Add optional file-based listings caching

Author: gutzbenj

docs/source/api.rst

@@ -47,7 +47,9 @@ Base Classes
    fsspec.core.OpenFiles
    fsspec.core.get_fs_token_paths
    fsspec.core.url_to_fs
-   fsspec.dircache.DirCache
+   fsspec.dircache.DisabledListingsCache
+   fsspec.dircache.MemoryListingsCache
+   fsspec.dircache.FileListingsCache
    fsspec.FSMap
    fsspec.generic.GenericFileSystem
    fsspec.registry.register_implementation
@@ -82,7 +84,13 @@ Base Classes
 
 .. autofunction:: fsspec.core.url_to_fs
 
-.. autoclass:: fsspec.dircache.DirCache
+.. autoclass:: fsspec.dircache.DisabledListingsCache
+   :members: __init__
+
+.. autoclass:: fsspec.dircache.MemoryListingsCache
+   :members: __init__
+
+.. autoclass:: fsspec.dircache.FileListingsCache
    :members: __init__
 
 .. autoclass:: fsspec.FSMap

docs/source/changelog.rst

@@ -1,6 +1,14 @@
 Changelog
 =========
 
+Dev
+--------
+
+Enhancements
+
+- add file-based listing cache using diskcache (#895)
+  warning: use new ``listings_cache_options`` instead of ``use_listings_cache`` etc.
+
 2024.3.1
 --------
 

docs/source/features.rst

@@ -181,15 +181,26 @@ Listings Caching
 ----------------
 
 For some implementations, getting file listings (i.e., ``ls`` and anything that
-depends on it) is expensive. These implementations use dict-like instances of
-:class:`fsspec.dircache.DirCache` to manage the listings.
-
-The cache allows for time-based expiry of entries with the ``listings_expiry_time``
-parameter, or LRU expiry with the ``max_paths`` parameter. These can be
-set on any implementation instance that uses listings caching; or to skip the
-caching altogether, use ``use_listings_cache=False``. That would be appropriate
-when the target location is known to be volatile because it is being written
-to from other sources.
+depends on it) is expensive. These implementations maye use either dict-like instances of
+:class:`fsspec.dircache.MemoryListingsCache` or file-based caching with instances of
+:class:`fsspec.dircache.FileListingsCache` to manage the listings.
+
+The listings cache can be controlled via the keyword ``listings_cache_options`` which is a dictionary.
+The type of cache that is used, can be controlled via the keyword ``cache_type`` (`disabled`, `memory` or `file`).
+The cache allows for time-based expiry of entries with the keyword ``expiry_time``. If the target location is known to
+be volatile because e.g. it is being written to from other sources we recommend to disable the listings cache.
+If you want to use the file-based caching, you can also provide the argument
+``directory`` to determine where the cache file is stored.
+
+Example for ``listings_cache_options``:
+
+.. code-block:: json
+
+        {
+            "cache_type": "file",
+            "expiry_time": 3600,
+            "directory": "/tmp/cache"
+        }
 
 When the ``fsspec`` instance writes to the backend, the method ``invalidate_cache``
 is called, so that subsequent listing of the given paths will force a refresh. In

fsspec/archive.py

@@ -49,7 +49,7 @@ def info(self, path, **kwargs):
     def ls(self, path, detail=True, **kwargs):
         self._get_dirs()
         paths = {}
-        for p, f in self.dir_cache.items():
+        for p, f in self.dircache.items():
             p = p.rstrip("/")
             if "/" in p:
                 root = p.rsplit("/", 1)[0]

fsspec/asyn.py

@@ -312,15 +312,23 @@ class AsyncFileSystem(AbstractFileSystem):
     mirror_sync_methods = True
     disable_throttling = False
 
-    def __init__(self, *args, asynchronous=False, loop=None, batch_size=None, **kwargs):
+    def __init__(
+        self,
+        *args,
+        asynchronous=False,
+        loop=None,
+        batch_size=None,
+        listings_cache_options=None,
+        **kwargs,
+    ):
         self.asynchronous = asynchronous
         self._pid = os.getpid()
         if not asynchronous:
             self._loop = loop or get_loop()
         else:
             self._loop = None
         self.batch_size = batch_size
-        super().__init__(*args, **kwargs)
+        super().__init__(listings_cache_options, *args, **kwargs)
 
     @property
     def loop(self):

fsspec/dircache.py

@@ -1,9 +1,44 @@
+import logging
 import time
 from collections.abc import MutableMapping
+from enum import Enum
 from functools import lru_cache
+from pathlib import Path
+from typing import Optional, Union
 
+logger = logging.getLogger(__name__)
 
-class DirCache(MutableMapping):
+
+class DisabledListingsCache(MutableMapping):
+    def __init__(self, *args, **kwargs):
+        pass
+
+    def __getitem__(self, item):
+        raise KeyError
+
+    def __setitem__(self, key, value):
+        pass
+
+    def __delitem__(self, key):
+        pass
+
+    def __iter__(self):
+        return iter(())
+
+    def __len__(self):
+        return 0
+
+    def clear(self):
+        pass
+
+    def __contains__(self, item):
+        return False
+
+    def __reduce__(self):
+        return (DisabledListingsCache, ())
+
+
+class MemoryListingsCache(MutableMapping):
     """
     Caching of directory listings, in a structure like::
 
@@ -26,19 +61,14 @@ class DirCache(MutableMapping):
 
     def __init__(
         self,
-        use_listings_cache=True,
-        listings_expiry_time=None,
+        expiry_time=None,
         max_paths=None,
-        **kwargs,
     ):
         """
 
         Parameters
         ----------
-        use_listings_cache: bool
-            If False, this cache never returns items, but always reports KeyError,
-            and setting items has no effect
-        listings_expiry_time: int or float (optional)
+        expiry_time: int or float (optional)
             Time in seconds that a listing is considered valid. If None,
             listings do not expire.
         max_paths: int (optional)
@@ -49,15 +79,14 @@ def __init__(
         self._times = {}
         if max_paths:
             self._q = lru_cache(max_paths + 1)(lambda key: self._cache.pop(key, None))
-        self.use_listings_cache = use_listings_cache
-        self.listings_expiry_time = listings_expiry_time
-        self.max_paths = max_paths
+        self._expiry_time = expiry_time
+        self._max_paths = max_paths
 
     def __getitem__(self, item):
-        if self.listings_expiry_time is not None:
-            if self._times.get(item, 0) - time.time() < -self.listings_expiry_time:
+        if self._expiry_time is not None:
+            if self._times.get(item, 0) - time.time() < -self._expiry_time:
                 del self._cache[item]
-        if self.max_paths:
+        if self._max_paths:
             self._q(item)
         return self._cache[item]  # maybe raises KeyError
 
@@ -75,12 +104,10 @@ def __contains__(self, item):
             return False
 
     def __setitem__(self, key, value):
-        if not self.use_listings_cache:
-            return
-        if self.max_paths:
+        if self._max_paths:
             self._q(key)
         self._cache[key] = value
-        if self.listings_expiry_time is not None:
+        if self._expiry_time is not None:
             self._times[key] = time.time()
 
     def __delitem__(self, key):
@@ -93,6 +120,99 @@ def __iter__(self):
 
     def __reduce__(self):
         return (
-            DirCache,
-            (self.use_listings_cache, self.listings_expiry_time, self.max_paths),
+            MemoryListingsCache,
+            (self._expiry_time, self._max_paths),
+        )
+
+
+class FileListingsCache(MutableMapping):
+    def __init__(
+        self,
+        expiry_time: Optional[int],
+        directory: Optional[Path],
+    ):
+        """
+
+        Parameters
+        ----------
+        expiry_time: int or float (optional)
+            Time in seconds that a listing is considered valid. If None,
+            listings do not expire.
+        directory: str (optional)
+            Directory path at which the listings cache file is stored. If None,
+            an autogenerated path at the user folder is created.
+
+        """
+        try:
+            import platformdirs
+            from diskcache import Cache
+        except ImportError as e:
+            raise ImportError(
+                "The optional dependencies ``platformdirs`` and ``diskcache`` are required for file-based dircache."
+            ) from e
+
+        if not directory:
+            directory = platformdirs.user_cache_dir(appname="fsspec")
+        directory = Path(directory) / "dircache" / str(expiry_time)
+
+        try:
+            directory.mkdir(exist_ok=True, parents=True)
+        except OSError as e:
+            logger.error(f"Directory for dircache could not be created at {directory}.")
+            raise e
+        else:
+            logger.info(f"Dircache located at {directory}.")
+
+        self._expiry_time = expiry_time
+        self._directory = directory
+        self._cache = Cache(directory=str(directory))
+
+    def __getitem__(self, item):
+        """Draw item as fileobject from cache, retry if timeout occurs"""
+        return self._cache.get(key=item, read=True, retry=True)
+
+    def clear(self):
+        self._cache.clear()
+
+    def __len__(self):
+        return len(list(self._cache.iterkeys()))
+
+    def __contains__(self, item):
+        value = self._cache.get(item, retry=True)  # None, if expired
+        if value:
+            return True
+        return False
+
+    def __setitem__(self, key, value):
+        self._cache.set(key=key, value=value, expire=self._expiry_time, retry=True)
+
+    def __delitem__(self, key):
+        del self._cache[key]
+
+    def __iter__(self):
+        return (k for k in self._cache.iterkeys() if k in self)
+
+    def __reduce__(self):
+        return (
+            FileListingsCache,
+            (self._expiry_time, self._directory),
         )
+
+
+class CacheType(Enum):
+    DISABLED = DisabledListingsCache
+    MEMORY = MemoryListingsCache
+    FILE = FileListingsCache
+
+
+def create_listings_cache(
+    cache_type: CacheType,
+    expiry_time: Optional[int],
+    **kwargs,
+) -> Optional[Union[MemoryListingsCache, FileListingsCache]]:
+    cache_map = {
+        CacheType.DISABLED: DisabledListingsCache,
+        CacheType.MEMORY: MemoryListingsCache,
+        CacheType.FILE: FileListingsCache,
+    }
+    return cache_map[cache_type](expiry_time, **kwargs)