improve startup if no eth0 is available (LP: #714807)

ChangeLog

@@ -21,6 +21,7 @@
    '#opt_include <filename>' or '#include <filename>' in cloud.cfg
  - allow /etc/hosts to be written from hosts.tmpl. which allows
    getting local-hostname into /etc/hosts (LP: #720440)
+ - better handle startup if there is no eth0 (LP: #714807)
 0.6.0:
  - change permissions of /var/log/cloud-init.log to accomodate
    syslog writing to it (LP: #704509)

cloud-init.py

@@ -27,6 +27,7 @@ import cloudinit.DataSource as ds
 import time
 import logging
 import errno
+import os
 
 def warn(wstr):
     sys.stderr.write("WARN:%s" % wstr)
@@ -75,29 +76,50 @@ def main():
     except Exception as e:
         warn("Failed to get and set output config: %s\n" % e)
 
-    msg = "cloud-init %s running: %s. up %s seconds" % (cmd, now, uptime)
-    sys.stderr.write(msg + "\n")
-    sys.stderr.flush()
-
     cloudinit.logging_set_from_cfg(cfg)
     log = logging.getLogger()
-    log.info(msg)
 
     try:
         cloudinit.initfs()
     except Exception as e:
         warn("failed to initfs, likely bad things to come: %s\n" % str(e))
 
+    nonet_path = "%s/%s" % (cloudinit.get_cpath("data"), "no-net")
 
+    if cmd == "start":
+        stop_files = ( cloudinit.get_ipath_cur("obj_pkl"), nonet_path )
+        # if starting as the network start, there are cases
+        # where everything is already done for us, and it makes
+        # most sense to exit early and silently
+        for f in stop_files:
+            try:
+                fp = open("/var/lib/cloud/instance/obj.pkl","r")
+                fp.close()
+            except:
+                continue
+            
+            log.debug("no need for cloud-init start to run (%s)\n", f)
+            sys.exit(0)
+    elif cmd == "start-local":
         # cache is not instance specific, so it has to be purged
         # but we want 'start' to benefit from a cache if
         # a previous start-local populated one
-    if cmd == "start-local":
         manclean = util.get_cfg_option_bool(cfg, 'manual_cache_clean',False)
         if manclean:
             log.debug("not purging cache, manual_cache_clean = True")
         cloudinit.purge_cache(not manclean)
 
+        try:
+            os.unlink(nonet_path)
+        except OSError as e:
+            if e.errno != errno.ENOENT: raise
+
+    msg = "cloud-init %s running: %s. up %s seconds" % (cmd, now, uptime)
+    sys.stderr.write(msg + "\n")
+    sys.stderr.flush()
+
+    log.info(msg)
+
     cloud = cloudinit.CloudInit(ds_deps=deps[cmd])
 
     try:

upstart/cloud-init-nonet.conf

@@ -0,0 +1,22 @@
+# cloud-init-no-net
+# the purpose of this job is
+#  * to block running of cloud-init until a non 'lo' interface is up
+#  * timeout if one doens't come up in a reasonable amount of time
+start on mounted MOUNTPOINT=/ and stopped cloud-init-local
+stop on net-device-up IFACE!=lo
+task
+
+console output
+
+script
+   # if a non 'lo' interface is up, exit immediately
+   grep -qv '^lo' /var/run/network/ifstate && exit 0
+
+   [ -f /var/lib/cloud/instance/obj.pkl ] && exit 0
+   sleep 10
+   echo $UPSTART_JOB "waiting for a network device."
+   sleep 60
+   echo $UPSTART_JOB "gave up waiting for a network device."
+   : > /var/lib/cloud/data/no-net
+end script
+# EOF

upstart/cloud-init.conf

@@ -1,7 +1,6 @@
 # cloud-init - the initial cloud-init job
 #   crawls metadata service, emits cloud-config
-start on (mounted MOUNTPOINT=/ and net-device-up IFACE=eth0 and \
+start on mounted MOUNTPOINT=/ and stopped cloud-init-nonet
-   stopped cloud-init-local )
 
 task