Merge branch '10236-disable-event-listening'
[arvados.git] / services / fuse / arvados_fuse / command.py
index e12d62a5e2e9767e26e10412a3aa712b052ee26c..3f89732bea25dcd1ca546fbef126227e9e0a9256 100644 (file)
@@ -4,6 +4,7 @@ import daemon
 import llfuse
 import logging
 import os
+import resource
 import signal
 import subprocess
 import sys
@@ -76,11 +77,17 @@ class ArgumentParser(argparse.ArgumentParser):
         self.add_argument('--file-cache', type=int, help="File data cache size, in bytes (default 256MiB)", default=256*1024*1024)
         self.add_argument('--directory-cache', type=int, help="Directory data cache size, in bytes (default 128MiB)", default=128*1024*1024)
 
+        self.add_argument('--disable-event-listening', action='store_true', help="Don't subscribe to events on the API server", dest="disable_event_listening", default=False)
+
         self.add_argument('--read-only', action='store_false', help="Mount will be read only (default)", dest="enable_write", default=False)
         self.add_argument('--read-write', action='store_true', help="Mount will be read-write", dest="enable_write", default=False)
 
         self.add_argument('--crunchstat-interval', type=float, help="Write stats to stderr every N seconds (default disabled)", default=0)
 
+        self.add_argument('--unmount-timeout',
+                          type=float, default=2.0,
+                          help="Time to wait for graceful shutdown after --exec program exits and filesystem is unmounted")
+
         self.add_argument('--exec', type=str, nargs=argparse.REMAINDER,
                             dest="exec_args", metavar=('command', 'args', '...', '--'),
                             help="""Mount, run a command, then unmount and exit""")
@@ -90,19 +97,12 @@ class Mount(object):
     def __init__(self, args, logger=logging.getLogger('arvados.arv-mount')):
         self.logger = logger
         self.args = args
+        self.listen_for_events = False
 
         self.args.mountpoint = os.path.realpath(self.args.mountpoint)
         if self.args.logfile:
             self.args.logfile = os.path.realpath(self.args.logfile)
 
-        # Daemonize as early as possible, so we don't accidentally close
-        # file descriptors we're using.
-        self.daemon_ctx = None
-        if not (self.args.exec_args or self.args.foreground):
-            os.chdir(self.args.mountpoint)
-            self.daemon_ctx = daemon.DaemonContext(working_directory='.')
-            self.daemon_ctx.open()
-
         try:
             self._setup_logging()
             self._setup_api()
@@ -113,21 +113,28 @@ class Mount(object):
 
     def __enter__(self):
         llfuse.init(self.operations, self.args.mountpoint, self._fuse_options())
-        if self.args.mode != 'by_pdh':
+        if self.listen_for_events and not self.args.disable_event_listening:
             self.operations.listen_for_events()
-        t = threading.Thread(None, lambda: llfuse.main())
-        t.start()
+        self.llfuse_thread = threading.Thread(None, lambda: self._llfuse_main())
+        self.llfuse_thread.daemon = True
+        self.llfuse_thread.start()
         self.operations.initlock.wait()
+        return self
 
     def __exit__(self, exc_type, exc_value, traceback):
         subprocess.call(["fusermount", "-u", "-z", self.args.mountpoint])
-        self.operations.destroy()
+        self.llfuse_thread.join(timeout=self.args.unmount_timeout)
+        if self.llfuse_thread.is_alive():
+            self.logger.warning("Mount.__exit__:"
+                                " llfuse thread still alive %fs after umount"
+                                " -- abandoning and exiting anyway",
+                                self.args.unmount_timeout)
 
     def run(self):
         if self.args.exec_args:
-            self._run_exec(self.args)
+            self._run_exec()
         else:
-            self._run_standalone(self.args)
+            self._run_standalone()
 
     def _fuse_options(self):
         """FUSE mount options; see mount.fuse(8)"""
@@ -141,8 +148,6 @@ class Mount(object):
         # Configure a log handler based on command-line switches.
         if self.args.logfile:
             log_handler = logging.FileHandler(self.args.logfile)
-        elif self.daemon_ctx:
-            log_handler = logging.NullHandler()
         else:
             log_handler = None
 
@@ -160,8 +165,11 @@ class Mount(object):
         self.api = arvados.safeapi.ThreadSafeApiCache(
             apiconfig=arvados.config.settings(),
             keep_params={
-                "block_cache": arvados.keep.KeepBlockCache(self.args.file_cache)
+                'block_cache': arvados.keep.KeepBlockCache(self.args.file_cache),
+                'num_retries': self.args.retries,
             })
+        # Do a sanity check that we have a working arvados host + token.
+        self.api.users().current().execute()
 
     def _setup_mount(self):
         self.operations = Operations(
@@ -189,6 +197,8 @@ class Mount(object):
 
         if self.args.collection is not None:
             # Set up the request handler with the collection at the root
+            # First check that the collection is readable
+            self.api.collections().get(uuid=self.args.collection).execute()
             self.args.mode = 'collection'
             dir_class = CollectionDirectory
             dir_args.append(self.args.collection)
@@ -234,7 +244,9 @@ class Mount(object):
             mount_readme = True
 
         if dir_class is not None:
-            self.operations.inodes.add_entry(dir_class(*dir_args))
+            ent = dir_class(*dir_args)
+            self.operations.inodes.add_entry(ent)
+            self.listen_for_events = ent.want_event_subscribe()
             return
 
         e = self.operations.inodes.add_entry(Directory(
@@ -264,6 +276,7 @@ class Mount(object):
         if name in ['', '.', '..'] or '/' in name:
             sys.exit("Mount point '{}' is not supported.".format(name))
         tld._entries[name] = self.operations.inodes.add_entry(ent)
+        self.listen_for_events = (self.listen_for_events or ent.want_event_subscribe())
 
     def _readme_text(self, api_host, user_email):
         return '''
@@ -281,58 +294,57 @@ From here, the following directories are available:
 '''.format(api_host, user_email)
 
     def _run_exec(self):
-        # Initialize the fuse connection
-        llfuse.init(self.operations, self.args.mountpoint, self._fuse_options())
-
-        # Subscribe to change events from API server
-        if self.args.mode != 'by_pdh':
-            self.operations.listen_for_events()
-
-        t = threading.Thread(None, lambda: llfuse.main())
-        t.start()
-
-        # wait until the driver is finished initializing
-        self.operations.initlock.wait()
-
         rc = 255
-        try:
-            sp = subprocess.Popen(self.args.exec_args, shell=False)
-
-            # forward signals to the process.
-            signal.signal(signal.SIGINT, lambda signum, frame: sp.send_signal(signum))
-            signal.signal(signal.SIGTERM, lambda signum, frame: sp.send_signal(signum))
-            signal.signal(signal.SIGQUIT, lambda signum, frame: sp.send_signal(signum))
-
-            # wait for process to complete.
-            rc = sp.wait()
-
-            # restore default signal handlers.
-            signal.signal(signal.SIGINT, signal.SIG_DFL)
-            signal.signal(signal.SIGTERM, signal.SIG_DFL)
-            signal.signal(signal.SIGQUIT, signal.SIG_DFL)
-        except Exception as e:
-            self.logger.exception(
-                'arv-mount: exception during exec %s', self.args.exec_args)
+        with self:
             try:
-                rc = e.errno
-            except AttributeError:
-                pass
-        finally:
-            subprocess.call(["fusermount", "-u", "-z", self.args.mountpoint])
-            self.operations.destroy()
+                sp = subprocess.Popen(self.args.exec_args, shell=False)
+
+                # forward signals to the process.
+                signal.signal(signal.SIGINT, lambda signum, frame: sp.send_signal(signum))
+                signal.signal(signal.SIGTERM, lambda signum, frame: sp.send_signal(signum))
+                signal.signal(signal.SIGQUIT, lambda signum, frame: sp.send_signal(signum))
+
+                # wait for process to complete.
+                rc = sp.wait()
+
+                # restore default signal handlers.
+                signal.signal(signal.SIGINT, signal.SIG_DFL)
+                signal.signal(signal.SIGTERM, signal.SIG_DFL)
+                signal.signal(signal.SIGQUIT, signal.SIG_DFL)
+            except Exception as e:
+                self.logger.exception(
+                    'arv-mount: exception during exec %s', self.args.exec_args)
+                try:
+                    rc = e.errno
+                except AttributeError:
+                    pass
         exit(rc)
 
     def _run_standalone(self):
         try:
             llfuse.init(self.operations, self.args.mountpoint, self._fuse_options())
 
+            if not self.args.foreground:
+                self.daemon_ctx = daemon.DaemonContext(
+                    working_directory=os.path.dirname(self.args.mountpoint),
+                    files_preserve=range(
+                        3, resource.getrlimit(resource.RLIMIT_NOFILE)[1]))
+                self.daemon_ctx.open()
+
             # Subscribe to change events from API server
-            self.operations.listen_for_events()
+            if self.listen_for_events and not self.args.disable_event_listening:
+                self.operations.listen_for_events()
 
-            llfuse.main()
+            self._llfuse_main()
         except Exception as e:
             self.logger.exception('arv-mount: exception during mount: %s', e)
             exit(getattr(e, 'errno', 1))
-        finally:
-            self.operations.destroy()
         exit(0)
+
+    def _llfuse_main(self):
+        try:
+            llfuse.main()
+        except:
+            llfuse.close(unmount=False)
+            raise
+        llfuse.close()