File descr alarm v01 (#10192)

4 files changed, 26 insertions, 1 deletions

diff --git a/collectors/apps.plugin/apps_plugin.c b/collectors/apps.plugin/apps_plugin.c
index b440bcd5ec..21146bb341 100644
--- a/collectors/apps.plugin/apps_plugin.c
+++ b/collectors/apps.plugin/apps_plugin.c
@@ -508,6 +508,7 @@ struct file_descriptor {
 static int
         all_files_len = 0,
         all_files_size = 0;
+        long double currentmaxfds = 0;
 
 // ----------------------------------------------------------------------------
 // read users and groups from files
@@ -2998,6 +2999,7 @@ static inline void aggregate_pid_fds_on_targets(struct pid_stat *p) {
     reallocate_target_fds(u);
     reallocate_target_fds(g);
 
+    long double currentfds = 0;
     size_t c, size = p->fds_size;
     struct pid_fd *fds = p->fds;
     for(c = 0; c < size ;c++) {
@@ -3006,10 +3008,15 @@ static inline void aggregate_pid_fds_on_targets(struct pid_stat *p) {
         if(likely(fd <= 0 || fd >= all_files_size))
             continue;
 
+        currentfds++;
+
         aggregate_fd_on_target(fd, w);
         aggregate_fd_on_target(fd, u);
         aggregate_fd_on_target(fd, g);
     }
+
+    if (currentfds >= currentmaxfds)
+        currentmaxfds = currentfds;
 }
 
 static inline void aggregate_pid_on_target(struct target *w, struct pid_stat *p, struct target *o) {
@@ -3607,6 +3614,10 @@ static void send_collected_data_to_netdata(struct target *root, const char *type
             if (unlikely(w->exposed && w->processes))
                 send_SET(w->name, w->openfiles);
         }
+        if (!strcmp("apps", type)){
+            kernel_uint_t usedfdpercentage = (kernel_uint_t) ((currentmaxfds * 100) / sysconf(_SC_OPEN_MAX));
+            fprintf(stdout, "VARIABLE fdperc = " KERNEL_UINT_FORMAT "\n", usedfdpercentage);
+        }
         send_END();
 
         send_BEGIN(type, "sockets", dt);
@@ -4184,6 +4195,7 @@ int main(int argc, char **argv) {
             exit(1);
         }
 
+        currentmaxfds = 0;
         calculate_netdata_statistics();
         normalize_utilization(apps_groups_root_target);
 
diff --git a/health/Makefile.am b/health/Makefile.am
index 0802dc750f..399d6df5ab 100644
--- a/health/Makefile.am
+++ b/health/Makefile.am
@@ -29,6 +29,7 @@ dist_healthconfig_DATA = \
     health.d/anomalies.conf \
     health.d/apache.conf \
     health.d/apcupsd.conf \
+    health.d/apps_plugin.conf \
     health.d/backend.conf \
     health.d/bcache.conf \
     health.d/beanstalkd.conf \
diff --git a/health/health.d/apps_plugin.conf b/health/health.d/apps_plugin.conf
new file mode 100644
index 0000000000..d3670f1c9d
--- /dev/null
+++ b/health/health.d/apps_plugin.conf
@@ -0,0 +1,13 @@
+# you can disable an alarm notification by setting the 'to' line to: silent
+
+   alarm: used_file_descriptors
+      on: apps.files
+   hosts: *
+    calc: $fdperc
+   units: %
+   every: 5s
+    warn: $this > (($status >= $WARNING)  ? (75) : (80))
+    crit: $this > (($status == $CRITICAL) ? (85) : (90))
+   delay: down 5m multiplier 1.5 max 1h
+    info: Peak percentage of file descriptors used
+      to: sysadmin
diff --git a/libnetdata/os.c b/libnetdata/os.c
index 0248eb627c..4271a917cd 100644
--- a/libnetdata/os.c
+++ b/libnetdata/os.c
@@ -215,4 +215,3 @@ int getsysctl_by_name(const char *name, void *ptr, size_t len) {
 }
 
 #endif // (TARGET_OS == OS_MACOS)
-