trackrdrd: shutdown cancels the monitoring threads (no need to wait for it)

           init script start checks if trackrdrd is running, waits for stop

trackrdrd/etc/trackrdrd

@@ -28,11 +28,31 @@ CMD="${TRACKRDRD_BIN}"
 CURL=/usr/bin/curl
 URL=http://localhost/ts-rcv?testSession=true
 
+START_RETRIES=5
+
 if [ ! -e ${TRACKRDRD_BIN} ]; then
     echo "${TRACKRDRD_BIN} not installed"
 fi
 
 function start_trackrdrd {
+    START_CHECKS=0
+    echo -n "Checking if trackrdrd is running ..."
+    while [ ${START_CHECKS} -lt ${START_RETRIES} ]; do
+        checkproc ${TRACKRDRD_BIN}
+        if [ $? -ne 0 ]; then
+            break
+        fi
+        echo -n "."
+        START_CHECKS=$(( $START_CHECKS + 1 ))
+        sleep 1
+        continue
+    done
+    if [ ${START_CHECKS} -eq ${START_RETRIES} ]; then
+        echo " trackrdrd still running, giving up"
+        logger "trackrdrd still running, start FAILED"
+        exit 1
+    fi
+    echo ""
     echo -n "Starting trackrdrd ... "
     startproc -q ${CMD}
     if [ $? -ne 0 ]; then
@@ -48,7 +68,7 @@ function stop_trackrdrd {
     killproc ${TRACKRDRD_BIN}
     if [ $? -ne 0 ]; then
 	echo "FAILED (killproc returned $?)"
-        logger "trackrdrd start FAILED (killproc returned $?)"
+        logger "trackrdrd stop FAILED (killproc returned $?)"
         exit 1
     fi
     ${CURL} ${URL} >/dev/null 2&>1
@@ -74,11 +94,6 @@ function check_varnishd {
 
 case "$1" in
     start)
-        checkproc ${TRACKRDRD_BIN}
-        if [ $? -eq 0 ]; then
-            echo "trackrdrd already running"
-            exit 1
-        fi
         check_varnishd
 	start_trackrdrd
 	exit 0
@@ -91,7 +106,6 @@ case "$1" in
         check_varnishd
         echo "Restarting trackrdrd"
         stop_trackrdrd
-	sleep 1
         start_trackrdrd
 	sleep 1
         getstatus_trackrdrd

trackrdrd/src/monitor.c

@@ -40,6 +40,32 @@
 
 static int run;
 
+static void
+log_output(void)
+{
+    LOG_Log(LOG_INFO,
+        "Data table: len=%u collisions=%u insert_probes=%u find_probes=%u "
+        "open=%u done=%u load=%.2f len_overflows=%u data_overflows=%u "
+        "occ_hi=%u seen=%u submitted=%u nodata=%u sent=%u failed=%u "
+        "wait_qfull=%u data_hi=%u",
+        tbl.len, tbl.collisions, tbl.insert_probes, tbl.find_probes,
+        tbl.open, tbl.done, 100.0 * ((float) tbl.open + tbl.done) / tbl.len,
+        tbl.len_overflows, tbl.data_overflows, tbl.occ_hi, tbl.seen,
+        tbl.submitted, tbl.nodata, tbl.sent, tbl.failed, tbl.wait_qfull,
+        tbl.data_hi);
+    if (config.monitor_workers)
+        WRK_Stats();
+}
+
+static void
+monitor_cleanup(void *arg)
+{
+    (void) arg;
+
+    log_output();
+    LOG_Log0(LOG_INFO, "Monitoring thread exiting");
+}
+
 void
 *MON_StatusThread(void *arg)
 {
@@ -51,7 +77,9 @@ void
     LOG_Log(LOG_INFO, "Monitor thread running every %.2f secs",
             t.tv_sec + ((float) t.tv_nsec * 10e-9));
     run = 1;
-    
+
+    pthread_cleanup_push(monitor_cleanup, arg);
+
     while (run) {
         int err;
         if (nanosleep(&t, NULL) != 0) {
@@ -68,20 +96,10 @@ void
                 pthread_exit(&err);
             }
         }
-        LOG_Log(LOG_INFO,
-            "Data table: len=%u collisions=%u insert_probes=%u find_probes=%u "
-            "open=%u done=%u load=%.2f len_overflows=%u data_overflows=%u "
-            "occ_hi=%u seen=%u submitted=%u nodata=%u sent=%u failed=%u "
-            "wait_qfull=%u data_hi=%u",
-            tbl.len, tbl.collisions, tbl.insert_probes, tbl.find_probes,
-            tbl.open, tbl.done, 100.0 * ((float) tbl.open + tbl.done) / tbl.len,
-            tbl.len_overflows, tbl.data_overflows, tbl.occ_hi, tbl.seen,
-            tbl.submitted, tbl.nodata, tbl.sent, tbl.failed, tbl.wait_qfull,
-            tbl.data_hi);
-        if (config.monitor_workers)
-            WRK_Stats();
+        log_output();
     }
 
+    pthread_cleanup_pop(0);
     LOG_Log0(LOG_INFO, "Monitoring thread exiting");
     pthread_exit((void *) NULL);
 }
@@ -90,6 +108,7 @@ void
 MON_StatusShutdown(pthread_t monitor)
 {
     run = 0;
+    AZ(pthread_cancel(monitor));
     AZ(pthread_join(monitor, NULL));
 }