Merge branch 'benbrown/lockfiles' into 'master'

Fix file locking errors with higher number of instances

See merge request baserock/ybd!371

2 files changed, 12 insertions, 6 deletions

diff --git a/ybd/app.py b/ybd/app.py
index 36fbebc..1040ab3 100644
--- a/ybd/app.py
+++ b/ybd/app.py
@@ -41,10 +41,14 @@ class RetryException(Exception):
         if config.get('last-retry-time'):
             wait = datetime.datetime.now() - config.get('last-retry-time')
             if wait.seconds < 1 and os.path.exists(lockfile(dn)):
-                with open(lockfile(dn), 'r') as L:
-                    call(['flock', '--shared', '--timeout',
-                          config.get('timeout', '60'), str(L.fileno())])
-                log(dn, 'Finished wait loop', verbose=True)
+                try:
+                    with open(lockfile(dn), 'r') as L:
+                        call(['flock', '--shared', '--timeout',
+                              config.get('timeout', '60'), str(L.fileno())])
+                        log(dn, 'Finished wait loop', verbose=True)
+                except IOError:
+                    # The process that had the lock is finished with it.
+                    pass
         config['last-retry-time'] = datetime.datetime.now()
         config['last-retry-dn'] = dn
         for dirname in config['sandboxes']:
diff --git a/ybd/assembly.py b/ybd/assembly.py
index 9699d6e..1ea9664 100644
--- a/ybd/assembly.py
+++ b/ybd/assembly.py
@@ -210,9 +210,11 @@ def shuffle(contents):
 @contextlib.contextmanager
 def claim(dn):
     with open(lockfile(dn), 'a') as L:
+        locked = False
         try:
             fcntl.flock(L, fcntl.LOCK_EX | fcntl.LOCK_NB)
-        except Exception as e:
+            locked = True
+        except IOError as e:
             if e.errno in (errno.EACCES, errno.EAGAIN):
                 # flock() will report EACCESS or EAGAIN when the lock fails.
                 raise RetryException(dn)
@@ -224,7 +226,7 @@ def claim(dn):
         try:
             yield
         finally:
-            if os.path.isfile(lockfile(dn)):
+            if locked and os.path.isfile(lockfile(dn)):
                 os.remove(lockfile(dn))