Docserver: Persist stat cache for versioned file systems

chrome/common/extensions/docs/server2/caching_file_system.py

 # Copyright (c) 2012 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
+import logging
 import posixpath
 import sys
 
 from file_system import FileSystem, StatInfo, FileNotFoundError
 from future import All, Future
 from path_util import AssertIsDirectory, IsDirectory, ToDirectory
 from third_party.json_schema_compiler.memoize import memoize
 
 
 class CachingFileSystem(FileSystem):
-  '''FileSystem which implements a caching layer on top of |file_system|. It's
-  smart, using Stat() to decided whether to skip Read()ing from |file_system|,
-  and only Stat()ing directories never files.
+  '''FileSystem which implements a caching layer on top of |file_system|. If
+  |fail_on_miss| is True then cache misses throw a FileNotFoundError rather than
+  falling back onto the underlying FileSystem.
+
+  If the underlying FileSystem is versioned (i.e., it implements GetVersion to
+  return something other than None), this will create a persistent stat cache
+  (keyed on the FileSystem instance's version) as an additional optimization.
   '''
-  def __init__(self, file_system, object_store_creator):
+  def __init__(self, file_system, object_store_creator, fail_on_miss=False):
     self._file_system = file_system
-    def create_object_store(category, **optargs):
+    self._fail_on_miss = fail_on_miss
+    def create_object_store(category, try_versioning=False, **optargs):
+      version = file_system.GetVersion()

[not at google - send to devlin, 2014/10/24 00:04:50]

It seems like these changes should be reflected in CachingFileSystemTest

[Ken Rockot(use gerrit already), 2014/10/24 21:26:53]

Done. Added a test to verify caching behavior on versioned file systems.

+      versioned = try_versioning and version is not None

[not at google - send to devlin, 2014/10/24 00:04:50]

And do you need to update CompiledFileSystem at all?

[Ken Rockot(use gerrit already), 2014/10/24 21:26:53]

I don't think so? This change only affects the caching of versioned filesystems,
of which Gitiles is the only one. I haven't given much thought to how/if
CompiledFS could use it, but as long as CompiledFS remains unversioned its
behavior is not impacted by this change.

+      if versioned:
+        identity = '%s/%s' % (file_system.GetIdentity(), version)
+      else:
+        identity = file_system.GetIdentity()
+      optargs['start_empty'] = optargs.get('start_empty', not versioned)

[Ken Rockot(use gerrit already), 2014/10/23 22:36:16]

This is gross. Is there a more clever way to express this kind of behavior? Also
we could just make start_empty a kw arg of create_object_store and nix optargs
altogether.

[not at google - send to devlin, 2014/10/24 00:04:50]

Maybe only have a single flag like "controls_versioning" which controls both
start_empty and versioned? They're mutually exclusive.

(the answer to your question, at face value, is no I don't think there's a more
clever way)

[Ken Rockot(use gerrit already), 2014/10/24 21:26:53]

OK - Leaving as-is then.

       return object_store_creator.Create(
           CachingFileSystem,
-          category='%s/%s' % (file_system.GetIdentity(), category),
+          category='%s/%s' % (identity, category),
           **optargs)
-    self._stat_cache = create_object_store('stat')
+    self._stat_cache = create_object_store('stat', try_versioning=True)
     # The read caches can start populated (start_empty=False) because file
     # updates are picked up by the stat, so it doesn't need the force-refresh
     # which starting empty is designed for. Without this optimisation, cron
     # runs are extra slow.
     self._read_cache = create_object_store('read', start_empty=False)
     self._walk_cache = create_object_store('walk', start_empty=False)
 
   def Refresh(self):
     return self._file_system.Refresh()
 
@@ skipping 12 matching lines @@
       '''
       if path == dir_path:
         return dir_stat
       # Was a file stat. Extract that file.
       file_version = dir_stat.child_versions.get(file_path)
       if file_version is None:
         raise FileNotFoundError('No stat found for %s in %s (found %s)' %
                                 (path, dir_path, dir_stat.child_versions))
       return StatInfo(file_version)
 
+    def raise_cache_miss(path):
+      raise FileNotFoundError('Got cache miss when trying to stat %s' % path)
+
     dir_stat = self._stat_cache.Get(dir_path).Get()
     if dir_stat is not None:
       return Future(callback=lambda: make_stat_info(dir_stat))
 
+    if self._fail_on_miss:
+      logging.warning('Bailing on stat cache miss for %s' % dir_path)
+      return Future(callback=lambda: raise_cache_miss(dir_path))
+
     def next(dir_stat):
       assert dir_stat is not None  # should have raised a FileNotFoundError
       # We only ever need to cache the dir stat.
       self._stat_cache.Set(dir_path, dir_stat)
       return make_stat_info(dir_stat)
     return self._MemoizedStatAsyncFromFileSystem(dir_path).Then(next)
 
   @memoize
   def _MemoizedStatAsyncFromFileSystem(self, dir_path):
     '''This is a simple wrapper to memoize Futures to directory stats, since
@@ skipping 89 matching lines @@
         for f in self.ReadSingle(root).Get():
           if IsDirectory(f):
             dirs.append(f)
           else:
             files.append(f)
         # Update the cache. This is a root -> (dirs, files, version) mapping.
         self._walk_cache.Set(root, (dirs, files, root_stat.version))
       return dirs, files
     return self._file_system.Walk(root, depth=depth, file_lister=file_lister)
 
-  def GetCommitID(self):
-    return self._file_system.GetCommitID()
-
-  def GetPreviousCommitID(self):
-    return self._file_system.GetPreviousCommitID()
-
   def GetIdentity(self):
     return self._file_system.GetIdentity()
 
+  def GetVersion(self):
+    return self._file_system.GetVersion()
+
   def __repr__(self):
     return '%s of <%s>' % (type(self).__name__, repr(self._file_system))