "throttle_serviced_ops");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Throttle Serviced I/O Operations (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "throttle_serviced_ops", NULL, "disk", "cgroup.throttle_serviced_ops", title, "operations/s", 41200,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "write", NULL, -1, 1, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->throttle_io_serviced.Read);
+ rrddim_set(st, "write", cg->throttle_io_serviced.Write);
+ rrdset_done(st);
+ }
+
+ if(cg->io_queued.updated) {
+ st = rrdset_find_bytype(type, "queued_ops");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Queued I/O Operations (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "queued_ops", NULL, "disk", "cgroup.queued_ops", title, "operations", 42000,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1, RRDDIM_ABSOLUTE);
+ rrddim_add(st, "write", NULL, -1, 1, RRDDIM_ABSOLUTE);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->io_queued.Read);
+ rrddim_set(st, "write", cg->io_queued.Write);
+ rrdset_done(st);
+ }
+
+ if(cg->io_merged.updated && cg->io_merged.Read + cg->io_merged.Write > 0) {
+ st = rrdset_find_bytype(type, "merged_ops");
+ if(!st) {
+ snprintfz(title, CHART_TITLE_MAX, "Merged I/O Operations (all disks) for cgroup %s", cg->chart_title);
+ st = rrdset_create(type, "merged_ops", NULL, "disk", "cgroup.merged_ops", title, "operations/s", 42100,
+ update_every, RRDSET_TYPE_LINE);
+
+ rrddim_add(st, "read", NULL, 1, 1024, RRDDIM_INCREMENTAL);
+ rrddim_add(st, "write", NULL, -1, 1024, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(st);
+
+ rrddim_set(st, "read", cg->io_merged.Read);
+ rrddim_set(st, "write", cg->io_merged.Write);
+ rrdset_done(st);
+ }
+ }
+
+ debug(D_CGROUP, "done updating cgroups charts");
}
// ----------------------------------------------------------------------------
// cgroups main
int do_sys_fs_cgroup(int update_every, unsigned long long dt) {
- static int cgroup_global_config_read = 0;
- static time_t last_run = 0;
- time_t now = time(NULL);
+ (void)dt;
- if(dt) {};
+ static int cgroup_global_config_read = 0;
+ static time_t last_run = 0;
+ time_t now = time(NULL);
- if(unlikely(!cgroup_global_config_read)) {
- read_cgroup_plugin_configuration();
- cgroup_global_config_read = 1;
- }
+ if(unlikely(!cgroup_global_config_read)) {
+ read_cgroup_plugin_configuration();
+ cgroup_global_config_read = 1;
+ }
- if(unlikely(cgroup_enable_new_cgroups_detected_at_runtime && now - last_run > cgroup_check_for_new_every)) {
- find_all_cgroups();
- last_run = now;
- }
+ if(unlikely(cgroup_enable_new_cgroups_detected_at_runtime && now - last_run > cgroup_check_for_new_every)) {
+ find_all_cgroups();
+ last_run = now;
+ }
- read_all_cgroups(cgroup_root);
- update_cgroup_charts(update_every);
+ read_all_cgroups(cgroup_root);
+ update_cgroup_charts(update_every);
- return 0;
+ return 0;
}
void *cgroups_main(void *ptr)
{
- if(ptr) { ; }
+ if(ptr) { ; }
- info("CGROUP Plugin thread created with task id %d", gettid());
+ info("CGROUP Plugin thread created with task id %d", gettid());
- if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
- error("Cannot set pthread cancel type to DEFERRED.");
+ if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
+ error("Cannot set pthread cancel type to DEFERRED.");
- if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
- error("Cannot set pthread cancel state to ENABLE.");
+ if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
+ error("Cannot set pthread cancel state to ENABLE.");
- struct rusage thread;
+ struct rusage thread;
- // when ZERO, attempt to do it
- int vdo_sys_fs_cgroup = 0;
- int vdo_cpu_netdata = !config_get_boolean("plugin:cgroups", "cgroups plugin resources", 1);
+ // when ZERO, attempt to do it
+ int vdo_sys_fs_cgroup = 0;
+ int vdo_cpu_netdata = !config_get_boolean("plugin:cgroups", "cgroups plugin resources", 1);
- // keep track of the time each module was called
- unsigned long long sutime_sys_fs_cgroup = 0ULL;
+ // keep track of the time each module was called
+ unsigned long long sutime_sys_fs_cgroup = 0ULL;
- // the next time we will run - aligned properly
- unsigned long long sunext = (time(NULL) - (time(NULL) % rrd_update_every) + rrd_update_every) * 1000000ULL;
- unsigned long long sunow;
+ // the next time we will run - aligned properly
+ unsigned long long sunext = (time(NULL) - (time(NULL) % rrd_update_every) + rrd_update_every) * 1000000ULL;
+ unsigned long long sunow;
- RRDSET *stcpu_thread = NULL;
+ RRDSET *stcpu_thread = NULL;
- for(;1;) {
- if(unlikely(netdata_exit)) break;
+ for(;1;) {
+ if(unlikely(netdata_exit)) break;
- // delay until it is our time to run
- while((sunow = timems()) < sunext)
- usleep((useconds_t)(sunext - sunow));
+ // delay until it is our time to run
+ while((sunow = time_usec()) < sunext)
+ sleep_usec(sunext - sunow);
- // find the next time we need to run
- while(timems() > sunext)
- sunext += rrd_update_every * 1000000ULL;
+ // find the next time we need to run
+ while(time_usec() > sunext)
+ sunext += rrd_update_every * 1000000ULL;
- if(unlikely(netdata_exit)) break;
+ if(unlikely(netdata_exit)) break;
- // BEGIN -- the job to be done
+ // BEGIN -- the job to be done
- if(!vdo_sys_fs_cgroup) {
- debug(D_PROCNETDEV_LOOP, "PROCNETDEV: calling do_sys_fs_cgroup().");
- sunow = timems();
- vdo_sys_fs_cgroup = do_sys_fs_cgroup(rrd_update_every, (sutime_sys_fs_cgroup > 0)?sunow - sutime_sys_fs_cgroup:0ULL);
- sutime_sys_fs_cgroup = sunow;
- }
- if(unlikely(netdata_exit)) break;
+ if(!vdo_sys_fs_cgroup) {
+ debug(D_PROCNETDEV_LOOP, "PROCNETDEV: calling do_sys_fs_cgroup().");
+ sunow = time_usec();
+ vdo_sys_fs_cgroup = do_sys_fs_cgroup(rrd_update_every, (sutime_sys_fs_cgroup > 0)?sunow - sutime_sys_fs_cgroup:0ULL);
+ sutime_sys_fs_cgroup = sunow;
+ }
+ if(unlikely(netdata_exit)) break;
- // END -- the job is done
+ // END -- the job is done
- // --------------------------------------------------------------------
+ // --------------------------------------------------------------------
- if(!vdo_cpu_netdata) {
- getrusage(RUSAGE_THREAD, &thread);
+ if(!vdo_cpu_netdata) {
+ getrusage(RUSAGE_THREAD, &thread);
- if(!stcpu_thread) stcpu_thread = rrdset_find("netdata.plugin_cgroups_cpu");
- if(!stcpu_thread) {
- stcpu_thread = rrdset_create("netdata", "plugin_cgroups_cpu", NULL, "proc.internal", NULL, "NetData CGroups Plugin CPU usage", "milliseconds/s", 132000, rrd_update_every, RRDSET_TYPE_STACKED);
+ if(!stcpu_thread) stcpu_thread = rrdset_find("netdata.plugin_cgroups_cpu");
+ if(!stcpu_thread) {
+ stcpu_thread = rrdset_create("netdata", "plugin_cgroups_cpu", NULL, "proc.internal", NULL, "NetData CGroups Plugin CPU usage", "milliseconds/s", 132000, rrd_update_every, RRDSET_TYPE_STACKED);
- rrddim_add(stcpu_thread, "user", NULL, 1, 1000, RRDDIM_INCREMENTAL);
- rrddim_add(stcpu_thread, "system", NULL, 1, 1000, RRDDIM_INCREMENTAL);
- }
- else rrdset_next(stcpu_thread);
+ rrddim_add(stcpu_thread, "user", NULL, 1, 1000, RRDDIM_INCREMENTAL);
+ rrddim_add(stcpu_thread, "system", NULL, 1, 1000, RRDDIM_INCREMENTAL);
+ }
+ else rrdset_next(stcpu_thread);
- rrddim_set(stcpu_thread, "user" , thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
- rrddim_set(stcpu_thread, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
- rrdset_done(stcpu_thread);
- }
- }
+ rrddim_set(stcpu_thread, "user" , thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
+ rrddim_set(stcpu_thread, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
+ rrdset_done(stcpu_thread);
+ }
+ }
- pthread_exit(NULL);
- return NULL;
+ pthread_exit(NULL);
+ return NULL;
}