gerrit.sh: Wait until the daemon is serving requests

Don't display "OK" until the daemon is able to serve requests.
This permits subsequent scripts to rely on our exit status to know
whether or not Gerrit Code Review is ready.

We now wait up to 90 seconds for the server to come online before
we report "OK" or "FAILED" to the console and the caller.

Because its difficult to compute the exact network addresses that
the daemon will be listening on, we instead pass it a magic cookie
as part of the arguments.  When the daemon's network listeners are
accepting connections the main thread writes that cookie into the
logs/gerrit.run file.  The calling script polls that file until
the cookie shows up.

Bug: issue 451
Change-Id: Ic5018b347a5037221c3d26c16c71173f5071d244
Signed-off-by: Shawn O. Pearce <sop@google.com>

gerrit-pgm/src/main/java/com/google/gerrit/pgm/Daemon.java

@@ -42,6 +42,9 @@ import org.kohsuke.args4j.Option;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
 import java.util.ArrayList;
 import java.util.List;
 
@@ -73,6 +76,9 @@ public class Daemon extends SiteProgram {
   @Option(name = "--console-log", usage = "Log to console (not $site_path/logs)")
   private boolean consoleLog;
 
+  @Option(name = "--run-id", usage = "Cookie to store in $site_path/logs/gerrit.run")
+  private String runId;
+
   private final LifecycleManager manager = new LifecycleManager();
   private Injector dbInjector;
   private Injector cfgInjector;
@@ -80,11 +86,16 @@ public class Daemon extends SiteProgram {
   private Injector sshInjector;
   private Injector webInjector;
   private Injector httpdInjector;
+  private File runFile;
 
   @Override
   public int run() throws Exception {
     mustHaveValidSite();
 
+    if (runId != null) {
+      runFile = new File(new File(getSitePath(), "logs"), "gerrit.run");
+    }
+
     if (httpd == null) {
       httpd = !slave;
     }
@@ -119,13 +130,33 @@ public class Daemon extends SiteProgram {
     }
 
     manager.start();
-    log.info("Gerrit Code Review " + myVersion() + " ready");
     RuntimeShutdown.add(new Runnable() {
       public void run() {
         log.info("caught shutdown, cleaning up");
+        if (runId != null) {
+          runFile.delete();
+        }
         manager.stop();
       }
     });
+
+    log.info("Gerrit Code Review " + myVersion() + " ready");
+    if (runId != null) {
+      try {
+        runFile.createNewFile();
+        runFile.setReadable(true, false);
+
+        FileOutputStream out = new FileOutputStream(runFile);
+        try {
+          out.write((runId + "\n").getBytes("UTF-8"));
+        } finally {
+          out.close();
+        }
+      } catch (IOException err) {
+        log.warn("Cannot write --run-id to " + runFile, err);
+      }
+    }
+
     RuntimeShutdown.waitFor();
     return 0;
   }

gerrit-pgm/src/main/java/com/google/gerrit/pgm/util/LogFileCompressor.java

@@ -99,6 +99,7 @@ public class LogFileCompressor implements Runnable {
     return ErrorLogFile.LOG_NAME.equals(name) //
         || "sshd_log".equals(name) //
         || "httpd_log".equals(name) //
+        || "gerrit.run".equals(name) //
         || name.endsWith(".pid");
   }
 

gerrit-pgm/src/main/resources/com/google/gerrit/pgm/gerrit.sh

@@ -201,6 +201,7 @@ test -r "$GERRIT_CONFIG" || {
 }
 
 GERRIT_PID="$GERRIT_SITE/logs/gerrit.pid"
+GERRIT_RUN="$GERRIT_SITE/logs/gerrit.run"
 
 ##################################################
 # Check for JAVA_HOME
@@ -360,6 +361,9 @@ case "$ACTION" in
 
     test -z "$UID" && UID=`id -u`
 
+    RUN_ID=$(date +%s).$$
+    RUN_ARGS="$RUN_ARGS --run-id=$RUN_ID"
+
     if test 1 = "$START_STOP_DAEMON" && type start-stop-daemon >/dev/null 2>&1
     then
       test $UID = 0 && CH_USER="-c $GERRIT_USER"
@@ -368,12 +372,14 @@ case "$ACTION" in
          -d "$GERRIT_SITE" \
          -a "$RUN_EXEC" -- $RUN_Arg1 "$RUN_Arg2" $RUN_Arg3 $RUN_ARGS
       then
-          sleep 1
+        : OK
-          if running "$GERRIT_PID" ; then
-            echo OK
       else
-            echo FAILED
+        rc=$?
+        if test $rc = 127; then
+          echo >&2 "fatal: start-stop-daemon failed"
+          rc=1
         fi
+        exit $rc 
       fi
     else
       if test -f "$GERRIT_PID" ; then
@@ -381,7 +387,7 @@ case "$ACTION" in
           echo "Already Running!!"
           exit 1
         else
-          rm -f "$GERRIT_PID"
+          rm -f "$GERRIT_PID" "$GERRIT_RUN"
         fi
       fi
 
@@ -399,9 +405,22 @@ case "$ACTION" in
         disown $PID
         echo $PID >"$GERRIT_PID"
       fi
-
-      echo OK
     fi
+
+    TIMEOUT=90  # seconds
+    sleep 1
+    while running "$GERRIT_PID" && test $TIMEOUT -gt 0 ; do
+      if test "x$RUN_ID" = "x$(cat $GERRIT_RUN 2>/dev/null)" ; then
+        echo OK
+        exit 0
+      fi
+
+      sleep 2
+      let TIMEOUT=$TIMEOUT-2
+    done
+
+    echo FAILED
+    exit 1
   ;;
 
   stop)
@@ -420,7 +439,7 @@ case "$ACTION" in
           fi
         fi
       fi
-      rm -f "$GERRIT_PID"
+      rm -f "$GERRIT_PID" "$GERRIT_RUN"
       echo OK
     else
       PID=`cat "$GERRIT_PID" 2>/dev/null`
@@ -431,7 +450,7 @@ case "$ACTION" in
         let TIMEOUT=$TIMEOUT-1
       done
       test $TIMEOUT -gt 0 || kill -9 $PID 2>/dev/null
-      rm -f "$GERRIT_PID"
+      rm -f "$GERRIT_PID" "$GERRIT_RUN"
       echo OK
     fi
   ;;