Change: set up worker

1 files changed, 107 insertions, 5 deletions

diff --git a/contractor b/contractor
index cc02613..934819b 100755
--- a/contractor
+++ b/contractor
@@ -21,7 +21,13 @@ def ssh(target, argv, quiet=False):
     stdout = None
     if quiet:
         stdout = PIPE
-    return subprocess.call(argv, stderr=STDOUT, stdout=stdout)
+    logging.debug(
+        'EXEC: {!r} stdout={!r} stderr={!r}'.format(
+            argv, stdout, STDOUT))
+    p = subprocess.Popen(argv, stderr=STDOUT, stdout=stdout)
+    out, err = p.communicate()
+    logging.debug('RESULT: {} {!r} {!r}'.format(p.returncode, out, err))
+    return out, err, p.returncode
 
 
 def rsync(filename, target):
@@ -63,11 +69,21 @@ class ContractorApplication(cliapp.Application):
 
         self.verbose(
             'uploading image to worker: {}'.format(bs.worker_image()))
-        m.upload_image(bs.worker_image())
+        m.upload_worker_image(bs.worker_image())
+
+        self.verbose('starting worker')
+        if not m.start_worker():
+            self.error('could not start worker')
+
+        # install = bs.install()
+        # if install:
+        #     self.verbose(
+        #         'installing packages: {}'.format(', '.join(install)))
+        # else:
+        #     self.verbose('no packages to install')
+        # m.install_on
 
-       # pkgs = build.get('install', [])
         # if pkgs:
-        #     self.verbose('Installing packages: {}'.format(', '.join(pkgs)))
         #     self.cmd_setup(
         #         ['env', 'DEBIAN_INTERFACE=noninteractive',
         #          'apt-get', 'install', '-y'] + pkgs)
@@ -370,8 +386,94 @@ class Manager:
         self.virsh(['destroy', 'worker'])
         self.virsh(['undefine', 'worker'])
 
-    def upload_image(self, filename):
+    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))
+
+        status = json.loads(out)
+        if not status:
+            return None
+
+        status.sort(key=lambda e: e['expiry-time'])
+        return status[-1]['ip-address']
+
+    def start_worker(self):
+        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')
+            return None
+
+        out, err, exit_code = self.ssh([
+            'virt-install',
+            '--connect', 'qemu:///system',
+            '--quiet',
+            '--name=worker',
+            '--memory=4096',
+            '--vcpus={}'.format(n-1),
+            '--cpu=host',
+            '--import',
+            '--os-variant=debian9',
+            '--disk=path=temp.img,cache=none',
+            '--disk=path=/dev/vdb,cache=none',
+            '--network=network=default',
+            '--graphics=spice',
+            '--noautoconsole',
+        ])
+        if exit_code != 0:
+            logging.error('Could not create worker VM')
+            return None
+
+        return self.wait_for_worker()
+
+    def wait_for_worker(self):
+        # We look up the IP and try to use it. The IP might be for a
+        # previous worker instance.
+        while True:
+            ip = self.worker_ip()
+            if ip is None:
+                continue
+            w = Worker(self, ip)
+            out, err, exit_code = w.ssh(['true'], quiet=True)
+            if exit_code == 0:
+                return w
+            time.sleep(2)
+
+
+class Worker:
+
+    def __init__(self, manager, ip):
+        self._manager = manager
+        self._target = 'worker@{}'.format(ip)
+
+    def ssh(self, argv, quiet=False):
+        argv = [shlex.quote(arg) for arg in argv]        
+        argv = ['ssh', self._target, '--'] + argv
+        return self._manager.ssh(argv, quiet=quiet)
+        
 
 ContractorApplication().run()