optimize statsd cleanup; work around freeze of musl-libc cleanup on exit; https://github.com/firehol/binary-packages/issues/4

3 files changed, 47 insertions, 30 deletions

diff --git a/makeself/jobs/70-netdata-git.install.sh b/makeself/jobs/70-netdata-git.install.sh
index b85481492c..5369205c8b 100755
--- a/makeself/jobs/70-netdata-git.install.sh
+++ b/makeself/jobs/70-netdata-git.install.sh
@@ -6,9 +6,10 @@ cd "${NETDATA_SOURCE_PATH}" || exit 1
 
 if [ ${NETDATA_BUILD_WITH_DEBUG} -eq 0 ]
 then
-    export CFLAGS="-static -O3"
+    export CFLAGS="-static -O3 -DNETDATA_WAIT_BEFORE_EXIT=2"
 else
     export CFLAGS="-static -O1 -ggdb -Wall -Wextra -Wformat-signedness -fstack-protector-all -D_FORTIFY_SOURCE=2 -DNETDATA_INTERNAL_CHECKS=1"
+#    export CFLAGS="-static -O1 -ggdb -Wall -Wextra -Wformat-signedness"
 fi
 
 if [ ! -z "${NETDATA_INSTALL_PATH}" -a -d "${NETDATA_INSTALL_PATH}/etc" ]
diff --git a/src/main.c b/src/main.c
index 0422b2047e..806028d476 100644
--- a/src/main.c
+++ b/src/main.c
@@ -6,28 +6,33 @@ void netdata_cleanup_and_exit(int ret) {
     netdata_exit = 1;
 
     error_log_limit_unlimited();
+    info("netdata is preparing to exit...");
 
-    debug(D_EXIT, "Called: netdata_cleanup_and_exit()");
+    // allow all the threads to cleanup by themselves
+    unsigned int w = (unsigned int)default_rrd_update_every + 2;
+    info("Giving %u secs to background threads to cleanup...", w);
+    sleep(w);
 
-    // cleanup the database
+    // kill all threads and childs
+    //info("Stopping all threads and child processes...");
+    //kill_childs();
+
+    // cleanup the database (delete files not needed)
+    info("Cleaning up the database...");
     rrdhost_cleanup_all();
 
+    // free the database
+    //info("Freeing database memory...");
+    //rrdhost_free_all();
+
     // unlink the pid
     if(pidfile[0]) {
+        info("Removing netdata PID file '%s'...", pidfile);
         if(unlink(pidfile) != 0)
             error("Cannot unlink pidfile '%s'.", pidfile);
     }
 
-#ifdef NETDATA_INTERNAL_CHECKS
-    // kill all childs
-    //kill_childs();
-
-    // free database
-    sleep(2);
-    rrdhost_free_all();
-#endif
-
-    info("netdata exiting. Bye bye...");
+    info("All done - netdata is now exiting - bye bye...");
     exit(ret);
 }
 
@@ -177,8 +182,7 @@ int killpid(pid_t pid, int sig)
     return ret;
 }
 
-void kill_childs()
-{
+void kill_childs() {
     error_log_limit_unlimited();
 
     siginfo_t info;
@@ -186,9 +190,9 @@ void kill_childs()
     struct web_client *w;
     for(w = web_clients; w ; w = w->next) {
         info("Stopping web client %s", w->client_ip);
-        pthread_cancel(w->thread);
-        // it is detached
-        // pthread_join(w->thread, NULL);
+        int ret;
+        if((ret = pthread_cancel(w->thread)) != 0)
+            error("pthread_cancel() failed with code %d.", ret);
 
         WEB_CLIENT_IS_OBSOLETE(w);
     }
@@ -197,9 +201,9 @@ void kill_childs()
     for (i = 0; static_threads[i].name != NULL ; i++) {
         if(static_threads[i].enabled) {
             info("Stopping %s thread", static_threads[i].name);
-            pthread_cancel(*static_threads[i].thread);
-            // it is detached
-            // pthread_join(*static_threads[i].thread, NULL);
+            int ret;
+            if((ret = pthread_cancel(*static_threads[i].thread)) != 0)
+                error("pthread_cancel() failed with code %d.", ret);
 
             static_threads[i].enabled = 0;
         }
diff --git a/src/statsd.c b/src/statsd.c
index 2d07a6126a..9499b1bc36 100644
--- a/src/statsd.c
+++ b/src/statsd.c
@@ -252,7 +252,10 @@ static struct statsd {
     size_t histogram_increase_step;
     double histogram_percentile;
     char *histogram_percentile_str;
+
     int threads;
+    pthread_t *collection_threads;
+
     LISTEN_SOCKETS sockets;
 } statsd = {
         .enabled = 1,
@@ -320,6 +323,7 @@ static struct statsd {
         .histogram_percentile = 95.0,
         .histogram_increase_step = 10,
         .threads = 0,
+        .collection_threads = NULL,
         .sockets = {
                 .config_section  = CONFIG_SECTION_STATSD,
                 .default_bind_to = "udp:localhost tcp:localhost",
@@ -1969,13 +1973,22 @@ int statsd_listen_sockets_setup(void) {
 }
 
 void statsd_main_cleanup(void *data) {
-    pthread_t *threads = data;
+    (void)data;
 
-    int i;
-    for(i = 0; i < statsd.threads ;i++)
-        pthread_cancel(threads[i]);
+    info("STATSD: cleaning up...");
 
+    if(statsd.collection_threads) {
+        int i;
+        for (i = 0; i < statsd.threads; i++) {
+            info("STATSD: stopping data collection thread %d...", i);
+            pthread_cancel(statsd.collection_threads[i]);
+        }
+    }
+
+    info("STATSD: closing sockets...");
     listen_sockets_close(&statsd.sockets);
+
+    info("STATSD: cleanup completed.");
 }
 
 void *statsd_main(void *ptr) {
@@ -2079,18 +2092,17 @@ void *statsd_main(void *ptr) {
         pthread_exit(NULL);
     }
 
-    pthread_t threads[statsd.threads];
+    statsd.collection_threads = callocz((size_t)statsd.threads, sizeof(pthread_t));
     int i;
-
     for(i = 0; i < statsd.threads ;i++) {
-        if(pthread_create(&threads[i], NULL, statsd_collector_thread, &i))
+        if(pthread_create(&statsd.collection_threads[i], NULL, statsd_collector_thread, &i))
             error("STATSD: failed to create child thread.");
 
-        else if(pthread_detach(threads[i]))
+        else if(pthread_detach(statsd.collection_threads[i]))
             error("STATSD: cannot request detach of child thread.");
     }
 
-    pthread_cleanup_push(statsd_main_cleanup, &threads);
+    pthread_cleanup_push(statsd_main_cleanup, NULL);
 
     // ----------------------------------------------------------------------------------------------------------------
     // statsd monitoring charts