]> git.proxmox.com Git - mirror_lxcfs.git/commitdiff
Calculate and read the average load.
authorzhang2639 <zhangshiqiang@hust.edu.cn>
Thu, 24 May 2018 10:42:37 +0000 (18:42 +0800)
committerzhang2639 <zhangshiqiang@hust.edu.cn>
Thu, 24 May 2018 10:42:37 +0000 (18:42 +0800)
Use load daemon to calculate the loadavg and use proc_loadavg_read() to read the loadavg.
calc_pid : find the process pid from cgroup path of a container.
calc_load : calculate the loadavg of a container.
refresh_load : refresh the loadavg of a container.
load_begin : traverse the hash table and update it.

Signed-off-by: zhang2639 <zhangshiqiang@hust.edu.cn>
Signed-off-by: yuwang <yuwang@linux.alibaba.com>
bindings.c
bindings.h
lxcfs.c

index d4ddd3300e1d2e41588e051f483d856454dab82c..b08cf43e725d38ccf0608dcfda4d91a36c35c3a1 100644 (file)
@@ -82,6 +82,21 @@ struct file_info {
 
 /* The function of hash table.*/
 #define LOAD_SIZE 100 /*the size of hash_table */
+#define FLUSH_TIME 5  /*the flush rate */
+#define DEPTH_DIR 3   /*the depth of per cgroup */
+/* The function of calculate loadavg .*/
+#define FSHIFT         11              /* nr of bits of precision */
+#define FIXED_1                (1<<FSHIFT)     /* 1.0 as fixed-point */
+#define EXP_1          1884            /* 1/exp(5sec/1min) as fixed-point */
+#define EXP_5          2014            /* 1/exp(5sec/5min) */
+#define EXP_15         2037            /* 1/exp(5sec/15min) */
+#define LOAD_INT(x) ((x) >> FSHIFT)
+#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
+/* 
+ * This parameter is used for proc_loadavg_read().
+ * 1 means use loadavg, 0 means not use.
+ */
+static int loadavg = 0;
 static int calc_hash(char *name)
 {
        unsigned int hash = 0;
@@ -4244,6 +4259,363 @@ err:
        free(memswusage_str);
        return rv;
 }
+/*
+ * Find the process pid from cgroup path.
+ * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid.
+ * @pid_buf : put pid to pid_buf.
+ * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ...
+ * @depth : the depth of cgroup in container.
+ * @sum : return the number of pid.
+ * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu
+ */
+static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd)
+{
+       DIR *dir;
+       int fd;
+       struct dirent *file;
+       FILE *f = NULL;
+       size_t linelen = 0;
+       char *line = NULL;
+       int pd;
+       char *path_dir, *path;
+       char **pid;
+
+       /* path = dpath + "/cgroup.procs" + /0 */
+       do {
+               path = malloc(strlen(dpath) + 20);
+       } while (!path);
+
+       strcpy(path, dpath);
+       fd = openat(cfd, path, O_RDONLY);
+       if (fd < 0)
+               goto out;
+
+       dir = fdopendir(fd);
+       if (dir == NULL) {
+               close(fd);
+               goto out;
+       }
+
+       while (((file = readdir(dir)) != NULL) && depth > 0) {
+               if (strncmp(file->d_name, ".", 1) == 0)
+                       continue;
+               if (strncmp(file->d_name, "..", 1) == 0)
+                       continue;
+               if (file->d_type == DT_DIR) {
+                       /* path + '/' + d_name +/0 */
+                       do {
+                               path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name));
+                       } while (!path_dir);
+                       strcpy(path_dir, path);
+                       strcat(path_dir, "/");
+                       strcat(path_dir, file->d_name);
+                       pd = depth - 1;
+                       sum = calc_pid(pid_buf, path_dir, pd, sum, cfd);
+                       free(path_dir);
+               }
+       }
+       closedir(dir);
+
+       strcat(path, "/cgroup.procs");
+       fd = openat(cfd, path, O_RDONLY);
+       if (fd < 0)
+               goto out;
+
+       f = fdopen(fd, "r");
+       if (!f) {
+               close(fd);
+               goto out;
+       }
+
+       while (getline(&line, &linelen, f) != -1) {
+               do {
+                       pid = realloc(*pid_buf, sizeof(char *) * (sum + 1));
+               } while (!pid);
+               *pid_buf = pid;
+               do {
+                       *(*pid_buf + sum) = malloc(strlen(line) + 1);
+               } while (*(*pid_buf + sum) == NULL);
+               strcpy(*(*pid_buf + sum), line);
+               sum++;
+       }
+       fclose(f);
+out:
+       free(path);
+       return sum;
+}
+/*
+ * calc_load calculates the load according to the following formula:
+ * load1 = load0 * exp + active * (1 - exp)
+ *
+ * @load1: the new loadavg.
+ * @load0: the former loadavg.
+ * @active: the total number of running pid at this moment.
+ * @exp: the fixed-point defined in the beginning.
+ */
+static unsigned long
+calc_load(unsigned long load, unsigned long exp, unsigned long active)
+{
+       unsigned long newload;
+
+       active = active > 0 ? active * FIXED_1 : 0;
+       newload = load * exp + active * (FIXED_1 - exp);
+       if (active >= load)
+               newload += FIXED_1 - 1;
+
+       return newload / FIXED_1;
+}
+
+/*
+ * Return 0 means that container p->cg is closed.
+ * Return -1 means that error occurred in refresh.
+ * Positive num equals the total number of pid.
+ */
+static int refresh_load(struct load_node *p, char *path)
+{
+       FILE *f = NULL;
+       char **idbuf;
+       char proc_path[256];
+       int i, ret, run_pid = 0, total_pid = 0, last_pid = 0;
+       char *line = NULL;
+       size_t linelen = 0;
+       int sum, length;
+       DIR *dp;
+       struct dirent *file;
+
+       do {
+               idbuf = malloc(sizeof(char *));
+       } while (!idbuf);
+       sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd);
+       /*  normal exit  */
+       if (sum == 0)
+               goto out;
+
+       for (i = 0; i < sum; i++) {
+               /*clean up '\n' */
+               length = strlen(idbuf[i])-1;
+               idbuf[i][length] = '\0';
+               ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]);
+               if (ret < 0 || ret > 255) {
+                       lxcfs_error("%s\n", "snprintf() failed in refresh_load.");
+                       i = sum;
+                       sum = -1;
+                       goto err_out;
+               }
+
+               dp = opendir(proc_path);
+               if (!dp) {
+                       lxcfs_error("%s\n", "Open proc_path failed in refresh_load.");
+                       continue;
+               }
+               while ((file = readdir(dp)) != NULL) {
+                       if (strncmp(file->d_name, ".", 1) == 0)
+                               continue;
+                       if (strncmp(file->d_name, "..", 1) == 0)
+                               continue;
+                       total_pid++;
+                       /* We make the biggest pid become last_pid.*/
+                       ret = atof(file->d_name);
+                       last_pid = (ret > last_pid) ? ret : last_pid;
+
+                       ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name);
+                       if (ret < 0 || ret > 255) {
+                               lxcfs_error("%s\n", "snprintf() failed in refresh_load.");
+                               i = sum;
+                               sum = -1;
+                               closedir(dp);
+                               goto err_out;
+                       }
+                       f = fopen(proc_path, "r");
+                       if (f != NULL) {
+                               while (getline(&line, &linelen, f) != -1) {
+                                       /* Find State */
+                                       if ((line[0] == 'S') && (line[1] == 't'))
+                                               break;
+                               }
+                       if ((line[7] == 'R') || (line[7] == 'D'))
+                               run_pid++;
+                       fclose(f);
+                       }
+               }
+               closedir(dp);
+       }
+       /*Calculate the loadavg.*/
+       p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid);
+       p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid);
+       p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid);
+       p->run_pid = run_pid;
+       p->total_pid = total_pid;
+       p->last_pid = last_pid;
+
+       free(line);
+err_out:       
+       for (; i > 0; i--)
+               free(idbuf[i-1]);
+out:
+       free(idbuf);
+       return sum;
+}
+/*
+ * Traverse the hash table and update it.
+ */
+void *load_begin(void *arg)
+{
+
+       char *path = NULL;
+       int i, sum, length, ret;
+       struct load_node *f;
+       int first_node;
+       clock_t time1, time2;
+
+       while (1) {
+               time1 = clock();
+               for (i = 0; i < LOAD_SIZE; i++) {
+                       pthread_mutex_lock(&load_hash[i].lock);
+                       if (load_hash[i].next == NULL) {
+                               pthread_mutex_unlock(&load_hash[i].lock);
+                               continue;
+                       }
+                       f = load_hash[i].next;
+                       first_node = 1;
+                       while (f) {
+                               length = strlen(f->cg) + 2;
+                               do {
+                                       /* strlen(f->cg) + '.' or '' + \0 */
+                                       path = malloc(length);
+                               } while (!path);
+
+                               ret = snprintf(path, length, "%s%s", *(f->cg) == '/' ? "." : "", f->cg);
+                               if (ret < 0 || ret > length - 1) {
+                                       /* snprintf failed, ignore the node.*/
+                                       lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg);
+                                       goto out;
+                               }
+                               sum = refresh_load(f, path);
+                               if (sum == 0) {
+                                       f = del_node(f, i);
+                               } else {
+out:                                   f = f->next;
+                               }
+                               free(path);
+                               /* load_hash[i].lock locks only on the first node.*/
+                               if (first_node == 1) {
+                                       first_node = 0;
+                                       pthread_mutex_unlock(&load_hash[i].lock);
+                               }
+                       }
+               }
+               time2 = clock();
+               usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC));
+       }
+}
+
+static int proc_loadavg_read(char *buf, size_t size, off_t offset,
+               struct fuse_file_info *fi)
+{
+       struct fuse_context *fc = fuse_get_context();
+       struct file_info *d = (struct file_info *)fi->fh;
+       pid_t initpid;
+       char *cg;
+       size_t total_len = 0;
+       char *cache = d->buf;
+       struct load_node *n;
+       int hash;
+       int cfd;
+       unsigned long a, b, c;
+
+       if (offset) {
+               if (offset > d->size)
+                       return -EINVAL;
+               if (!d->cached)
+                       return 0;
+               int left = d->size - offset;
+               total_len = left > size ? size : left;
+               memcpy(buf, cache + offset, total_len);
+               return total_len;
+       }
+       if (!loadavg)
+               return read_file("/proc/loadavg", buf, size, d);
+
+       initpid = lookup_initpid_in_store(fc->pid);
+       if (initpid <= 0)
+               initpid = fc->pid;
+       cg = get_pid_cgroup(initpid, "cpu");
+       if (!cg)
+               return read_file("/proc/loadavg", buf, size, d);
+
+       prune_init_slice(cg);
+       hash = calc_hash(cg);
+       n = locate_node(cg, hash);
+
+       /* First time */
+       if (n == NULL) {
+               if (!find_mounted_controller("cpu", &cfd)) {
+                       /*
+                        * In locate_node() above, pthread_rwlock_unlock() isn't used
+                        * because delete is not allowed before read has ended.
+                        */
+                       pthread_rwlock_unlock(&load_hash[hash].rdlock);
+                       return 0;
+               }
+               do {
+                       n = malloc(sizeof(struct load_node));
+               } while (!n);
+
+               do {
+                       n->cg = malloc(strlen(cg)+1);
+               } while (!n->cg);
+               strcpy(n->cg, cg);
+               n->avenrun[0] = 0;
+               n->avenrun[1] = 0;
+               n->avenrun[2] = 0;
+               n->run_pid = 0;
+               n->total_pid = 1;
+               n->last_pid = initpid;
+               n->cfd = cfd;
+               insert_node(&n, hash);
+       }
+       a = n->avenrun[0] + (FIXED_1/200);
+       b = n->avenrun[1] + (FIXED_1/200);
+       c = n->avenrun[2] + (FIXED_1/200);
+       total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n",
+               LOAD_INT(a), LOAD_FRAC(a),
+               LOAD_INT(b), LOAD_FRAC(b),
+               LOAD_INT(c), LOAD_FRAC(c),
+               n->run_pid, n->total_pid, n->last_pid);
+       pthread_rwlock_unlock(&load_hash[hash].rdlock);
+       if (total_len < 0 || total_len >=  d->buflen) {
+               lxcfs_error("%s\n", "Failed to write to cache");
+               return 0;
+       }
+       d->size = (int)total_len;
+       d->cached = 1;
+
+       if (total_len > size)
+               total_len = size;
+       memcpy(buf, d->buf, total_len);
+       return total_len;
+}
+/* Return a positive number on success, return 0 on failure.*/
+pthread_t load_daemon(int load_use)
+{
+       int ret;
+       pthread_t pid;
+
+       ret = init_load();
+       if (ret == -1) {
+               lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!");
+               return 0;
+       }
+       ret = pthread_create(&pid, NULL, load_begin, NULL);
+       if (ret != 0) {
+               lxcfs_error("%s\n", "Create pthread fails in load_daemon!");
+               load_free();
+               return 0;
+       }
+       /* use loadavg, here loadavg = 1*/
+       loadavg = load_use;
+       return pid;
+}
 
 static off_t get_procfile_size(const char *which)
 {
index c663287b15638f53a593bfe55823f0ef2b2df255..a34ac5fa964d2bd05114280936cd2ca331c2244c 100644 (file)
@@ -32,5 +32,6 @@ extern int proc_open(const char *path, struct fuse_file_info *fi);
 extern int proc_read(const char *path, char *buf, size_t size, off_t offset,
                struct fuse_file_info *fi);
 extern int proc_access(const char *path, int mask);
+extern pthread_t load_daemon(int load_use);
 
 #endif /* __LXCFS_BINDINGS_H */
diff --git a/lxcfs.c b/lxcfs.c
index 8370c35e2bdcb148e03c64d862facf188f22408c..fe9c250ea057e2714997b90090f6826a748c152b 100644 (file)
--- a/lxcfs.c
+++ b/lxcfs.c
@@ -745,8 +745,9 @@ static void usage()
 {
        fprintf(stderr, "Usage:\n");
        fprintf(stderr, "\n");
-       fprintf(stderr, "lxcfs [-f|-d] [-p pidfile] mountpoint\n");
+       fprintf(stderr, "lxcfs [-f|-d] -l [-p pidfile] mountpoint\n");
        fprintf(stderr, "  -f running foreground by default; -d enable debug output \n");
+       fprintf(stderr, "  -l use loadavg \n");
        fprintf(stderr, "  Default pidfile is %s/lxcfs.pid\n", RUNTIME_PATH);
        fprintf(stderr, "lxcfs -h\n");
        exit(1);
@@ -848,6 +849,12 @@ int main(int argc, char *argv[])
        char *pidfile = NULL, *saveptr = NULL, *token = NULL, *v = NULL;
        size_t pidfile_len;
        bool debug = false, nonempty = false;
+       pthread_t pid;
+       int s = 0;
+       char *error;
+       bool load_use = false;
+       pthread_t (*load_daemon)(int);
+       void (*load_free)(void);
        /*
         * what we pass to fuse_main is:
         * argv[0] -s [-f|-d] -o allow_other,directio argv[1] NULL
@@ -859,6 +866,9 @@ int main(int argc, char *argv[])
        swallow_arg(&argc, argv, "-s");
        swallow_arg(&argc, argv, "-f");
        debug = swallow_arg(&argc, argv, "-d");
+       if (swallow_arg(&argc, argv, "-l")) {
+               load_use = true;
+       }
        if (swallow_option(&argc, argv, "-o", &v)) {
                /* Parse multiple values */
                for (; (token = strtok_r(v, ",", &saveptr)); v = NULL) {
@@ -912,6 +922,19 @@ int main(int argc, char *argv[])
        if ((pidfd = set_pidfile(pidfile)) < 0)
                goto out;
 
+       if (load_use == true) {
+               dlerror();    /* Clear any existing error */
+
+               load_daemon = (pthread_t (*)(int)) dlsym(dlopen_handle, "load_daemon");
+               error = dlerror();
+               if (error != NULL) {
+                       lxcfs_error("load_daemon fails:%s\n", error);
+                       goto out;
+               }
+               pid = load_daemon(1);
+               if (pid == 0)
+                       goto out;
+       }
        if (!fuse_main(nargs, newargv, &lxcfs_ops, NULL))
                ret = EXIT_SUCCESS;