Check that all machines are locked, and add an option to lock machines instead of...

diff --git a/teuthology/run.py b/teuthology/run.py
index d96f71f1fee4376e17ab98f97c6aedc447f5267d..8e75cc3a1ae3e950ce666543bb896a1a23c815f1 100644 (file)
--- a/teuthology/run.py
+++ b/teuthology/run.py
@@ -48,6 +48,12 @@ def parse_args():
         '--owner',
         help='job owner'
         )
+    parser.add_argument(
+        '--lock',
+        action='store_true',
+        default=False,
+        help='lock machines for the duration of the run',
+        )
 
     args = parser.parse_args()
     return args
@@ -87,24 +93,13 @@ def main():
             yaml.safe_dump(ctx.config, f, default_flow_style=False)
 
     log.debug('\n  '.join(['Config:', ] + yaml.safe_dump(ctx.config, default_flow_style=False).splitlines()))
-    log.info('Opening connections...')
-
-    from orchestra import connection, remote
-    import orchestra.cluster
-
-    remotes = [remote.Remote(name=t, ssh=connection.connect(t))
-               for t in ctx.config['targets']]
-    ctx.cluster = orchestra.cluster.Cluster()
-    for rem, roles in zip(remotes, ctx.config['roles']):
-        ctx.cluster.add(rem, roles)
 
     ctx.summary = {}
 
-    if ctx.owner is not None:
-        ctx.summary['owner'] = ctx.owner
-    else:
+    if ctx.owner is None:
         from teuthology.misc import get_user
-        ctx.summary['owner'] = get_user()
+        ctx.owner = get_user()
+    ctx.summary['owner'] = ctx.owner
 
     if ctx.description is not None:
         ctx.summary['description'] = ctx.description
@@ -113,7 +108,17 @@ def main():
         assert 'kernel' not in task, \
             'kernel installation shouldn be a base-level item, not part of the tasks list'
 
-    init_tasks = [{'internal.check_conflict': None}]
+    init_tasks = []
+    if ctx.lock:
+        assert 'targets' not in ctx.config, \
+            'You cannot specify targets in a config file when using the --lock option'
+        init_tasks.append({'internal.lock_machines': len(ctx.config['roles'])})
+
+    init_tasks.extend([
+            {'internal.check_lock': None},
+            {'internal.connect': None},
+            {'internal.check_conflict': None},
+            ])
     if 'kernel' in ctx.config:
         init_tasks.append({'kernel': ctx.config['kernel']})
     init_tasks.extend([

diff --git a/teuthology/task/internal.py b/teuthology/task/internal.py
index 305bf11163e41d579365d98f98f0b011ef1268de..bbc31e76a29dbf9756ae7b9f2d2b9b9de7feece0 100644 (file)
--- a/teuthology/task/internal.py
+++ b/teuthology/task/internal.py
@@ -5,6 +5,7 @@ import logging
 import os
 import tarfile
 
+from teuthology import lock
 from teuthology import misc as teuthology
 from teuthology import safepath
 from orchestra import run
@@ -42,6 +43,52 @@ def base(ctx, config):
             )
 
 
+@contextlib.contextmanager
+def lock_machines(ctx, config):
+    log.info('Locking machines...')
+    assert isinstance(config, int), 'config must be an integer'
+    newly_locked = lock.lock_many(config, ctx.owner)
+    if len(newly_locked) != config:
+        log.error('Could not lock enough machines, unlocking and exiting...')
+        for machine in newly_locked:
+            lock.unlock(machine, ctx.owner)
+        assert 0
+    ctx.config['targets'] = newly_locked
+    try:
+        yield
+    finally:
+        log.info('Unlocking machines...')
+        for machine in newly_locked:
+            lock.unlock(machine, ctx.owner)
+
+def check_lock(ctx, config):
+    log.info('Checking locks...')
+    for machine in ctx.config['targets']:
+        status = lock.get_status(machine)
+        log.debug('machine status is %s', repr(status))
+        assert status is not None, \
+            'could not read lock status for {name}'.format(name=machine)
+        assert status['up'], 'machine {name} is marked down'.format(name=machine)
+        assert status['locked'], \
+            'machine {name} is not locked'.format(name=machine)
+        assert status['locked_by'] == ctx.owner, \
+            'machine {name} is locked by {user}, not {owner}'.format(
+            name=machine,
+            user=status['locked_by'],
+            owner=ctx.owner,
+            )
+
+def connect(ctx, config):
+    log.info('Opening connections...')
+    from orchestra import connection, remote
+    import orchestra.cluster
+
+    remotes = [remote.Remote(name=t, ssh=connection.connect(t))
+               for t in ctx.config['targets']]
+    ctx.cluster = orchestra.cluster.Cluster()
+    for rem, roles in zip(remotes, ctx.config['roles']):
+        ctx.cluster.add(rem, roles)
+
 def check_conflict(ctx, config):
     log.info('Checking for old test directory...')
     processes = ctx.cluster.run(