Fix: some error handling

1 files changed, 52 insertions, 40 deletions

diff --git a/contractor b/contractor
index 934819b..5c90bd1 100755
--- a/contractor
+++ b/contractor
@@ -64,24 +64,29 @@ class ContractorApplication(cliapp.Application):
         bs = self.load_build_spec(args[0])
         m = self.manager()
 
+        # This might fail. We ignore the failure.
         self.verbose('stopping worker')
         m.stop_worker()
 
         self.verbose(
             'uploading image to worker: {}'.format(bs.worker_image()))
-        m.upload_worker_image(bs.worker_image())
+        if m.upload_worker_image(bs.worker_image()) != 0:
+            self.error('could not upload image to worker')
+            sys.exit(1)
 
         self.verbose('starting worker')
-        if not m.start_worker():
+        w = m.start_worker()
+        if not w:
             self.error('could not start worker')
+            sys.exit(1)
 
-        # install = bs.install()
-        # if install:
-        #     self.verbose(
-        #         'installing packages: {}'.format(', '.join(install)))
-        # else:
-        #     self.verbose('no packages to install')
-        # m.install_on
+        install = bs.install()
+        if install:
+            self.verbose(
+                'installing packages: {}'.format(', '.join(install)))
+            m.install_on
+        else:
+            self.verbose('no packages to install')
 
         # if pkgs:
         #     self.cmd_setup(
@@ -387,42 +392,17 @@ class Manager:
         self.virsh(['undefine', 'worker'])
 
     def upload_worker_image(self, filename):
-        rsync(filename, '{}:worker.img'.format(self._target))
-
-    def copy_worker_image(self):
-        self.ssh(['cp', 'worker.img', 'temp.img'])
-
-    def get_cpu_count(self):
-        out, err, exit_code = self.ssh(['lscpu'], quiet=True)
-        if exit_code != 0:
-            logging.error('lscpu on manager failed: {!r}'.format(err))
-            return None
-
-        for line in out.decode('UTF8').splitlines():
-            if line.startswith('CPU(s):'):
-                return int(line.split()[-1])
-
-        logging.error('Could not find number of CPUs on manager')
-        return None
-
-    def worker_ip(self):
-        filename = '/var/lib/libvirt/dnsmasq/virbr0.status'
-        out, err, exit_code = self.ssh(['cat', filename], quiet=True)
-        if exit_code != 0:
-            logging.error('Could not read dnsmasq status file')
-            return None
-        logging.debug('virbr0.status from manager: {!r}'.format(out))
+        return rsync(filename, '{}:worker.img'.format(self._target))
 
-        status = json.loads(out)
-        if not status:
+    def start_worker(self):
+        if self.copy_worker_image() != 0:
+            logging.error('Could not copy worker image for new instance')
             return None
 
-        status.sort(key=lambda e: e['expiry-time'])
-        return status[-1]['ip-address']
-
-    def start_worker(self):
+        # We ignore failures on these, as they shouldn't matter.
         self.virsh(['net-autostart', 'default'])
         self.virsh(['net-start', 'default'])
+
         n = self.get_cpu_count()
         if not isinstance(n, int):
             logging.error('Could not start worker due to missing CPU count')
@@ -450,6 +430,38 @@ class Manager:
 
         return self.wait_for_worker()
 
+    def copy_worker_image(self):
+        out, err, exit_code = self.ssh(['cp', 'worker.img', 'temp.img'])
+        return exit_code
+
+    def get_cpu_count(self):
+        out, err, exit_code = self.ssh(['lscpu'], quiet=True)
+        if exit_code != 0:
+            logging.error('lscpu on manager failed: {!r}'.format(err))
+            return None
+
+        for line in out.decode('UTF8').splitlines():
+            if line.startswith('CPU(s):'):
+                return int(line.split()[-1])
+
+        logging.error('Could not find number of CPUs on manager')
+        return None
+
+    def worker_ip(self):
+        filename = '/var/lib/libvirt/dnsmasq/virbr0.status'
+        out, err, exit_code = self.ssh(['cat', filename], quiet=True)
+        if exit_code != 0:
+            logging.error('Could not read dnsmasq status file')
+            return None
+        logging.debug('virbr0.status from manager: {!r}'.format(out))
+
+        status = json.loads(out)
+        if not status:
+            return None
+
+        status.sort(key=lambda e: e['expiry-time'])
+        return status[-1]['ip-address']
+
     def wait_for_worker(self):
         # We look up the IP and try to use it. The IP might be for a
         # previous worker instance.