Improve performance

changelog.txt

@@ -7,20 +7,23 @@ The **signac** package follows `semantic versioning <https://semver.org/>`_.
 Version 2
 =========
 
-[2.2.0] -- 2023-xx-xx
+[2.2.0] -- 2024-xx-xx
 ---------------------
 
 Added
 +++++
 
  - Official support for Python 3.12 (#957).
+ - ``Job.statepoint_mapping`` - cached and read only access to job statepoints. Faster than
+   ``Job.statepoint`` (#975).
 
 Changed
 +++++++
 
  - Restrict allowable tar file features in Python 3.12 (#957).
  - linked views now can contain spaces and other characters except directory separators (#926).
  - linked views now can be created on Windows, if 'Developer mode' is enabled (#430).
+ - Increase performance for many usage patterns (#975).
 
 Fixed
 +++++

signac/job.py

@@ -11,6 +11,7 @@
 import shutil
 from copy import deepcopy
 from threading import RLock
+from types import MappingProxyType
 from typing import FrozenSet
 
 from synced_collections.backends.collection_json import (
@@ -248,7 +249,8 @@ class Job:
 
     Jobs can be opened by ``statepoint`` or ``id_``. If both values are
     provided, it is the user's responsibility to ensure that the values
-    correspond.
+    correspond. Set ``directory_known`` to ``True`` when the job directory
+    is known to exist - this skips some expensive isdir checks.
 
     Parameters
     ----------
@@ -258,6 +260,8 @@ class Job:
         State point for the job. (Default value = None)
     id_ : str, optional
         The job identifier. (Default value = None)
+    directory_known : bool, optional
+        Set to true when the job directory is known to exist. (Default value = False)
 
     """
 
@@ -274,30 +278,34 @@ class Job:
     KEY_DATA = "signac_data"
     "The job's datastore key."
 
-    def __init__(self, project, statepoint=None, id_=None):
+    def __init__(self, project, statepoint=None, id_=None, directory_known=False):
         self._project = project
         self._lock = RLock()
         self._initialize_lazy_properties()
+        self._directory_known = directory_known
 
         if statepoint is None and id_ is None:
             raise ValueError("Either statepoint or id_ must be provided.")
         elif statepoint is not None:
-            self._statepoint_requires_init = False
             try:
                 self._id = calc_id(statepoint) if id_ is None else id_
             except TypeError:
                 raise KeyTypeError
-            self._statepoint = _StatePointDict(
-                jobs=[self], filename=self._statepoint_filename, data=statepoint
-            )
 
-            # Update the project's state point cache immediately if opened by state point
-            self._project._register(self.id, statepoint)
+            self._statepoint_mapping = statepoint
+            self._statepoint_requires_init = True
         else:
             # Only an id was provided. State point will be loaded lazily.
             self._id = id_
             self._statepoint_requires_init = True
 
+            # Fetch the statepoint mapping from the project's cache. Don't load it
+            # from disk on a cache miss (will be loaded on demand).
+            try:
+                self._statepoint_mapping = project._sp_cache[id_]
+            except KeyError:
+                self._statepoint_mapping = None
+
     def _initialize_lazy_properties(self):
         """Initialize all properties that are designed to be loaded lazily."""
         with self._lock:
@@ -334,7 +342,7 @@ def __str__(self):
 
     def __repr__(self):
         return "{}(project={}, statepoint={})".format(
-            self.__class__.__name__, repr(self._project), self.statepoint
+            self.__class__.__name__, repr(self._project), self.statepoint_mapping
         )
 
     @property
@@ -406,6 +414,32 @@ def update_statepoint(self, update, overwrite=False):
         statepoint.update(update)
         self.statepoint = statepoint
 
+    @property
+    def statepoint_mapping(self):
+        """Get a copy of the job's statepoint as a read-only mapping.
+
+        `statepoint_mapping` uses the statepoint cache to provide fast access to the
+        job's statepoint for reading.
+
+        .. note::
+
+            Create and update the statepoint cache with ``signac update-cache`` on the
+            command line.
+
+        .. seealso::
+
+            Use `statepoint` to modify the job's statepoint.
+
+        Returns
+        -------
+        Mapping
+            Returns the job's state point.
+        """
+        if self._statepoint_mapping is None:
+            self._statepoint_mapping = self._project._get_statepoint(self._id)
+
+        return MappingProxyType(self._statepoint_mapping)
+
     @property
     def statepoint(self):
         """Get or set the job's state point.
@@ -416,6 +450,10 @@ def statepoint(self):
         `Modifying the State Point
         <https://docs.signac.io/en/latest/jobs.html#modifying-the-state-point>`_.
 
+        .. tip::
+
+            Use `statepoint_mapping` for faster access to read the statepoint.
+
         .. warning::
 
             The state point object behaves like a dictionary in most cases,
@@ -443,14 +481,25 @@ def statepoint(self):
         """
         with self._lock:
             if self._statepoint_requires_init:
-                # Load state point data lazily (on access).
-                self._statepoint = _StatePointDict(
-                    jobs=[self], filename=self._statepoint_filename
-                )
-                statepoint = self._statepoint.load(self.id)
+                if self._statepoint_mapping is None:
+                    # Load state point data lazily (on access).
+                    self._statepoint = _StatePointDict(
+                        jobs=[self],
+                        filename=self._statepoint_filename,
+                    )
+                    statepoint = self._statepoint.load(self.id)
+
+                    # Update the project's state point cache when loaded lazily
+                    self._project._register(self.id, statepoint)
+                    self._statepoint_mapping = statepoint
+                else:
+                    # Create _StatePointDict lazily with a known statepoint dict.
+                    self._statepoint = _StatePointDict(
+                        jobs=[self],
+                        filename=self._statepoint_filename,
+                        data=self._statepoint_mapping,
+                    )
 
-                # Update the project's state point cache when loaded lazily
-                self._project._register(self.id, statepoint)
                 self._statepoint_requires_init = False
 
         return self._statepoint
@@ -510,7 +559,7 @@ def document(self):
         """
         with self._lock:
             if self._document is None:
-                self.init()
+                self.init(validate_statepoint=False)
                 fn_doc = os.path.join(self.path, self.FN_DOCUMENT)
                 self._document = BufferedJSONAttrDict(
                     filename=fn_doc, write_concern=True
@@ -591,9 +640,9 @@ def stores(self):
         """
         with self._lock:
             if self._stores is None:
-                self.init()
+                self.init(validate_statepoint=False)
                 self._stores = H5StoreManager(self.path)
-        return self.init()._stores
+        return self._stores
 
     @property
     def data(self):
@@ -640,7 +689,7 @@ def project(self):
         """
         return self._project
 
-    def init(self, force=False):
+    def init(self, force=False, validate_statepoint=True):
         """Initialize the job's workspace directory.
 
         This function will do nothing if the directory and the job state point
@@ -656,6 +705,10 @@ def init(self, force=False):
             Overwrite any existing state point files, e.g., to repair them if
             they got corrupted (Default value = False).
 
+        validate_statepoint : bool, optional
+            When True (the default), load the job statepoint and ensure that it matches
+            the id. When False, exit early when the job directory exists.
+
         Returns
         -------
         Job
@@ -671,6 +724,15 @@ def init(self, force=False):
         """
         with self._lock:
             try:
+                # Fast early exit when not validating.
+                if not validate_statepoint:
+                    if self._directory_known:
+                        return self
+
+                    if os.path.isdir(self.path):
+                        self._directory_known = True
+                        return self
+
                 # Attempt early exit if the state point file exists and is valid.
                 try:
                     statepoint = self.statepoint.load(self.id)
@@ -687,6 +749,8 @@ def init(self, force=False):
                         )
                         raise
 
+                    self._directory_known = True
+
                     # The state point save will not overwrite an existing file on
                     # disk unless force is True, so the subsequent load will catch
                     # when a preexisting invalid file was present.
@@ -760,6 +824,8 @@ def remove(self):
                     self._document = None
                 self._stores = None
 
+            self._directory_known = False
+
     def move(self, project):
         """Move this job to project.
 
@@ -899,7 +965,7 @@ def open(self):
 
         """
         self._cwd.append(os.getcwd())
-        self.init()
+        self.init(validate_statepoint=False)
         logger.info(f"Enter workspace '{self.path}'.")
         os.chdir(self.path)