[feat] Sha1 Store enhancements (#1026)

* refactor SHA1_Store

- renamed the class
- added created_on field and refactored how init is done
- wrap long lines

* wrapped longer lines

* rename json file ref_count.json

* make sha1_buf_size an argument

* update gitignore

* added tmp_dir

* added new sha1_store add and tests

* update chdir

* add debug to test

* fixing unit test for 1.8
Co-authored-by: Min Xu <min.xu.public@gmail.com>

.gitignore

@@ -12,10 +12,6 @@
 *.egg-info/
 .testmondata
 
-# experimental weigit
-fairscale/experimental/wgit/dev
-.wgit
-
 # Build and release
 build/
 dist/

fairscale/experimental/wgit/repo.py

@@ -11,7 +11,7 @@ import sys
 from typing import Dict, Tuple, Union
 
 from .pygit import PyGit
-from .sha1_store import SHA1_store
+from .sha1_store import SHA1_Store
 
 
 class Repo:
@@ -42,28 +42,27 @@ class Repo:
         exists = self._exists(self.wgit_parent)
         if not exists and init:
             # No weigit repo exists and is being initialized with init=True
-            # Make .wgit directory, create sha1_refs
+            # Make .wgit directory, create sha1_store
             weigit_dir = self.wgit_parent.joinpath(".wgit")
             weigit_dir.mkdir(parents=False, exist_ok=True)
 
             # Initializing sha1_store only after wgit has been initialized!
-            self._sha1_store = SHA1_store(weigit_dir, init=True)
+            self._sha1_store = SHA1_Store(weigit_dir, init=True)
 
             # # Make the .wgit a git repo
             gitignore_files = [
                 self._sha1_store_path.name,
-                self._sha1_store.ref_file_path.name,
             ]
             self._pygit = PyGit(weigit_dir, gitignore=gitignore_files)
 
         elif exists and init:
             # if weigit repo already exists and init is being called, wrap the existing .wgit/.git repo with PyGit
-            self._sha1_store = SHA1_store(self.path)
+            self._sha1_store = SHA1_Store(self.path)
             self._pygit = PyGit(self.path)
 
         elif exists and not init:
             # weigit exists and non-init commands are triggered
-            self._sha1_store = SHA1_store(self.path)
+            self._sha1_store = SHA1_Store(self.path)
             self._pygit = PyGit(self.path)
 
         else:
@@ -86,7 +85,10 @@ class Repo:
             metadata_file, parent_sha1 = self._process_metadata_file(rel_file_path)
 
             # add the file to the sha1_store
-            sha1_hash = self._sha1_store.add(file_path, parent_sha1)
+            # TODO (Min): We don't add parent sha1 tracking to sha1 store due to
+            #             de-duplication & dependency tracking can create cycles.
+            #             We need to figure out a way to handle deletion.
+            sha1_hash = self._sha1_store.add(file_path)
 
             # write metadata to the metadata-file
             self._write_metadata(metadata_file, file_path, sha1_hash)
@@ -183,7 +185,7 @@ class Repo:
         metadata_d = dict()
         for file in self.path.iterdir():  # iterate over the .wgit directory
             # exlude all the .wgit files and directory
-            if file.name not in {"sha1_store", "sha1_refs.json", ".git", ".gitignore"}:
+            if file.name not in {"sha1_store", ".git", ".gitignore"}:
                 # perform a directory walk on the metadata_file directories to find the metadata files
                 for path in file.rglob("*"):
                     if path.is_file():
@@ -275,16 +277,15 @@ class Repo:
 
     def _weigit_repo_exists(self, check_dir: pathlib.Path) -> bool:
         """Returns True if a valid WeiGit repo exists in the path: check_dir"""
-        wgit_exists, sha1_refs, git_exists, gitignore_exists = self._weight_repo_file_check(check_dir)
-        return wgit_exists and sha1_refs and git_exists and gitignore_exists
+        wgit_exists, git_exists, gitignore_exists = self._weight_repo_file_check(check_dir)
+        return wgit_exists and git_exists and gitignore_exists
 
     def _weight_repo_file_check(self, check_dir: Path) -> tuple:
         """Returns a tuple of boolean corresponding to the existence of each .wgit internally required files."""
         wgit_exists = check_dir.joinpath(".wgit").exists()
-        sha1_refs = check_dir.joinpath(".wgit/sha1_refs.json").exists()
         git_exists = check_dir.joinpath(".wgit/.git").exists()
         gitignore_exists = check_dir.joinpath(".wgit/.gitignore").exists()
-        return wgit_exists, sha1_refs, git_exists, gitignore_exists
+        return wgit_exists, git_exists, gitignore_exists
 
 
 class RepoStatus(Enum):

fairscale/experimental/wgit/sha1_store.py

@@ -4,160 +4,257 @@
 # LICENSE file in the root directory of this source tree.
 
 
+from collections import OrderedDict
 import hashlib
 import json
 from pathlib import Path
 import shutil
 import sys
-from typing import Union
+import tempfile
+import time
+from typing import Any, Dict, Union, cast
+
+import torch
+from torch import Tensor
 
 from .utils import ExitCode
 
+# This is a fixed dir name we use for sha1_store. It should not be changed
+# for backward compatibility reasons.
+SHA1_STORE_DIR_NAME = "sha1_store"
+
 
-class SHA1_store:
+class SHA1_Store:
     """
-    Represent the sha1_store within the WeiGit repo for handling added file to the store and managing references.
+    This class represents a SHA1 checksum based storage dir for state_dict
+    and tensors.
+
+    This means the same content will not be stored multiple times, resulting
+    in space savings. (a.k.a. de-duplication)
+
+    To make things easier for the callers, this class accept input data
+    as files, state_dict or tensors. This class always returns in-memory
+    data, not on-disk files. This class doesn't really care or know the actually
+    data types. It uses torch.save() and torch.load() to do serialization.
+
+    A key issue is dealing with content deletion. We use a reference counting
+    algorithm, which means the caller must have symmetrical add/remove calls
+    for each object.
+
+    We used to support children-parent dependency graph and ref counting, but
+    it is flawed since a grand-child can have the same SHA1 as the grand-parent,
+    resulting in a cycle. This means caller must compute which parent is safe
+    to delete in a version tracking graph. The lesson here is that content
+    addressibility and dependency graphs do not mix well.
 
     Args:
-        weigit_path (pathlib.Path)
-            The path to the weigit repo where a sha1_store will be created, or if already exists will be wrapped.
-        init (bool, optional)
-            - If ``True``, initializes a new sha1_store in the weigit_path. Initialization creates a `sha1_store` directory within WeiGit repo in ./<weigit_path>/,
-                and a `sha1_refs.json` withiin ./<weigit_path>/.
-            - If ``False``, a new sha1_store is not initialized and the existing sha1_store is simply wrapped, populating the `name`, `path` and the `ref_file_path` attributes.
+        parent_path (Path):
+            The parent path in which a SHA1_Store will be created.
+        init (bool, optional):
+            - If ``True``, initializes a new SHA1_Store in the parent_path. Initialization
+              creates a `sha1_store` directory in ./<parent_path>/,
+              and a `ref_count.json` within ./<parent_path>/.
+            - If ``False``, a new `sha1_store` dir is not initialized and the existing
+              `sha1_store` is used to init this class, populating `_json_dict`, and other
+              attributes.
             - Default: False
+        sha1_buf_size (int):
+            Buffer size used for checksumming. Default: 100MB.
+        tmp_dir (str):
+            Dir for temporary files if input is an in-memory object.
     """
 
-    def __init__(self, weigit_path: Path, init: bool = False) -> None:
-        """Create or wrap (if already exists) a sha1_store within the WeiGit repo."""
-        # should use the sha1_refs.json to track the parent references.
-        self.name = "sha1_store"
-        self.path = weigit_path.joinpath(self.name)
-        self.ref_file_path = weigit_path.joinpath("sha1_refs.json")
+    def __init__(
+        self, parent_path: Path, init: bool = False, sha1_buf_size: int = 100 * 1024 * 1024, tmp_dir: str = ""
+    ) -> None:
+        """Create or wrap (if already exists) a sha1_store."""
+        self._path = parent_path.joinpath(SHA1_STORE_DIR_NAME)
+        self._ref_file_path = self._path.joinpath("ref_count.json")
+        self._sha1_buf_size = sha1_buf_size
+        self._json_dict: Dict[str, Any] = {"created_on": time.ctime()}
 
-        self._weigit_path = weigit_path
-        # initialize the sha1_store
-        if init:
+        # Initialize the sha1_store if not exist and init==True.
+        if init and not self._path.exists():
             try:
-                if not self.path.exists():
-                    Path.mkdir(self.path, parents=False, exist_ok=False)
-                    self.ref_file_path.touch(exist_ok=False)
+                Path.mkdir(self._path, parents=False, exist_ok=False)
             except FileExistsError as error:
                 sys.stderr.write(f"An exception occured while creating Sha1_store: {repr(error)}\n")
                 sys.exit(ExitCode.FILE_EXISTS_ERROR)
+            # Create a new json file for this new store.
+            self._store_json_dict()
 
-    def add(self, file_path: Path, parent_sha1: str) -> str:
+        # This is an internal error since caller of this our own wgit code.
+        assert self._path.exists(), "SHA1 store does not exist and init==False"
+
+        # Init temp dir.
+        if tmp_dir:
+            # Caller supplied tmp dir
+            assert Path(tmp_dir).is_dir(), "incorrect input"
+            self._tmp_dir = Path(tmp_dir)
+        else:
+            # Default tmp dir, need to clean it.
+            self._tmp_dir = self._path.joinpath("tmp")
+            shutil.rmtree(self._tmp_dir, ignore_errors=True)
+            self._tmp_dir.mkdir()
+
+    def _load_json_dict(self) -> None:
+        """Loading json dict from disk."""
+        with open(self._ref_file_path, "r") as f:
+            self._json_dict = json.load(f)
+
+    def _store_json_dict(self) -> None:
+        """Storing json dict to disk."""
+        with open(self._ref_file_path, "w", encoding="utf-8") as f:
+            json.dump(self._json_dict, f, ensure_ascii=False, indent=4)
+
+    def add(self, file_or_obj: Union[Path, Tensor, OrderedDict]) -> str:
         """
-        Adds a file/checkpoint to the internal sha1_store and the sha1 references accordingly.
-        First, a sha1 hash is calculated. Utilizing the sha1 hash string, the actual file in <in_file_path> is moved
-        within the sha1_store and the sha1 reference file is updated accordingly with the information of their parents
-        node (if exists) and whether the new version is a leaf node or not.
+        Adds a file/object to the internal sha1_store and the sha1 references
+        accordingly.
+
+        First, a sha1 hash is calculated. Utilizing the sha1 hash string, the actual file
+        in <file_or_obj> is moved within the sha1_store and the reference file is updated.
+        If the input is an object, it will be store in the self._tmp_dir and then moved.
 
         Args:
-            in_file_path (str): path to the file to be added to the sha1_store.
+            file_or_obj (str or tensor or OrderedDict):
+                Path to the file to be added to the sha1_store or an in-memory object
+                that can be handled by torch.save.
         """
+        # Use `isinstance` not type() == Path since pathlib returns OS specific
+        # Path types, which inherit from the Path class.
+        if isinstance(file_or_obj, (Path, str)):
+            # Make sure it is a valid file.
+            torch.load(cast(Union[Path, str], file_or_obj))
+            file_path = Path(file_or_obj)
+            remove_tmp = False
+        elif isinstance(file_or_obj, (Tensor, OrderedDict)):
+            # Serialize the object into a tmp file.
+            file_path = self._get_tmp_file_path()
+            torch.save(cast(Union[Tensor, OrderedDict], file_or_obj), file_path)
+            remove_tmp = True
+        else:
+            assert False, f"incorrect input {type(file_or_obj)}"
+
+        # Get SHA1 from the file.
+        assert isinstance(file_path, Path), type(file_path)
         sha1_hash = self._get_sha1_hash(file_path)
-        # use the sha1_hash to create a directory with first2 sha naming convention
+
+        # Add reference.
+        ref_count = self._add_ref(sha1_hash, True)
+
+        if ref_count == 1:
+            # First time adding
+
+            # Create the file dir, if needed.
+            repo_fdir = self._sha1_to_dir(sha1_hash)
+            if not repo_fdir.exists():
                 try:
-            repo_fdir = self.path.joinpath(sha1_hash[:2])
-            repo_fdir.mkdir(exist_ok=True)
+                    repo_fdir.mkdir(exist_ok=True, parents=True)
                 except FileExistsError as error:
                     sys.stderr.write(f"An exception occured: {repr(error)}\n")
                     sys.exit(ExitCode.FILE_EXISTS_ERROR)
+
+            # Transfer the file to the internal sha1_store
+            repo_fpath = repo_fdir.joinpath(sha1_hash)
             try:
-            # First transfer the file to the internal sha1_store
-            repo_fpath = Path.cwd().joinpath(repo_fdir, sha1_hash[2:])
                 shutil.copy2(file_path, repo_fpath)
-
-            # Create the dependency Graph and track reference
-            self._add_ref(sha1_hash, parent_sha1)
-
             except BaseException as error:
-            # in case of failure: Cleans up the sub-directories created to store sha1-named checkpoints
+                # Something went wrong, perhaps out of space, or race condition due to lack of locking.
+                # TODO (Min): proper handle the error and recover when we learn more here.
                 sys.stderr.write(f"An exception occured: {repr(error)}\n")
-            shutil.rmtree(repo_fdir)
+                ref_count = self._add_ref(sha1_hash, False)
+
+        # Clean up if needed.
+        if remove_tmp:
+            file_path.unlink()
+
         return sha1_hash
 
-    def _get_sha1_hash(self, file_path: Union[str, Path]) -> str:
-        """return the sha1 hash of a file
+    def get(self, sha1: str) -> Union[Tensor, OrderedDict]:
+        """Get data from a SHA1
+
+        Args:
+            sha1 (str):
+                SHA1 of the object to get.
+
+        Returns:
+            (Tensor or OrderedDict):
+                In-memory object.
+        """
+        raise NotImplementedError()
+
+    def delete(self, sha1: str) -> None:
+        """Delete a SHA1
 
         Args:
-            file_path (str, Path): Path to the file whose sha1 hash is to be calculalated and returned.
+            sha1 (str):
+                SHA1 of the object to delete.
+
         """
-        SHA1_BUF_SIZE = 104857600  # Reading file in 100MB chunks
+        raise NotImplementedError()
+
+    def _get_sha1_hash(self, file_path: Union[str, Path]) -> str:
+        """Return the sha1 hash of a file
+
+        Args:
+            file_path (str, Path):
+                Path to the file whose sha1 hash is to be calculalated and returned.
 
+        Returns:
+            (str):
+                The SHA1 computed.
+        """
         sha1 = hashlib.sha1()
         with open(file_path, "rb") as f:
             while True:
-                data = f.read(SHA1_BUF_SIZE)
+                data = f.read(self._sha1_buf_size)
                 if not data:
                     break
                 sha1.update(data)
         return sha1.hexdigest()
 
-    def _add_ref(self, current_sha1_hash: str, parent_hash: str) -> None:
+    def _get_tmp_file_path(self) -> Path:
+        """Helper to get a tmp file name under self.tmp_dir."""
+        return Path(tempfile.mkstemp(dir=self._tmp_dir)[1])
+
+    def _sha1_to_dir(self, sha1: str) -> Path:
+        """Helper to get the internal dir for a file based on its SHA1"""
+        # Using first 2 letters of the sha1, which results 26 * 26 = 676 subdirs under the top
+        # level. Then, using another 2 letters for sub-sub-dir. If each dir holds 1000 files, this
+        # can hold 450 millions files.
+        # NOTE: this can NOT be changed for backward compatible reasons once in production.
+        assert len(sha1) > 4, "sha1 too short"
+        part1, part2 = sha1[:2], sha1[2:4]
+        return self._path.joinpath(part1, part2)
+
+    def _add_ref(self, current_sha1_hash: str, inc: bool) -> int:
         """
-        Populates the sha1_refs.json file when file is added and keeps track of reference to earlier file additions.
-        If the sha1_refs.json file is empty, then a new tracking entry of the added file is logged in the sha1_refs file.
-        If the file already has an entry, first it checks if the incoming new added file is a new version of any of the
-        existing entries. If it is, then logs the tracking info as a new version of that existing entry.
-        Otherwise a new entry for the new added file is created for tracking.
+        Update the reference count.
+
+        If the reference counting file does not have this sha1, then a new tracking
+        entry of the added.
 
         Args:
-            file_path (pathlib.Path)
-                Path to the incoming added file.
-            current_sha1_hash (str)
+            current_sha1_hash (str):
                 The sha1 hash of the incoming added file.
+            inc (bool):
+                Increment or decrement.
+
+        Returns:
+            (int):
+                Resulting ref count.
         """
-        # Check the current state of the reference file and check if the added file already has an entry.
-        sha1_refs_empty = self._sha1_refs_file_state()
-
-        # if the file is empty: add the first entry
-        if sha1_refs_empty:
-            with open(self.ref_file_path) as f:
-                ref_data = {current_sha1_hash: {"parent": "ROOT", "ref_count": 1, "is_leaf": True}}
-            self._write_to_json(self.ref_file_path, ref_data)
-        else:
-            # Open sha1 reference file and check if there is a parent_hash not equal to Root?
-            # if Yes, find parent and add the child. Else, just add a new entry
-            with open(self.ref_file_path, "r") as f:
-                ref_data = json.load(f)
-            if parent_hash != "ROOT":
-                # get the last head and replace it's child from HEAD -> this sha1
-                ref_data[parent_hash]["is_leaf"] = False
-                ref_data[current_sha1_hash] = {"parent": parent_hash, "ref_count": 1, "is_leaf": True}
-            else:
-                ref_data[current_sha1_hash] = {"parent": "ROOT", "ref_count": 1, "is_leaf": True}
+        self._load_json_dict()
 
-            self._write_to_json(self.ref_file_path, ref_data)
+        # Init the entry if needed.
+        if current_sha1_hash not in self._json_dict:
+            self._json_dict[current_sha1_hash] = 0
 
-    def _sha1_refs_file_state(self) -> bool:
-        """
-        Checks the state of the sha1 reference file, whether the file is empty or not.
-        If not empty, it checks whether the input file in <file_path> has an older entry (version)
-        in the reference file.
+        # Update the ref count.
+        self._json_dict[current_sha1_hash] += 1 if inc else -1
+        assert self._json_dict[current_sha1_hash] >= 0, "negative ref count"
 
-        Args:
-            file_path (pathlib.Path)
-                input File whose entry will be checked if it exists in the reference file.
-        """
-        try:
-            with open(self.ref_file_path, "r") as f:
-                ref_data = json.load(f)
-            sha1_refs_empty: bool = False
-        except json.JSONDecodeError as error:
-            if not self.ref_file_path.stat().st_size:
-                sha1_refs_empty = True
-        return sha1_refs_empty
-
-    def _write_to_json(self, file: Path, data: dict) -> None:
-        """
-        Populates a json file with data.
-        Args:
-            file (pathlib.Path)
-                path to the file to be written in.
-            data (pathlib.Path)
-                Data to be written in the file.
-        """
-        with open(file, "w", encoding="utf-8") as f:
-            json.dump(data, f, ensure_ascii=False, indent=4)
+        self._store_json_dict()
+
+        return self._json_dict[current_sha1_hash]

stubs/torch/serialization.pyi

 import os
 import pickle
+from pathlib import Path
 from typing import Any, BinaryIO, Callable, IO, Union
 
 DEFAULT_PROTOCOL: int = 2
@@ -7,4 +8,4 @@ DEFAULT_PROTOCOL: int = 2
 def save(obj, f: Union[str, os.PathLike, BinaryIO, IO[bytes]],
          pickle_module: Any=pickle, pickle_protocol: int=DEFAULT_PROTOCOL, _use_new_zipfile_serialization: bool=True) -> None: ...
 
-def load(f: Union[str, BinaryIO], map_location) -> Any: ...
+def load(f: Union[str, BinaryIO, Path], map_location=None) -> Any: ...

tests/experimental/wgit/test_api.py

@@ -10,9 +10,10 @@ import random
 import shutil
 
 import pytest
+import torch
+from torch import nn
 
-from fairscale.experimental.wgit import repo as api
-from fairscale.experimental.wgit.repo import RepoStatus
+from fairscale.experimental.wgit.repo import Repo, RepoStatus
 
 
 @pytest.fixture
@@ -32,14 +33,13 @@ def create_test_dir():
     # create random checkpoints
     size_list = [30e5, 35e5, 40e5, 40e5]
     for i, size in enumerate(size_list):
-        with open(f"checkpoint_{i}.pt", "wb") as f:
-            f.write(os.urandom(int(size)))
+        torch.save(nn.Linear(1, int(size)), f"checkpoint_{i}.pt")
     return test_dir
 
 
 @pytest.fixture
 def repo():
-    repo = api.Repo(Path.cwd(), init=True)
+    repo = Repo(Path.cwd(), init=True)
     return repo
 
 
@@ -48,8 +48,8 @@ def test_setup(create_test_dir):
 
 
 def test_api_init(capsys, repo):
-    repo = api.Repo(Path.cwd(), init=True)
-    assert Path(".wgit/sha1_refs.json").is_file()
+    repo = Repo(Path.cwd(), init=True)
+    assert Path(".wgit/sha1_store").is_dir()
     assert Path(".wgit/.gitignore").is_file()
     assert Path(".wgit/.git").exists()
     assert Path(".wgit/.gitignore").exists()
@@ -80,7 +80,7 @@ def test_api_commit(capsys, repo):
 def test_api_status(capsys, repo):
     # delete the repo and initialize a new one:
     shutil.rmtree(".wgit")
-    repo = api.Repo(Path.cwd(), init=True)
+    repo = Repo(Path.cwd(), init=True)
 
     # check status before any file is added
     out = repo.status()
@@ -99,8 +99,7 @@ def test_api_status(capsys, repo):
     assert out == {key_list[0]: RepoStatus.CLEAN}
 
     # check status after a new change has been made to the file
-    with open(chkpt0, "wb") as f:
-        f.write(os.urandom(int(15e5)))
+    torch.save(nn.Linear(1, int(15e5)), chkpt0)
     out = repo.status()
     assert out == {key_list[0]: RepoStatus.CHANGES_NOT_ADDED}
 

tests/experimental/wgit/test_cli.py

@@ -9,13 +9,18 @@ from pathlib import Path
 import shutil
 
 import pytest
+import torch
+from torch import nn
 
 import fairscale.experimental.wgit.cli as cli
-from fairscale.experimental.wgit.sha1_store import SHA1_store
+from fairscale.experimental.wgit.sha1_store import SHA1_Store
 
 
-@pytest.fixture
+@pytest.fixture(scope="module")
 def create_test_dir():
+    """This setup function runs once per test of this module and
+    it creates a repo, in the process, testing the init function.
+    """
     curr_dir = Path.cwd()
     parent_dir = "experimental"
     test_dir = curr_dir.joinpath(parent_dir, "wgit_testing/")
@@ -30,28 +35,27 @@ def create_test_dir():
     # create random checkpoints
     size_list = [30e5, 35e5, 40e5]
     for i, size in enumerate(size_list):
-        with open(f"checkpoint_{i}.pt", "wb") as f:
-            f.write(os.urandom(int(size)))
-    return test_dir
-
+        torch.save(nn.Linear(1, int(size)), f"checkpoint_{i}.pt")
 
-def test_setup(create_test_dir):
+    # Test init.
     cli.main(["init"])
-    assert str(create_test_dir.stem) == "wgit_testing"
+    assert str(test_dir.stem) == "wgit_testing"
+
+    return test_dir
 
 
-def test_cli_init(capsys):
+def test_cli_init(create_test_dir, capsys):
     # Check if the json and other files have been created by the init
-    assert Path(".wgit/sha1_refs.json").is_file()
+    assert Path(".wgit/sha1_store").is_dir()
     assert Path(".wgit/.gitignore").is_file()
     assert Path(".wgit/.git").exists()
 
 
-def test_cli_add(capsys):
+def test_cli_add(create_test_dir, capsys):
     chkpt0 = "checkpoint_0.pt"
-    cli.main(["add", "checkpoint_0.pt"])
+    cli.main(["add", chkpt0])
 
-    sha1_store = SHA1_store(
+    sha1_store = SHA1_Store(
         Path.cwd().joinpath(".wgit"),
         init=False,
     )

tests/experimental/wgit/test_sha1_store.py

@@ -3,148 +3,111 @@
 # This source code is licensed under the BSD license found in the
 # LICENSE file in the root directory of this source tree.
 
-from dataclasses import dataclass
-import json
 import os
 from pathlib import Path
 import shutil
 
 import pytest
+import torch
+from torch import nn
 
-from fairscale.experimental.wgit.repo import Repo
-from fairscale.experimental.wgit.sha1_store import SHA1_store
+from fairscale.experimental.wgit.sha1_store import SHA1_Store
+from fairscale.internal import torch_version
 
+# Get the absolute path of the parent at the beginning before any os.chdir(),
+# so that we can proper clean it up at any CWD.
+PARENT_DIR = Path("sha1_store_testing").resolve()
 
-@pytest.fixture
-def sha1_configs():
-    @dataclass
-    class Sha1StorePaths:
-        test_dirs = Path("temp_wgit_testing/.wgit")
-        test_path = Path.cwd().joinpath(test_dirs)
-        sha1_ref = test_path.joinpath("sha1_refs.json")
-        chkpt1a_dir = test_path.joinpath("checkpoint_1a")
-        chkpt1b_dir = test_path.joinpath("checkpoint_1b")
-        chkpt1c_dir = test_path.joinpath("checkpoint_1c")
-        checkpoint_1a = test_path.joinpath("checkpoint_1a", "checkpoint_1.pt")
-        checkpoint_1b = test_path.joinpath("checkpoint_1b", "checkpoint_1.pt")
-        checkpoint_1c = test_path.joinpath("checkpoint_1c", "checkpoint_1.pt")
-        checkpoint_2 = test_path.joinpath("checkpoint_1a", "checkpoint_2.pt")
-        checkpoint_3 = test_path.joinpath("checkpoint_1a", "checkpoint_3.pt")
-        metadata_1 = test_path.joinpath("checkpoint_1.pt")
-        metadata_2 = test_path.joinpath("checkpoint_2.pt")
-        metadata_3 = test_path.joinpath("checkpoint_3.pt")
 
-    return Sha1StorePaths
+@pytest.fixture(scope="function")
+def sha1_store(request):
+    """A fixture for setup and teardown.
 
+    This only runs once per test function. So don't make this too slow.
 
-@pytest.fixture
-def sha1_store(sha1_configs):
-    repo = Repo(sha1_configs.test_path.parent, init=False)
-    sha1_store = SHA1_store(sha1_configs.test_dirs, init=False)
-    return repo, sha1_store
+    Tests must be written in a way that either all of the tests run
+    in the order they appears in this file or a specific test is
+    run separately by the user. Either way, the test should work.
+    """
+    # Attach a teardown function.
+    def teardown():
+        os.chdir(PARENT_DIR.joinpath("..").resolve())
+        shutil.rmtree(PARENT_DIR, ignore_errors=True)
 
+    request.addfinalizer(teardown)
 
-def test_setup(sha1_configs):
-    # Set up the testing directory
-    sha1_configs.test_dirs.mkdir(parents=True, exist_ok=True)  # create test .wgit dir
+    # Teardown in case last run didn't clean it up.
+    teardown()
 
-    # Create the test checkpoint files
-    sha1_configs.chkpt1a_dir.mkdir(exist_ok=False)
-    sha1_configs.chkpt1b_dir.mkdir(exist_ok=False)
-    sha1_configs.chkpt1c_dir.mkdir(exist_ok=False)
+    # Get an empty sha1 store.
+    PARENT_DIR.mkdir()
+    sha1_store = SHA1_Store(PARENT_DIR, init=True)
+
+    return sha1_store
+
+
+def test_sha1_add_file(sha1_store):
+    os.chdir(PARENT_DIR)
 
     # Create random checkpoints
     size_list = [25e5, 27e5, 30e5, 35e5, 40e5]
     chkpts = [
-        sha1_configs.checkpoint_1a,
-        sha1_configs.checkpoint_1b,
-        sha1_configs.checkpoint_1c,
-        sha1_configs.checkpoint_2,
-        sha1_configs.checkpoint_3,
+        "checkpoint_1a.pt",
+        "checkpoint_1b.pt",
+        "checkpoint_1c.pt",
+        "checkpoint_2.pt",
+        "checkpoint_3.pt",
     ]
 
     for file, size in zip(chkpts, size_list):
-        with open(file, "wb") as f:
-            f.write(os.urandom(int(size)))
-
-    repo = Repo(sha1_configs.test_path.parent, init=True)
-    sha1_store = SHA1_store(sha1_configs.test_dirs, init=True)
-
-    return sha1_store
-
-
-def test_sha1_add(sha1_configs, sha1_store):
-    repo, sha1_store = sha1_store
-
-    # Add checkpoint_1: Create the meta_data
-    chkpt1 = sha1_configs.checkpoint_1a
-    metadata_file, parent_sha1 = repo._process_metadata_file(chkpt1.name)
-
-    sha1_hash = sha1_store.add(sha1_configs.checkpoint_1a, parent_sha1)
-    repo._write_metadata(metadata_file, chkpt1, sha1_hash)
-
-    # for checkpoint 1
-    metadata_file = sha1_configs.test_path.joinpath(sha1_configs.checkpoint_1a.name)
-
-    with open(metadata_file, "r") as file:
-        metadata = json.load(file)
-    assert metadata["SHA1"]["__sha1_full__"] == sha1_hash
-
-
-def test_sha1_refs(sha1_configs, sha1_store):
-    repo, sha1_store = sha1_store
-
-    def add_checkpoint(checkpoint):
-        metadata_file, parent_sha1 = repo._process_metadata_file(checkpoint.name)
-        sha1_hash = sha1_store.add(checkpoint, parent_sha1)
-        repo._write_metadata(metadata_file, checkpoint, sha1_hash)
-        return sha1_hash
-
-    with open(sha1_configs.sha1_ref, "r") as file:
-        refs_data = json.load(file)
-
-    # get checkpoint1 sha1
-    sha1_chkpt1a_hash = sha1_store._get_sha1_hash(sha1_configs.checkpoint_1a)
-    assert refs_data[sha1_chkpt1a_hash]["parent"] == "ROOT"
-    assert refs_data[sha1_chkpt1a_hash]["ref_count"] == 1
-
-    ck1a_sha1_hash = sha1_store._get_sha1_hash(sha1_configs.checkpoint_1a)
-
-    # add checkpoint new version of checkpoint-1
-    ck1b_sha1_hash = add_checkpoint(sha1_configs.checkpoint_1b)
-
-    # Add new checkpoints 2 and 3
-    ck2_sha1_hash = add_checkpoint(sha1_configs.checkpoint_2)
-    ck3_sha1_hash = add_checkpoint(sha1_configs.checkpoint_3)
-
-    # add another version of checkpoint 1
-    ck1c_sha1_hash = add_checkpoint(sha1_configs.checkpoint_1c)
-
-    # load ref file after Sha1 add
-    with open(sha1_configs.sha1_ref, "r") as file:
-        refs_data = json.load(file)
-
-    # Tests for same file versions
-    assert refs_data[ck1b_sha1_hash]["parent"] == ck1a_sha1_hash
-    assert refs_data[ck1c_sha1_hash]["parent"] == ck1b_sha1_hash
-    assert refs_data[ck1b_sha1_hash]["ref_count"] == 1
-    assert refs_data[ck1a_sha1_hash]["is_leaf"] is False
-    assert refs_data[ck1a_sha1_hash]["is_leaf"] is False
-    assert refs_data[ck1b_sha1_hash]["is_leaf"] is False
-    assert refs_data[ck1c_sha1_hash]["is_leaf"] is True
-
-    # Tests for new files
-    assert refs_data[ck2_sha1_hash]["parent"] == "ROOT"
-    assert refs_data[ck2_sha1_hash]["is_leaf"] is True
-    assert refs_data[ck3_sha1_hash]["parent"] == "ROOT"
-    assert refs_data[ck3_sha1_hash]["is_leaf"] is True
-
-
-def test_tear_down(sha1_configs):
-    # clean up: delete the .wgit directory created during this Test
-    # Making sure the current directory is ./temp_wgit_testing before removing test dir
-    test_parent_dir = sha1_configs.test_path.parent
-    if (test_parent_dir.stem == "temp_wgit_testing") and (sha1_configs.test_path.stem == ".wgit"):
-        shutil.rmtree(test_parent_dir)
-    else:
-        raise Exception("Exception in testing directory tear down!")
+        torch.save(nn.Linear(1, int(size)), file)
+
+    # Add those 5 random files.
+    for c in chkpts:
+        sha1_store.add(c)
+
+    # Add a fixed data twice.
+    module = nn.Linear(100, 100, bias=False)
+    module.weight.data = torch.zeros(100, 100)
+    zeros_file = "zeros.pt"
+    torch.save(module.state_dict(), zeros_file)
+    sha1_store.add(zeros_file)
+    sha1_store.add(zeros_file)
+
+    # Assert the ref counts are 1,1,1,1,1 and 2
+    sha1_store._load_json_dict()
+    json_dict = sha1_store._json_dict
+    if torch_version() >= (1, 9, 0):
+        # torch 1.8 LTS doesn't produce deterministic checkpoint file from fixed tensors/state_dict.
+        key = "da3e19590de8f77fcf7a09c888c526b0149863a0"
+        assert key in json_dict.keys() and json_dict[key] == 2, json_dict
+    del json_dict["created_on"]
+    assert sorted(json_dict.values()) == [1, 1, 1, 1, 1, 2], json_dict
+
+
+def test_sha1_add_state_dict(sha1_store):
+    os.chdir(PARENT_DIR)
+    # add once
+    for i in range(3):
+        sha1_store.add(nn.Linear(10, 10).state_dict())
+    # add twice
+    for i in range(3):
+        sd = nn.Linear(8, 8).state_dict()
+        sha1_store.add(sd)
+        sha1_store.add(sd)
+
+    sha1_store._load_json_dict()
+    json_dict = sha1_store._json_dict
+    del json_dict["created_on"]
+    assert sorted(json_dict.values()) == [1, 1, 1, 2, 2, 2], json_dict
+
+
+def test_sha1_add_tensor(sha1_store):
+    os.chdir(PARENT_DIR)
+    sha1_store.add(torch.Tensor([1.0, 5.5, 3.4]))
+    sha1_store._load_json_dict()
+    json_dict = sha1_store._json_dict
+    if torch_version() >= (1, 9, 0):
+        # torch 1.8 LTS doesn't produce deterministic checkpoint file from fixed tensors/state_dict.
+        key = "71df4069a03a766eacf9f03eea50968e87eae9f8"
+        assert key in json_dict.keys() and json_dict[key] == 1, json_dict