Cause make start to wait for master start.

masters/master-common-rules.mk

Index: masters/master-common-rules.mk
diff --git a/masters/master-common-rules.mk b/masters/master-common-rules.mk
index 73d249aa6998ccf25c0c06eccca290ecca7a102b..a8b72e654c4b338ef119c8dca7894b33327c6d67 100644
--- a/masters/master-common-rules.mk
+++ b/masters/master-common-rules.mk
@@ -63,8 +63,15 @@ ifndef NO_REVISION_AUDIT
    || echo 'Running send_monitoring_event failed, skipping sending events.' \
   ) 2>&1 | tee -a actions.log
 endif
+	# There is a race condition between startup and when twistd.pid is written.
+	# Anyone issuing a second `make start` before the twistd.pid is written will
+	# spawn two masters. This is hopefully unlikely, but a proper solution would
+	# be to use flock (not available on OSX). The critical section contains BOTH
+	# the twistd start and the wait for the twistd.pid.
 	@echo 'Now running Buildbot master.'
 	PYTHONPATH=$(PYTHONPATH) python $(SCRIPTS_DIR)/common/twistd --no_save -y buildbot.tac
+	@echo 'Waiting for creation of twistd.pid...'
+	while `test ! -f twistd.pid`; do sleep 1; done;

[dsansome, 2016/03/18 00:01:57]

What if twistd.pid is never created?  Does master manager have its own timeout?

 
 ifeq ($(BUILDBOT_PATH),$(BUILDBOT8_PATH))
 start-prof: bootstrap
@@ -111,6 +118,9 @@ last-restart:
 wait:
 	while `test -f twistd.pid`; do sleep 1; done;
 
+waitforstart:
+	while `test ! -f twistd.pid`; do sleep 1; done;
+
 restart: notify stop wait start log
 
 restart-prof: stop wait start-prof log