3
0
mirror of https://github.com/jlu5/PyLink.git synced 2024-11-30 23:09:23 +01:00
PyLink/structures.py

126 lines
4.4 KiB
Python
Raw Normal View History

"""
structures.py - PyLink data structures module.
This module contains custom data structures that may be useful in various situations.
"""
import collections
import json
class KeyedDefaultdict(collections.defaultdict):
"""
Subclass of defaultdict allowing the key to be passed to the default factory.
"""
def __missing__(self, key):
if self.default_factory is None:
# If there is no default factory, just let defaultdict handle it
super().__missing__(self, key)
else:
value = self[key] = self.default_factory(key)
return value
class DataStore:
# will come into play with subclassing and db version upgrading
initial_version = 1
def __init__(self, name, filename, db_format='json', save_frequency={'seconds': 30}):
self.name = name
self._filename = os.path.abspath(os.path.expanduser(filename))
self._tmp_filename = self._filename + '.tmp'
log.debug('(db:{}) database path set to {}'.format(self.name, self._filename))
self._format = db_format
log.debug('(db:{}) format set to {}'.format(self.name, self._format))
self._save_frequency = timedelta(**save_frequency).total_seconds()
log.debug('(db:{}) saving every {} seconds'.format(self.name, self._save_frequency))
def create_or_load(self):
log.debug('(db:{}) creating/loading datastore using {}'.format(self.name, self._format))
if self._format == 'json':
self._store = {}
self._store_lock = threading.Lock()
log.debug('(db:{}) loading json data store from {}'.format(self.name, self._filename))
try:
self._store = json.loads(open(self._filename, 'r').read())
except (ValueError, IOError, FileNotFoundError):
log.exception('(db:{}) failed to load existing db, creating new one in memory'.format(self.name))
self.put('db.version', self.initial_version)
else:
raise Exception('(db:{}) Data store format [{}] not recognised'.format(self.name, self._format))
def save_callback(self, starting=False):
"""Start the DB save loop."""
if self._format == 'json':
# don't actually save the first time
if not starting:
self.save()
# schedule saving in a loop.
self.exportdb_timer = threading.Timer(self._save_frequency, self.save_callback)
self.exportdb_timer.name = 'PyLink {} save_callback Loop'.format(self.name)
self.exportdb_timer.start()
else:
raise Exception('(db:{}) Data store format [{}] not recognised'.format(self.name, self._format))
def save(self):
log.debug('(db:{}) saving datastore'.format(self.name))
if self._format == 'json':
with open(self._tmp_filename, 'w') as store_file:
store_file.write(json.dumps(self._store))
os.rename(self._tmp_filename, self._filename)
# single keys
def __contains__(self, key):
if self._format == 'json':
return key in self._store
def get(self, key, default=None):
if self._format == 'json':
return self._store.get(key, default)
def put(self, key, value):
if self._format == 'json':
# make sure we can serialize the given data
# so we don't choke later on saving the db out
json.dumps(value)
self._store[key] = value
return True
def delete(self, key):
if self._format == 'json':
try:
with self._store_lock:
del self._store[key]
except KeyError:
# key is already gone, nothing to do
...
return True
# multiple keys
def list_keys(self, prefix=None):
"""Return all key names. If prefix given, return only keys that start with it."""
if self._format == 'json':
keys = []
with self._store_lock:
for key in self._store:
if prefix is None or key.startswith(prefix):
keys.append(key)
return keys
def delete_keys(self, prefix):
"""Delete all keys with the given prefix."""
if self._format == 'json':
with self._store_lock:
for key in tuple(self._store):
if key.startswith(prefix):
del self._store[key]