# -*- coding: utf-8 -*- # memorystore.py # Copyright (C) 2014 LEAP # # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program. If not, see . """ In-memory transient store for a LEAPIMAPServer. """ import contextlib import logging import threading import weakref from collections import defaultdict from copy import copy from twisted.internet import defer from twisted.internet.task import LoopingCall from twisted.python import log from zope.interface import implements from leap.common.check import leap_assert_type from leap.mail import size from leap.mail.decorators import deferred from leap.mail.utils import empty from leap.mail.messageflow import MessageProducer from leap.mail.imap import interfaces from leap.mail.imap.fields import fields from leap.mail.imap.messageparts import MessagePartType, MessagePartDoc from leap.mail.imap.messageparts import RecentFlagsDoc from leap.mail.imap.messageparts import MessageWrapper from leap.mail.imap.messageparts import ReferenciableDict logger = logging.getLogger(__name__) # The default period to do writebacks to the permanent # soledad storage, in seconds. SOLEDAD_WRITE_PERIOD = 10 @contextlib.contextmanager def set_bool_flag(obj, att): """ Set a boolean flag to True while we're doing our thing. Just to let the world know. """ setattr(obj, att, True) try: yield True except RuntimeError as exc: logger.exception(exc) finally: setattr(obj, att, False) class MemoryStore(object): """ An in-memory store to where we can write the different parts that we split the messages into and buffer them until we write them to the permanent storage. It uses MessageWrapper instances to represent the message-parts, which are indexed by mailbox name and UID. It also can be passed a permanent storage as a paremeter (any implementor of IMessageStore, in this case a SoledadStore). In this case, a periodic dump of the messages stored in memory will be done. The period of the writes to the permanent storage is controled by the write_period parameter in the constructor. """ implements(interfaces.IMessageStore, interfaces.IMessageStoreWriter) # TODO We will want to index by chash when we transition to local-only # UIDs. WRITING_FLAG = "_writing" _last_uid_lock = threading.Lock() def __init__(self, permanent_store=None, write_period=SOLEDAD_WRITE_PERIOD): """ Initialize a MemoryStore. :param permanent_store: a IMessageStore implementor to dump messages to. :type permanent_store: IMessageStore :param write_period: the interval to dump messages to disk, in seconds. :type write_period: int """ self._permanent_store = permanent_store self._write_period = write_period # Internal Storage: messages self._msg_store = {} # Internal Storage: payload-hash """ {'phash': weakreaf.proxy(dict)} """ self._phash_store = {} # Internal Storage: content-hash:fdoc """ chash-fdoc-store keeps references to the flag-documents indexed by content-hash. {'chash': {'mbox-a': weakref.proxy(dict), 'mbox-b': weakref.proxy(dict)} } """ self._chash_fdoc_store = {} # Internal Storage: recent-flags store """ recent-flags store keeps one dict per mailbox, with the document-id of the u1db document and the set of the UIDs that have the recent flag. {'mbox-a': {'doc_id': 'deadbeef', 'set': {1,2,3,4} } } """ # TODO this will have to transition to content-hash # indexes after we move to local-only UIDs. self._rflags_store = defaultdict( lambda: {'doc_id': None, 'set': set([])}) """ last-uid store keeps the count of the highest UID per mailbox. {'mbox-a': 42, 'mbox-b': 23} """ self._last_uid = {} # New and dirty flags, to set MessageWrapper State. self._new = set([]) self._new_deferreds = {} self._dirty = set([]) self._rflags_dirty = set([]) self._dirty_deferreds = {} # Flag for signaling we're busy writing to the disk storage. setattr(self, self.WRITING_FLAG, False) if self._permanent_store is not None: # this producer spits its messages to the permanent store # consumer using a queue. We will use that to put # our messages to be written. self.producer = MessageProducer(permanent_store, period=0.1) # looping call for dumping to SoledadStore self._write_loop = LoopingCall(self.write_messages, permanent_store) # We can start the write loop right now, why wait? self._start_write_loop() def _start_write_loop(self): """ Start loop for writing to disk database. """ if not self._write_loop.running: self._write_loop.start(self._write_period, now=True) def _stop_write_loop(self): """ Stop loop for writing to disk database. """ if self._write_loop.running: self._write_loop.stop() # IMessageStore # XXX this would work well for whole message operations. # We would have to add a put_flags operation to modify only # the flags doc (and set the dirty flag accordingly) def create_message(self, mbox, uid, message, notify_on_disk=True): """ Create the passed message into this MemoryStore. By default we consider that any message is a new message. :param mbox: the mailbox :type mbox: basestring :param uid: the UID for the message :type uid: int :param message: a to be added :type message: MessageWrapper :param notify_on_disk: :type notify_on_disk: bool :return: a Deferred. if notify_on_disk is True, will be fired when written to the db on disk. Otherwise will fire inmediately :rtype: Deferred """ print "adding new doc to memstore %s (%s)" % (mbox, uid) key = mbox, uid self._add_message(mbox, uid, message, notify_on_disk) d = defer.Deferred() d.addCallback(lambda result: log.msg("message save: %s" % result)) self._new.add(key) # We store this deferred so we can keep track of the pending # operations internally. self._new_deferreds[key] = d if notify_on_disk: # Caller wants to be notified when the message is on disk # so we pass the deferred that will be fired when the message # has been written. return d else: # Caller does not care, just fired and forgot, so we pass # a defer that will inmediately have its callback triggered. return defer.succeed('fire-and-forget:%s' % str(key)) def put_message(self, mbox, uid, message, notify_on_disk=True): """ Put an existing message. :param mbox: the mailbox :type mbox: basestring :param uid: the UID for the message :type uid: int :param message: a to be added :type message: MessageWrapper :param notify_on_disk: :type notify_on_disk: bool :return: a Deferred. if notify_on_disk is True, will be fired when written to the db on disk. Otherwise will fire inmediately :rtype: Deferred """ key = mbox, uid d = defer.Deferred() d.addCallback(lambda result: log.msg("message PUT save: %s" % result)) self._dirty.add(key) self._dirty_deferreds[key] = d self._add_message(mbox, uid, message, notify_on_disk) #print "dirty ", self._dirty #print "new ", self._new return d def _add_message(self, mbox, uid, message, notify_on_disk=True): # XXX have to differentiate between notify_new and notify_dirty # TODO defaultdict the hell outa here... key = mbox, uid msg_dict = message.as_dict() FDOC = MessagePartType.fdoc.key HDOC = MessagePartType.hdoc.key CDOCS = MessagePartType.cdocs.key DOCS_ID = MessagePartType.docs_id.key try: store = self._msg_store[key] except KeyError: self._msg_store[key] = {FDOC: {}, HDOC: {}, CDOCS: {}, DOCS_ID: {}} store = self._msg_store[key] fdoc = msg_dict.get(FDOC, None) if fdoc: if not store.get(FDOC, None): store[FDOC] = ReferenciableDict({}) store[FDOC].update(fdoc) # content-hash indexing chash = fdoc.get(fields.CONTENT_HASH_KEY) chash_fdoc_store = self._chash_fdoc_store if not chash in chash_fdoc_store: chash_fdoc_store[chash] = {} chash_fdoc_store[chash][mbox] = weakref.proxy( store[FDOC]) hdoc = msg_dict.get(HDOC, None) if hdoc: if not store.get(HDOC, None): store[HDOC] = ReferenciableDict({}) store[HDOC].update(hdoc) docs_id = msg_dict.get(DOCS_ID, None) if docs_id: if not store.get(DOCS_ID, None): store[DOCS_ID] = {} store[DOCS_ID].update(docs_id) cdocs = message.cdocs for cdoc_key in cdocs.keys(): if not store.get(CDOCS, None): store[CDOCS] = {} cdoc = cdocs[cdoc_key] # first we make it weak-referenciable referenciable_cdoc = ReferenciableDict(cdoc) store[CDOCS][cdoc_key] = referenciable_cdoc phash = cdoc.get(fields.PAYLOAD_HASH_KEY, None) if not phash: continue self._phash_store[phash] = weakref.proxy(referenciable_cdoc) def prune(seq, store): for key in seq: if key in store and empty(store.get(key)): store.pop(key) prune((FDOC, HDOC, CDOCS, DOCS_ID), store) #print "after adding: " #import pprint; pprint.pprint(self._msg_store[key]) def get_docid_for_fdoc(self, mbox, uid): """ Get Soledad document id for the flags-doc for a given mbox and uid. """ fdoc = self._permanent_store.get_flags_doc(mbox, uid) if not fdoc: return None doc_id = fdoc.doc_id return doc_id def get_message(self, mbox, uid): """ Get a MessageWrapper for the given mbox and uid combination. :return: MessageWrapper or None """ key = mbox, uid msg_dict = self._msg_store.get(key, None) if msg_dict: new, dirty = self._get_new_dirty_state(key) return MessageWrapper(from_dict=msg_dict, new=new, dirty=dirty, memstore=weakref.proxy(self)) else: return None def remove_message(self, mbox, uid): """ Remove a Message from this MemoryStore. """ # XXX For the moment we are only removing the flags and headers # docs. The rest we leave there polluting your hard disk, # until we think about a good way of deorphaning. # XXX implement elijah's idea of using a PUT document as a # token to ensure consistency in the removal. try: key = mbox, uid self._new.discard(key) self._dirty.discard(key) self._msg_store.pop(key, None) except Exception as exc: logger.exception(exc) # IMessageStoreWriter def write_messages(self, store): """ Write the message documents in this MemoryStore to a different store. """ # For now, we pass if the queue is not empty, to avoid duplicate # queuing. # We would better use a flag to know when we've already enqueued an # item. # XXX this could return the deferred for all the enqueued operations if not self.producer.is_queue_empty(): return print "Writing messages to Soledad..." with set_bool_flag(self, self.WRITING_FLAG): for rflags_doc_wrapper in self.all_rdocs_iter(): self.producer.push(rflags_doc_wrapper) for msg_wrapper in self.all_new_dirty_msg_iter(): self.producer.push(msg_wrapper) # MemoryStore specific methods. def get_uids(self, mbox): """ Get all uids for a given mbox. """ all_keys = self._msg_store.keys() return [uid for m, uid in all_keys if m == mbox] # last_uid def get_last_uid(self, mbox): """ Get the highest UID for a given mbox. It will be the highest between the highest uid in the message store for the mailbox, and the soledad integer cache. """ uids = self.get_uids(mbox) last_mem_uid = uids and max(uids) or 0 last_soledad_uid = self.get_last_soledad_uid(mbox) return max(last_mem_uid, last_soledad_uid) def get_last_soledad_uid(self, mbox): """ Get last uid for a given mbox from the soledad integer cache. """ return self._last_uid.get(mbox, 0) def set_last_soledad_uid(self, mbox, value): """ Set last uid for a given mbox in the soledad integer cache. SoledadMailbox should prime this value during initialization. Other methods (during message adding) SHOULD call `increment_last_soledad_uid` instead. """ leap_assert_type(value, int) print "setting last soledad uid for ", mbox, "to", value # if we already have a vlue here, don't do anything with self._last_uid_lock: if not self._last_uid.get(mbox, None): self._last_uid[mbox] = value def increment_last_soledad_uid(self, mbox): """ Increment by one the soledad integer cache for the last_uid for this mbox, and fire a defer-to-thread to update the soledad value. The caller should lock the call tho this method. """ with self._last_uid_lock: self._last_uid[mbox] += 1 value = self._last_uid[mbox] self.write_last_uid(mbox, value) return value @deferred def write_last_uid(self, mbox, value): """ Increment the soledad cache, """ leap_assert_type(value, int) if self._permanent_store: self._permanent_store.write_last_uid(mbox, value) # Counting sheeps... def count_new_mbox(self, mbox): """ Count the new messages by inbox. """ return len([(m, uid) for m, uid in self._new if mbox == mbox]) def count_new(self): """ Count all the new messages in the MemoryStore. """ return len(self._new) def get_cdoc_from_phash(self, phash): """ Return a content-document by its payload-hash. """ doc = self._phash_store.get(phash, None) # XXX return None for consistency? # XXX have to keep a mapping between phash and its linkage # info, to know if this payload is been already saved or not. # We will be able to get this from the linkage-docs, # not yet implemented. new = True dirty = False return MessagePartDoc( new=new, dirty=dirty, store="mem", part=MessagePartType.cdoc, content=doc, doc_id=None) def get_fdoc_from_chash(self, chash, mbox): """ Return a flags-document by its content-hash and a given mailbox. :return: MessagePartDoc, or None. """ docs_dict = self._chash_fdoc_store.get(chash, None) fdoc = docs_dict.get(mbox, None) if docs_dict else None # a couple of special cases. # 1. We might have a doc with empty content... if empty(fdoc): return None # 2. ...Or the message could exist, but being flagged for deletion. # We want to create a new one in this case. # Hmmm what if the deletion is un-done?? We would end with a # duplicate... if fdoc and fields.DELETED_FLAG in fdoc[fields.FLAGS_KEY]: return None # XXX get flags new = True dirty = False return MessagePartDoc( new=new, dirty=dirty, store="mem", part=MessagePartType.fdoc, content=fdoc, doc_id=None) def all_msg_iter(self): """ Return generator that iterates through all messages in the store. """ return (self.get_message(*key) for key in sorted(self._msg_store.keys())) def all_new_dirty_msg_iter(self): """ Return geneator that iterates through all new and dirty messages. """ return (self.get_message(*key) for key in sorted(self._msg_store.keys()) if key in self._new or key in self._dirty) def all_msg_dict_for_mbox(self, mbox): """ Return all the message dicts for a given mbox. """ return [self._msg_store[(mb, uid)] for mb, uid in self._msg_store if mb == mbox] def all_deleted_uid_iter(self, mbox): """ Return generator that iterates through the UIDs for all messags with deleted flag in a given mailbox. """ all_deleted = [ msg['fdoc']['uid'] for msg in self.all_msg_dict_for_mbox(mbox) if msg.get('fdoc', None) and fields.DELETED_FLAG in msg['fdoc']['flags']] return all_deleted # new, dirty flags def _get_new_dirty_state(self, key): """ Return `new` and `dirty` flags for a given message. """ # XXX should return *first* the news, and *then* the dirty... return map(lambda _set: key in _set, (self._new, self._dirty)) def set_new(self, key): """ Add the key value to the `new` set. """ self._new.add(key) def unset_new(self, key): """ Remove the key value from the `new` set. """ #print "Unsetting NEW for: %s" % str(key) self._new.discard(key) deferreds = self._new_deferreds d = deferreds.get(key, None) if d: # XXX use a namedtuple for passing the result # when we check it in the other side. d.callback('%s, ok' % str(key)) deferreds.pop(key) def set_dirty(self, key): """ Add the key value to the `dirty` set. """ self._dirty.add(key) def unset_dirty(self, key): """ Remove the key value from the `dirty` set. """ #print "Unsetting DIRTY for: %s" % str(key) self._dirty.discard(key) deferreds = self._dirty_deferreds d = deferreds.get(key, None) if d: # XXX use a namedtuple for passing the result # when we check it in the other side. d.callback('%s, ok' % str(key)) deferreds.pop(key) # Recent Flags # TODO --- nice but unused def set_recent_flag(self, mbox, uid): """ Set the `Recent` flag for a given mailbox and UID. """ self._rflags_dirty.add(mbox) self._rflags_store[mbox]['set'].add(uid) # TODO --- nice but unused def unset_recent_flag(self, mbox, uid): """ Unset the `Recent` flag for a given mailbox and UID. """ self._rflags_store[mbox]['set'].discard(uid) def set_recent_flags(self, mbox, value): """ Set the value for the set of the recent flags. Used from the property in the MessageCollection. """ self._rflags_dirty.add(mbox) self._rflags_store[mbox]['set'] = set(value) def load_recent_flags(self, mbox, flags_doc): """ Load the passed flags document in the recent flags store, for a given mailbox. :param flags_doc: A dictionary containing the `doc_id` of the Soledad flags-document for this mailbox, and the `set` of uids marked with that flag. """ self._rflags_store[mbox] = flags_doc def get_recent_flags(self, mbox): """ Get the set of UIDs with the `Recent` flag for this mailbox. :return: set, or None """ rflag_for_mbox = self._rflags_store.get(mbox, None) if not rflag_for_mbox: return None return self._rflags_store[mbox]['set'] def all_rdocs_iter(self): """ Return an iterator through all in-memory recent flag dicts, wrapped under a RecentFlagsDoc namedtuple. Used for saving to disk. :rtype: generator """ # XXX use enums DOC_ID = "doc_id" SET = "set" rflags_store = self._rflags_store def get_rdoc(mbox, rdict): mbox_rflag_set = rdict[SET] recent_set = copy(mbox_rflag_set) # zero it! mbox_rflag_set.difference_update(mbox_rflag_set) return RecentFlagsDoc( doc_id=rflags_store[mbox][DOC_ID], content={ fields.TYPE_KEY: fields.TYPE_RECENT_VAL, fields.MBOX_KEY: mbox, fields.RECENTFLAGS_KEY: list(recent_set) }) return (get_rdoc(mbox, rdict) for mbox, rdict in rflags_store.items() if not empty(rdict[SET])) # Methods that mirror the IMailbox interface def remove_all_deleted(self, mbox): """ Remove all messages flagged \\Deleted from this Memory Store only. Called from `expunge` """ mem_deleted = self.all_deleted_uid_iter(mbox) for uid in mem_deleted: self.remove_message(mbox, uid) return mem_deleted def expunge(self, mbox): """ Remove all messages flagged \\Deleted, from the Memory Store and from the permanent store also. """ # TODO expunge should add itself as a callback to the ongoing # writes. soledad_store = self._permanent_store try: # 1. Stop the writing call self._stop_write_loop() # 2. Enqueue a last write. #self.write_messages(soledad_store) # 3. Should wait on the writebacks to finish ??? # FIXME wait for this, and add all the rest of the method # as a callback!!! except Exception as exc: logger.exception(exc) # Now, we...: try: # 1. Delete all messages marked as deleted in soledad. # XXX this could be deferred for faster operation. if soledad_store: sol_deleted = soledad_store.remove_all_deleted(mbox) else: sol_deleted = [] # 2. Delete all messages marked as deleted in memory. mem_deleted = self.remove_all_deleted(mbox) all_deleted = set(mem_deleted).union(set(sol_deleted)) print "deleted ", all_deleted except Exception as exc: logger.exception(exc) finally: self._start_write_loop() return all_deleted # Dump-to-disk controls. @property def is_writing(self): """ Property that returns whether the store is currently writing its internal state to a permanent storage. Used to evaluate whether the CHECK command can inform that the field is clear to proceed, or waiting for the write operations to complete is needed instead. :rtype: bool """ # FIXME this should return a deferred !!! # XXX ----- can fire when all new + dirty deferreds # are done (gatherResults) return getattr(self, self.WRITING_FLAG) def put_part(self, part_type, value): """ Put the passed part into this IMessageStore. `part` should be one of: fdoc, hdoc, cdoc """ # XXX turn that into a enum # Memory management. def get_size(self): """ Return the size of the internal storage. Use for calculating the limit beyond which we should flush the store. """ return size.get_size(self._msg_store)