ZeroNet/src/Site/Site.py

1763 lines
72 KiB
Python

import os
import json
import logging
import re
import time
import random
import sys
import hashlib
import itertools
import collections
import base64
import gevent
import gevent.pool
import gevent.lock
import util
from Config import config
from Peer import Peer
from Worker import WorkerManager
from Debug import Debug
from Content import ContentManager
from .SiteStorage import SiteStorage
from Crypt import CryptHash
from util import helper
from util import Diff
from util import GreenletManager
from Plugin import PluginManager
from File import FileServer
from .SiteAnnouncer import SiteAnnouncer
from . import SiteManager
from . import SiteHelpers
def lerp(val_min, val_max, scale):
return scale * (val_max - val_min) + val_min
class ScaledTimeoutHandler:
def __init__(self, val_min, val_max, handler=None, scaler=None):
self.val_min = val_min
self.val_max = val_max
self.timestamp = 0
self.handler = handler
self.scaler = scaler
self.log = logging.getLogger("ScaledTimeoutHandler")
def isExpired(self, scale):
interval = lerp(self.val_min, self.val_max, scale)
expired_at = self.timestamp + interval
now = time.time()
expired = (now > expired_at)
if expired:
self.log.debug(
"Expired: [%d..%d]: scale=%f, interval=%f",
self.val_min, self.val_max, scale, interval)
return expired
def done(self):
self.timestamp = time.time()
def run(self, *args, **kwargs):
do_run = kwargs["force"] or self.isExpired(self.scaler())
if do_run:
result = self.handler(*args, **kwargs)
if result:
self.done()
return result
else:
return None
class BackgroundPublisher:
def __init__(self, site, published=[], limit=5, inner_path="content.json", diffs={}):
self.site = site
self.threads = gevent.pool.Pool(None)
self.inner_path = inner_path
self.stages = [
{
"interval": ScaledTimeoutHandler(60, 60),
"max_tries": 2,
"tries": 0,
"limit": 0,
"done": False
},
{
"interval": ScaledTimeoutHandler(60 * 10, 60 * 10),
"max_tries": 5,
"tries": 0,
"limit": 0,
"done": False
}
]
self.reinit(published=published, limit=limit, diffs=diffs)
def reinit(self, published=[], limit=5, diffs={}):
self.threads.kill()
self.published = published
self.diffs = diffs
i = 0
for stage in self.stages:
stage["nr"] = i
stage["limit"] = limit * (2 + i)
stage["tries"] = 0
stage["done"] = False
stage["thread"] = None
if i > 0:
stage["interval"].done()
i += 1
def isStageComplete(self, stage):
if not stage["done"]:
stage["done"] = len(self.published) >= stage["limit"]
if not stage["done"]:
stage["done"] = stage["tries"] >= stage["max_tries"]
return stage["done"]
def isComplete(self):
for stage in self.stages:
if not self.isStageComplete(stage):
return False
return True
def process(self):
for stage in self.stages:
if not self.isStageComplete(stage):
self.processStage(stage)
break
return self.isComplete()
def processStage(self, stage):
if not stage["interval"].isExpired(0):
return
if len(self.site.peers) < stage["limit"]:
self.site.announce(mode="more")
if not stage["thread"]:
peers = list(self.site.peers.values())
random.shuffle(peers)
stage["thread"] = self.threads.spawn(self.site.publisher,
self.inner_path, peers, self.published, stage["limit"], diffs=self.diffs, max_retries=1)
stage["tries"] += 1
stage["interval"].done()
self.site.log.info("Background publisher: Stage #%s: %s published to %s/%s peers",
stage["nr"], self.inner_path, len(self.published), stage["limit"])
def finalize(self):
self.threads.kill()
self.site.log.info("Background publisher: Published %s to %s peers", self.inner_path, len(self.published))
@PluginManager.acceptPlugins
class Site(object):
def __init__(self, address, allow_create=True, settings=None):
self.address = str(re.sub("[^A-Za-z0-9]", "", address)) # Make sure its correct address
self.address_hash = hashlib.sha256(self.address.encode("ascii")).digest()
self.address_sha1 = hashlib.sha1(self.address.encode("ascii")).digest()
self.address_short = "%s..%s" % (self.address[:6], self.address[-4:]) # Short address for logging
self.log = logging.getLogger("Site:%s" % self.address_short)
self.addEventListeners()
self.periodic_maintenance_handlers = [
ScaledTimeoutHandler(
config.site_announce_interval_max * 60,
config.site_announce_interval_min * 60,
handler=self.periodicMaintenanceHandler_announce,
scaler=self.getAnnounceRating),
ScaledTimeoutHandler(
config.site_peer_check_interval_max * 60,
config.site_peer_check_interval_min * 60,
handler=self.periodicMaintenanceHandler_peer_check,
scaler=self.getAnnounceRating),
ScaledTimeoutHandler(
config.site_update_check_interval_max * 60,
config.site_update_check_interval_min * 60,
handler=self.periodicMaintenanceHandler_general,
scaler=self.getActivityRating)
]
self.background_publishers = {}
self.background_publishers_lock = gevent.lock.RLock()
# FZS = forced zero "since"
self.fzs_range = 20
self.fzs_interval = 30 * 60
self.fzs_count = random.randint(0, self.fzs_range / 4)
self.fzs_timestamp = 0
##############################################
self.connection_server = None
if "main" in sys.modules: # import main has side-effects, breaks tests
import main
if "file_server" in dir(main): # Use global file server by default if possible
self.connection_server = main.file_server
else:
main.file_server = FileServer()
self.connection_server = main.file_server
else:
self.connection_server = FileServer()
##############################################
self.content = None # Load content.json
self.peers = {} # Key: ip:port, Value: Peer.Peer
self.peers_recent = collections.deque(maxlen=150)
self.peer_blacklist = SiteManager.peer_blacklist # Ignore this peers (eg. myself)
self.greenlet_manager = GreenletManager.GreenletManager(self.connection_server.site_pool) # Running greenlets
self.worker_manager = WorkerManager(self) # Handle site download from other peers
self.bad_files = {} # SHA check failed files, need to redownload {"inner.content": 1} (key: file, value: failed accept)
self.content_updated = None # Content.js update time
self.last_online_update = 0
self.startup_announce_done = 0
self.notifications = [] # Pending notifications displayed once on page load [error|ok|info, message, timeout]
self.page_requested = False # Page viewed in browser
self.websockets = [] # Active site websocket connections
self.loadSettings(settings) # Load settings from sites.json
self.storage = SiteStorage(self, allow_create=allow_create) # Save and load site files
self.content_manager = ContentManager(self)
self.content_manager.loadContents() # Load content.json files
self.announcer = SiteAnnouncer(self) # Announce and get peer list from other nodes
self.peer_connector = SiteHelpers.PeerConnector(self) # Connect more peers in background by request
self.persistent_peer_req = None # The persistent peer requirement, managed by maintenance handler
if not self.settings.get("wrapper_key"): # To auth websocket permissions
self.settings["wrapper_key"] = CryptHash.random()
self.log.debug("New wrapper key: %s" % self.settings["wrapper_key"])
if not self.settings.get("ajax_key"): # To auth websocket permissions
self.settings["ajax_key"] = CryptHash.random()
self.log.debug("New ajax key: %s" % self.settings["ajax_key"])
def __str__(self):
return "Site %s" % self.address_short
def __repr__(self):
return "<%s>" % self.__str__()
# Load site settings from data/sites.json
def loadSettings(self, settings=None):
if not settings:
settings = json.load(open("%s/sites.json" % config.data_dir)).get(self.address)
if settings:
self.settings = settings
if "check_files_timestamp" not in settings:
settings["check_files_timestamp"] = 0
if "verify_files_timestamp" not in settings:
settings["verify_files_timestamp"] = 0
if "cache" not in settings:
settings["cache"] = {}
if "size_files_optional" not in settings:
settings["size_optional"] = 0
if "optional_downloaded" not in settings:
settings["optional_downloaded"] = 0
if "downloaded" not in settings:
settings["downloaded"] = settings.get("added")
self.bad_files = settings["cache"].get("bad_files", {})
settings["cache"]["bad_files"] = {}
# Give it minimum 10 tries after restart
for inner_path in self.bad_files:
self.bad_files[inner_path] = min(self.bad_files[inner_path], 20)
else:
self.settings = {
"check_files_timestamp": 0,
"verify_files_timestamp": 0,
"own": False,
"serving": True,
"permissions": [],
"cache": {"bad_files": {}},
"size_files_optional": 0,
"added": int(time.time()),
"downloaded": None,
"optional_downloaded": 0,
"size_optional": 0
} # Default
if config.download_optional == "auto":
self.settings["autodownloadoptional"] = True
# Add admin permissions to homepage
if self.address in (config.homepage, config.updatesite) and "ADMIN" not in self.settings["permissions"]:
self.settings["permissions"].append("ADMIN")
return
# Save site settings to data/sites.json
def saveSettings(self):
if not SiteManager.site_manager.sites:
SiteManager.site_manager.sites = {}
if not SiteManager.site_manager.sites.get(self.address):
SiteManager.site_manager.sites[self.address] = self
SiteManager.site_manager.load(False)
SiteManager.site_manager.saveDelayed()
# Returns True if any site-related activity should be interrupted
# due to connection server being stopped or site being deleted
def isStopping(self):
return self.connection_server.stopping or self.settings.get("deleting", False)
# Returns False if any network activity for the site should not happen
def isServing(self):
if self.connection_server.isOfflineMode():
return False
elif self.isStopping():
return False
else:
return self.settings["serving"]
# Spawns a thread that will be waited for on server being stopped (and killed after a timeout).
# Short cut to self.greenlet_manager.spawn()
def spawn(self, *args, **kwargs):
thread = self.greenlet_manager.spawn(*args, **kwargs)
return thread
# Spawns a thread that will be waited for on server being stopped (and killed after a timeout).
# Short cut to self.greenlet_manager.spawnLater()
def spawnLater(self, *args, **kwargs):
thread = self.greenlet_manager.spawnLater(*args, **kwargs)
return thread
def checkSendBackLRU(self, peer, inner_path, remote_modified):
return SiteManager.site_manager.checkSendBackLRU(self, peer, inner_path, remote_modified)
def addToSendBackLRU(self, peer, inner_path, modified):
return SiteManager.site_manager.addToSendBackLRU(self, peer, inner_path, modified)
def getSettingsCache(self):
back = {}
back["bad_files"] = self.bad_files
back["hashfield"] = base64.b64encode(self.content_manager.hashfield.tobytes()).decode("ascii")
return back
# Max site size in MB
def getSizeLimit(self):
return self.settings.get("size_limit", int(config.size_limit))
def isFileVerificationExpired(self, check_files_interval, verify_files_interval):
now = time.time()
check_files_timestamp = self.settings.get("check_files_timestamp", 0)
verify_files_timestamp = self.settings.get("verify_files_timestamp", 0)
if check_files_interval is None:
check_files_expiration = now + 1
else:
check_files_expiration = check_files_timestamp + check_files_interval
if verify_files_interval is None:
verify_files_expiration = now + 1
else:
verify_files_expiration = verify_files_timestamp + verify_files_interval
if verify_files_expiration < now:
return "verify"
if check_files_expiration < now:
return "check"
return False
# Next size limit based on current size
def getNextSizeLimit(self):
size_limits = [10, 20, 50, 100, 200, 500, 1000, 2000, 5000, 10000, 20000, 50000, 100000]
size = self.settings.get("size", 0)
for size_limit in size_limits:
if size * 1.2 < size_limit * 1024 * 1024:
return size_limit
return 999999
def isAddedRecently(self):
return time.time() - self.settings.get("added", 0) < 60 * 60 * 24
# Download all files from content.json
def downloadContent(self, inner_path, download_files=True, peer=None, check_modifications=False, diffs={}):
s = time.time()
if config.verbose:
self.log.debug(
"DownloadContent %s: Started. (download_files: %s, check_modifications: %s, diffs: %s)..." %
(inner_path, download_files, check_modifications, diffs.keys())
)
if not inner_path.endswith("content.json"):
return False
found = self.needFile(inner_path, update=self.bad_files.get(inner_path))
content_inner_dir = helper.getDirname(inner_path)
if not found:
self.log.debug("DownloadContent %s: Download failed, check_modifications: %s" % (inner_path, check_modifications))
if check_modifications: # Download failed, but check modifications if its succed later
self.onFileDone.once(lambda file_name: self.checkModifications(0), "check_modifications")
return False # Could not download content.json
if config.verbose:
self.log.debug("DownloadContent got %s" % inner_path)
sub_s = time.time()
changed, deleted = self.content_manager.loadContent(inner_path, load_includes=False)
if config.verbose:
self.log.debug("DownloadContent %s: loadContent done in %.3fs" % (inner_path, time.time() - sub_s))
if inner_path == "content.json":
self.saveSettings()
if peer: # Update last received update from peer to prevent re-sending the same update to it
peer.last_content_json_update = self.content_manager.contents[inner_path]["modified"]
# Verify size limit
if inner_path == "content.json":
site_size_limit = self.getSizeLimit() * 1024 * 1024
content_size = len(json.dumps(self.content_manager.contents[inner_path], indent=1)) + sum([file["size"] for file in list(self.content_manager.contents[inner_path].get("files", {}).values()) if file["size"] >= 0]) # Size of new content
if site_size_limit < content_size:
# Not enought don't download anything
self.log.debug("DownloadContent Size limit reached (site too big please increase limit): %.2f MB > %.2f MB" % (content_size / 1024 / 1024, site_size_limit / 1024 / 1024))
return False
# Start download files
file_threads = []
if download_files:
for file_relative_path in list(self.content_manager.contents[inner_path].get("files", {}).keys()):
file_inner_path = content_inner_dir + file_relative_path
# Try to diff first
diff_success = False
diff_actions = diffs.get(file_relative_path)
if diff_actions and self.bad_files.get(file_inner_path):
try:
s = time.time()
new_file = Diff.patch(self.storage.open(file_inner_path, "rb"), diff_actions)
new_file.seek(0)
time_diff = time.time() - s
s = time.time()
diff_success = self.content_manager.verifyFile(file_inner_path, new_file)
time_verify = time.time() - s
if diff_success:
s = time.time()
new_file.seek(0)
self.storage.write(file_inner_path, new_file)
time_write = time.time() - s
s = time.time()
self.onFileDone(file_inner_path)
time_on_done = time.time() - s
self.log.debug(
"DownloadContent Patched successfully: %s (diff: %.3fs, verify: %.3fs, write: %.3fs, on_done: %.3fs)" %
(file_inner_path, time_diff, time_verify, time_write, time_on_done)
)
except Exception as err:
self.log.debug("DownloadContent Failed to patch %s: %s" % (file_inner_path, err))
diff_success = False
if not diff_success:
# Start download and dont wait for finish, return the event
res = self.needFile(file_inner_path, blocking=False, update=self.bad_files.get(file_inner_path), peer=peer)
if res is not True and res is not False: # Need downloading and file is allowed
file_threads.append(res) # Append evt
# Optionals files
if inner_path == "content.json":
self.spawn(self.updateHashfield)
for file_relative_path in list(self.content_manager.contents[inner_path].get("files_optional", {}).keys()):
file_inner_path = content_inner_dir + file_relative_path
if file_inner_path not in changed and not self.bad_files.get(file_inner_path):
continue
if not self.isDownloadable(file_inner_path):
continue
# Start download and dont wait for finish, return the event
res = self.pooledNeedFile(
file_inner_path, blocking=False, update=self.bad_files.get(file_inner_path), peer=peer
)
if res is not True and res is not False: # Need downloading and file is allowed
file_threads.append(res) # Append evt
# Wait for includes download
include_threads = []
for file_relative_path in list(self.content_manager.contents[inner_path].get("includes", {}).keys()):
file_inner_path = content_inner_dir + file_relative_path
include_thread = self.spawn(self.downloadContent, file_inner_path, download_files=download_files, peer=peer)
include_threads.append(include_thread)
if config.verbose:
self.log.debug("DownloadContent %s: Downloading %s includes..." % (inner_path, len(include_threads)))
gevent.joinall(include_threads)
if config.verbose:
self.log.debug("DownloadContent %s: Includes download ended" % inner_path)
if check_modifications: # Check if every file is up-to-date
self.checkModifications(0)
if config.verbose:
self.log.debug("DownloadContent %s: Downloading %s files, changed: %s..." % (inner_path, len(file_threads), len(changed)))
gevent.joinall(file_threads)
if config.verbose:
self.log.debug("DownloadContent %s: ended in %.3fs (tasks left: %s)" % (
inner_path, time.time() - s, len(self.worker_manager.tasks)
))
# If no file tasks have been started, worker_manager.checkComplete()
# never called. So call it explicitly.
self.greenlet_manager.spawn(self.worker_manager.checkComplete)
return True
# Return bad files with less than 3 retry
def getReachableBadFiles(self):
if not self.bad_files:
return False
return [bad_file for bad_file, retry in self.bad_files.items() if retry < 3]
# Retry download bad files
def retryBadFiles(self, force=False):
self.checkBadFiles()
self.log.debug("Retry %s bad files" % len(self.bad_files))
content_inner_paths = []
file_inner_paths = []
for bad_file, tries in list(self.bad_files.items()):
if force or random.randint(0, min(40, tries)) < 4: # Larger number tries = less likely to check every 15min
if bad_file.endswith("content.json"):
content_inner_paths.append(bad_file)
else:
file_inner_paths.append(bad_file)
if content_inner_paths:
self.pooledDownloadContent(content_inner_paths, only_if_bad=True)
if file_inner_paths:
self.pooledDownloadFile(file_inner_paths, only_if_bad=True)
def checkBadFiles(self):
for bad_file in list(self.bad_files.keys()):
file_info = self.content_manager.getFileInfo(bad_file)
if bad_file.endswith("content.json"):
if file_info is False and bad_file != "content.json":
del self.bad_files[bad_file]
self.log.debug("No info for file: %s, removing from bad_files" % bad_file)
else:
if file_info is False or not file_info.get("size"):
del self.bad_files[bad_file]
self.log.debug("No info or size for file: %s, removing from bad_files" % bad_file)
# Download all files of the site
@util.Noparallel(blocking=False)
def download(self, check_size=False, blind_includes=False, retry_bad_files=True):
if not self.connection_server:
self.log.debug("No connection server found, skipping download")
return False
s = time.time()
self.log.debug(
"Start downloading, bad_files: %s, check_size: %s, blind_includes: %s, isAddedRecently: %s" %
(self.bad_files, check_size, blind_includes, self.isAddedRecently())
)
if self.isAddedRecently():
self.spawn(self.announce, mode="start", force=True)
else:
self.spawn(self.announce, mode="update")
if check_size: # Check the size first
valid = self.downloadContent("content.json", download_files=False) # Just download content.json files
if not valid:
return False # Cant download content.jsons or size is not fits
# Download everything
valid = self.downloadContent("content.json", check_modifications=blind_includes)
if retry_bad_files:
self.onComplete.once(lambda: self.retryBadFiles(force=True))
self.log.debug("Download done in %.3fs" % (time.time() - s))
return valid
def pooledDownloadContent(self, inner_paths, pool_size=100, only_if_bad=False):
self.log.debug("New downloadContent pool: len: %s, only if bad: %s" % (len(inner_paths), only_if_bad))
self.worker_manager.started_task_num += len(inner_paths)
pool = gevent.pool.Pool(pool_size)
num_skipped = 0
site_size_limit = self.getSizeLimit() * 1024 * 1024
for inner_path in inner_paths:
if not only_if_bad or inner_path in self.bad_files:
pool.spawn(self.downloadContent, inner_path)
else:
num_skipped += 1
self.worker_manager.started_task_num -= 1
if self.settings["size"] > site_size_limit * 0.95:
self.log.warning("Site size limit almost reached, aborting downloadContent pool")
for aborted_inner_path in inner_paths:
if aborted_inner_path in self.bad_files:
del self.bad_files[aborted_inner_path]
self.worker_manager.removeSolvedFileTasks(mark_as_good=False)
break
pool.join()
self.log.debug("Ended downloadContent pool len: %s, skipped: %s" % (len(inner_paths), num_skipped))
def pooledDownloadFile(self, inner_paths, pool_size=100, only_if_bad=False):
self.log.debug("New downloadFile pool: len: %s, only if bad: %s" % (len(inner_paths), only_if_bad))
self.worker_manager.started_task_num += len(inner_paths)
pool = gevent.pool.Pool(pool_size)
num_skipped = 0
for inner_path in inner_paths:
if not only_if_bad or inner_path in self.bad_files:
pool.spawn(self.needFile, inner_path, update=True)
else:
num_skipped += 1
self.worker_manager.started_task_num -= 1
self.log.debug("Ended downloadFile pool len: %s, skipped: %s" % (len(inner_paths), num_skipped))
# Update worker, try to find client that supports listModifications command
def updater(self, peers_try, queried, need_queries, since):
threads = []
while 1:
if not peers_try or len(queried) >= need_queries: # Stop after 3 successful query
break
peer = peers_try.pop(0)
if peer in queried:
continue
if config.verbose:
self.log.debug("CheckModifications: Trying to get updates from: %s Left: %s" % (peer, peers_try))
res = None
with gevent.Timeout(20, exception=False):
res = peer.listModified(since)
if not res or "modified_files" not in res:
continue # Failed query
queried.append(peer)
modified_contents = []
send_back = []
send_back_limit = config.send_back_limit
send_back_skipped = 0
peer_modified_files = res["modified_files"]
my_modified_files = self.content_manager.listModified(since)
inner_paths = itertools.chain(peer_modified_files.keys(), my_modified_files.keys())
seen_inner_paths = {}
for inner_path in inner_paths: # Check if the peer has newer files than we have
if seen_inner_paths.get(inner_path, False):
continue
seen_inner_paths[inner_path] = True
peer_modified = int(peer_modified_files.get(inner_path, 0))
my_modified = int(my_modified_files.get(inner_path, 0))
diff = peer_modified - my_modified
if diff == 0:
continue
has_newer = diff > 0
has_older = diff < 0
if inner_path not in self.bad_files and not self.content_manager.isArchived(inner_path, peer_modified):
if has_newer: # We don't have this file or we have older version
modified_contents.append(inner_path)
self.bad_files[inner_path] = self.bad_files.get(inner_path, 1)
if has_older: # The remote peer doesn't have this file or it has older version
if self.checkSendBackLRU(peer, inner_path, peer_modified):
send_back_skipped += 1
else:
send_back.append(inner_path)
inner_paths = None
seen_inner_paths = None
if modified_contents:
self.log.info("CheckModifications: %s new modified files from %s" % (len(modified_contents), peer))
modified_contents.sort(key=lambda inner_path: 0 - peer_modified_files[inner_path]) # Download newest first
for inner_path in modified_contents:
self.log.info("CheckModifications: %s: %s > %s" % (
inner_path, peer_modified_files.get(inner_path, 0), my_modified_files.get(inner_path, 0)
))
t = self.spawn(self.pooledDownloadContent, modified_contents, only_if_bad=True)
threads.append(t)
if send_back:
self.log.info("CheckModifications: %s has older versions of %s files" % (peer, len(send_back)))
if len(send_back) > send_back_limit:
self.log.info("CheckModifications: choosing %s random files to publish back" % (send_back_limit))
random.shuffle(send_back)
send_back = send_back[0:send_back_limit]
for inner_path in send_back:
self.log.info("CheckModifications: %s: %s < %s" % (
inner_path, peer_modified_files.get(inner_path, 0), my_modified_files.get(inner_path, 0)
))
self.spawn(self.publisher, inner_path, [peer], [], 1, save_to_send_back_lru=True)
if send_back_skipped:
self.log.info("CheckModifications: %s has older versions of %s files, skipped according to send back LRU" % (peer, send_back_skipped))
self.log.debug("CheckModifications: Waiting for %s pooledDownloadContent" % len(threads))
gevent.joinall(threads)
# We need, with some rate, to perform the full check of modifications,
# "since the beginning of time", instead of the partial one.
def getForcedZeroSince(self):
now = time.time()
if self.fzs_timestamp + self.fzs_interval > now:
return False
self.fzs_count -= 1
if self.fzs_count < 1:
self.fzs_count = random.randint(0, self.fzs_range)
self.fzs_timestamp = now
return True
return False
# Check modified content.json files from peers and add modified files to bad_files
# Return: Successfully queried peers [Peer, Peer...]
def checkModifications(self, since=None):
s = time.time()
peers_try = [] # Try these peers
queried = [] # Successfully queried from these peers
peer_limit = 10
updater_limit = 3
need_queries = 3
# Wait for peers
if not self.peers:
self.announce(mode="update")
for wait in range(10):
time.sleep(5 + wait)
self.log.debug("CheckModifications: Waiting for peers...")
if self.peers:
break
if since is None:
if self.getForcedZeroSince():
since = 0
else:
margin = 60 * 60 * 24
since = self.settings.get("modified", margin) - margin
if since == 0:
peer_limit *= 4
need_queries *= 4
peers_try = self.getConnectedPeers() + self.getConnectablePeers(peer_limit)
self.log.debug(
"CheckModifications: Trying to get listModifications from %s peers, %s connected, since: %s" %
(len(peers_try), len(self.getConnectedPeers()), since)
)
updaters = []
for i in range(updater_limit):
updaters.append(self.spawn(self.updater, peers_try, queried, need_queries, since))
for r in range(10):
gevent.joinall(updaters, timeout=5+r)
if len(queried) >= need_queries or len(peers_try) == 0:
break
self.log.debug("CheckModifications: Waiting... (%s) succesfully queried: %s, left: %s" %
(r + 1, len(queried), len(peers_try)))
self.log.debug("CheckModifications: Queried listModifications from %s peers in %.3fs since %s" % (
len(queried), time.time() - s, since))
time.sleep(0.1)
return queried
def invalidateUpdateTime(self, invalid_interval):
a, b = invalid_interval
if b is None:
b = time.time()
if a is None:
a = b
if a <= self.last_online_update and self.last_online_update <= b:
self.last_online_update = 0
self.log.debug("Update time invalidated")
def isUpdateTimeValid(self):
if not self.last_online_update:
return False
expirationThreshold = 60 * 60 * 6
return self.last_online_update > time.time() - expirationThreshold
def refreshUpdateTime(self, valid=True):
if valid:
self.last_online_update = time.time()
else:
self.last_online_update = 0
# Update content.json from peers and download changed files
# Return: None
@util.Noparallel()
def update(self, announce=False, check_files=False, verify_files=False, since=None):
online = self.connection_server.isInternetOnline()
self.content_manager.loadContent("content.json", load_includes=False) # Reload content.json
self.content_updated = None # Reset content updated time
if verify_files:
check_files = True
if verify_files:
self.updateWebsocket(verifying=True)
elif check_files:
self.updateWebsocket(checking=True)
if check_files:
if verify_files:
self.storage.updateBadFiles(quick_check=False) # Full-featured checksum verification
else:
self.storage.updateBadFiles(quick_check=True) # Quick check and mark bad files based on file size
# Don't update the timestamps in case of the application being shut down,
# so we can make another try next time.
if not self.isStopping():
self.settings["check_files_timestamp"] = time.time()
if verify_files:
self.settings["verify_files_timestamp"] = time.time()
if verify_files:
self.updateWebsocket(verified=True)
elif check_files:
self.updateWebsocket(checked=True)
if not self.isServing():
return False
if announce:
self.updateWebsocket(updating=True)
self.announce(mode="update", force=True)
reqs = [
self.peer_connector.newReq(4, 4, 30),
self.peer_connector.newReq(2, 2, 60),
self.peer_connector.newReq(1, 1, 120)
]
nr_connected_peers = self.waitForPeers(reqs);
if nr_connected_peers < 1:
return
self.updateWebsocket(updating=True)
# Remove files that no longer in content.json
self.checkBadFiles()
# Full update, we can reset bad files
if check_files and since == 0:
self.bad_files = {}
queried = self.checkModifications(since)
changed, deleted = self.content_manager.loadContent("content.json", load_includes=False)
if self.bad_files:
self.log.debug("Bad files: %s" % self.bad_files)
self.spawn(self.retryBadFiles, force=True)
if len(queried) == 0:
# Failed to query modifications
self.content_updated = False
else:
self.content_updated = time.time()
self.sendMyHashfield()
self.updateHashfield()
online = online and self.connection_server.isInternetOnline()
self.refreshUpdateTime(valid=online)
self.updateWebsocket(updated=True)
# To be called from FileServer
@util.Noparallel(queue=True, ignore_args=True)
def update2(self, check_files=False, verify_files=False):
self.update(check_files=check_files, verify_files=verify_files)
# Update site by redownload all content.json
def redownloadContents(self):
# Download all content.json again
content_threads = []
for inner_path in list(self.content_manager.contents.keys()):
content_threads.append(self.needFile(inner_path, update=True, blocking=False))
self.log.debug("Waiting %s content.json to finish..." % len(content_threads))
gevent.joinall(content_threads)
# Publish worker
def publisher(self, inner_path, peers, published, limit, diffs={}, event_done=None, cb_progress=None, max_retries=2, save_to_send_back_lru=False):
file_size = self.storage.getSize(inner_path)
content_json_modified = self.content_manager.contents[inner_path]["modified"]
body = self.storage.read(inner_path)
while 1:
if not peers or len(published) >= limit:
if event_done:
event_done.set(True)
break # All peers done, or published engouht
peer = peers.pop()
if peer in published:
continue
if peer.last_content_json_update == content_json_modified:
self.log.debug("%s already received this update for %s, skipping" % (peer, inner_path))
continue
if peer.connection and peer.connection.last_ping_delay: # Peer connected
# Timeout: 5sec + size in kb + last_ping
timeout = 5 + int(file_size / 1024) + peer.connection.last_ping_delay
else: # Peer not connected
# Timeout: 10sec + size in kb
timeout = 10 + int(file_size / 1024)
result = {"exception": "Timeout"}
for retry in range(max_retries):
try:
with gevent.Timeout(timeout, False):
result = peer.publish(self.address, inner_path, body, content_json_modified, diffs)
if result:
break
except Exception as err:
self.log.error("Publish error: %s" % Debug.formatException(err))
result = {"exception": Debug.formatException(err)}
# We add to the send back lru not only on success, but also on errors.
# Some peers returns None. (Why?)
# Anyway, we tried our best in delivering possibly lost updates.
if save_to_send_back_lru:
self.addToSendBackLRU(peer, inner_path, content_json_modified)
if result and "ok" in result:
published.append(peer)
if cb_progress and len(published) <= limit:
cb_progress(len(published), limit)
self.log.info("[OK] %s: %s %s/%s" % (peer.key, result["ok"], len(published), limit))
else:
if result == {"exception": "Timeout"}:
peer.onConnectionError("Publish timeout")
self.log.info("[FAILED] %s: %s" % (peer.key, result))
time.sleep(0.01)
def addBackgroundPublisher(self, published=[], limit=5, inner_path="content.json", diffs={}):
with self.background_publishers_lock:
if self.background_publishers.get(inner_path, None):
background_publisher = self.background_publishers[inner_path]
background_publisher.reinit(published=published, limit=limit, diffs=diffs)
else:
background_publisher = BackgroundPublisher(self, published=published, limit=limit, inner_path=inner_path, diffs=diffs)
self.background_publishers[inner_path] = background_publisher
self.spawn(background_publisher.process)
def processBackgroundPublishers(self):
with self.background_publishers_lock:
for inner_path, background_publisher in list(self.background_publishers.items()):
background_publisher.process()
if background_publisher.isComplete():
background_publisher.finalize()
del self.background_publishers[inner_path]
def getPeersForForegroundPublishing(self, limit):
# Wait for some peers to appear
reqs = [
self.peer_connector.newReq(limit, limit / 2, 10), # some of them...
self.peer_connector.newReq(1, 1, 60) # or at least one...
]
self.waitForPeers(reqs, update_websocket=False)
peers = self.getConnectedPeers()
random.shuffle(peers)
# Prefer newer clients.
# Trying to deliver foreground updates to the latest version clients,
# expecting that they have better networking facilities.
# Note: background updates SHOULD NOT discriminate peers by their rev number,
# otherwise it can cause troubles in delivering updates to older versions.
peers = sorted(peers, key=lambda peer: peer.connection.handshake.get("rev", 0) < config.rev - 100)
# Add more, non-connected peers if necessary
if len(peers) < limit * 2 and len(self.peers) > len(peers):
peers += self.getRecentPeers(limit * 2)
peers = set(peers)
return peers
# Update content.json on peers
@util.Noparallel()
def publish(self, limit="default", inner_path="content.json", diffs={}, cb_progress=None):
published = [] # Successfully published (Peer)
publishers = [] # Publisher threads
if limit == "default":
limit = 5
threads = limit
peers = self.getPeersForForegroundPublishing(limit)
self.log.info("Publishing %s to %s/%s peers (connected: %s) diffs: %s (%.2fk)..." % (
inner_path,
limit, len(self.peers), len(self.getConnectedPeers()),
list(diffs.keys()), float(len(str(diffs))) / 1024
))
if not peers:
self.addBackgroundPublisher(published=published, limit=limit, inner_path=inner_path, diffs=diffs)
return 0 # No peers found
event_done = gevent.event.AsyncResult()
for i in range(min(len(peers), limit, threads)):
publisher = self.spawn(self.publisher, inner_path, peers, published, limit, diffs, event_done, cb_progress)
publishers.append(publisher)
event_done.get() # Wait for done
if len(published) < min(len(self.peers), limit):
time.sleep(0.2) # If less than we need sleep a bit
if len(published) == 0:
gevent.joinall(publishers) # No successful publish, wait for all publisher
# Publish to more peers in the background
self.log.info(
"Published %s to %s peers, publishing to more peers in the background" %
(inner_path, len(published))
)
self.addBackgroundPublisher(published=published, limit=limit, inner_path=inner_path, diffs=diffs)
# Send my hashfield to every connected peer if changed
self.spawn(self.sendMyHashfield, 100)
return len(published)
# Copy this site
@util.Noparallel()
def clone(self, address, privatekey=None, address_index=None, root_inner_path="", overwrite=False):
import shutil
new_site = SiteManager.site_manager.need(address, all_file=False)
default_dirs = [] # Dont copy these directories (has -default version)
for dir_name in os.listdir(self.storage.directory):
if "-default" in dir_name:
default_dirs.append(dir_name.replace("-default", ""))
self.log.debug("Cloning to %s, ignore dirs: %s, root: %s" % (address, default_dirs, root_inner_path))
# Copy root content.json
if not new_site.storage.isFile("content.json") and not overwrite:
# New site: Content.json not exist yet, create a new one from source site
if "size_limit" in self.settings:
new_site.settings["size_limit"] = self.settings["size_limit"]
# Use content.json-default is specified
if self.storage.isFile(root_inner_path + "/content.json-default"):
content_json = self.storage.loadJson(root_inner_path + "/content.json-default")
else:
content_json = self.storage.loadJson("content.json")
if "domain" in content_json:
del content_json["domain"]
content_json["title"] = "my" + content_json["title"]
content_json["cloned_from"] = self.address
content_json["clone_root"] = root_inner_path
content_json["files"] = {}
if address_index:
content_json["address_index"] = address_index # Site owner's BIP32 index
new_site.storage.writeJson("content.json", content_json)
new_site.content_manager.loadContent(
"content.json", add_bad_files=False, delete_removed_files=False, load_includes=False
)
# Copy files
for content_inner_path, content in list(self.content_manager.contents.items()):
file_relative_paths = list(content.get("files", {}).keys())
# Sign content.json at the end to make sure every file is included
file_relative_paths.sort()
file_relative_paths.sort(key=lambda key: key.replace("-default", "").endswith("content.json"))
for file_relative_path in file_relative_paths:
file_inner_path = helper.getDirname(content_inner_path) + file_relative_path # Relative to content.json
file_inner_path = file_inner_path.strip("/") # Strip leading /
if not file_inner_path.startswith(root_inner_path):
self.log.debug("[SKIP] %s (not in clone root)" % file_inner_path)
continue
if file_inner_path.split("/")[0] in default_dirs: # Dont copy directories that has -default postfixed alternative
self.log.debug("[SKIP] %s (has default alternative)" % file_inner_path)
continue
file_path = self.storage.getPath(file_inner_path)
# Copy the file normally to keep the -default postfixed dir and file to allow cloning later
if root_inner_path:
file_inner_path_dest = re.sub("^%s/" % re.escape(root_inner_path), "", file_inner_path)
file_path_dest = new_site.storage.getPath(file_inner_path_dest)
else:
file_inner_path_dest = file_inner_path
file_path_dest = new_site.storage.getPath(file_inner_path)
self.log.debug("[COPY] %s to %s..." % (file_inner_path, file_path_dest))
dest_dir = os.path.dirname(file_path_dest)
if not os.path.isdir(dest_dir):
os.makedirs(dest_dir)
if file_inner_path_dest.replace("-default", "") == "content.json": # Don't copy root content.json-default
continue
shutil.copy(file_path, file_path_dest)
# If -default in path, create a -default less copy of the file
if "-default" in file_inner_path_dest:
file_path_dest = new_site.storage.getPath(file_inner_path_dest.replace("-default", ""))
if new_site.storage.isFile(file_inner_path_dest.replace("-default", "")) and not overwrite:
# Don't overwrite site files with default ones
self.log.debug("[SKIP] Default file: %s (already exist)" % file_inner_path)
continue
self.log.debug("[COPY] Default file: %s to %s..." % (file_inner_path, file_path_dest))
dest_dir = os.path.dirname(file_path_dest)
if not os.path.isdir(dest_dir):
os.makedirs(dest_dir)
shutil.copy(file_path, file_path_dest)
# Sign if content json
if file_path_dest.endswith("/content.json"):
new_site.storage.onUpdated(file_inner_path_dest.replace("-default", ""))
new_site.content_manager.loadContent(
file_inner_path_dest.replace("-default", ""), add_bad_files=False,
delete_removed_files=False, load_includes=False
)
if privatekey:
new_site.content_manager.sign(file_inner_path_dest.replace("-default", ""), privatekey, remove_missing_optional=True)
new_site.content_manager.loadContent(
file_inner_path_dest, add_bad_files=False, delete_removed_files=False, load_includes=False
)
if privatekey:
new_site.content_manager.sign("content.json", privatekey, remove_missing_optional=True)
new_site.content_manager.loadContent(
"content.json", add_bad_files=False, delete_removed_files=False, load_includes=False
)
# Rebuild DB
if new_site.storage.isFile("dbschema.json"):
new_site.storage.closeDb()
try:
new_site.storage.rebuildDb()
except Exception as err:
self.log.error(err)
return new_site
@util.Pooled(100)
def pooledNeedFile(self, *args, **kwargs):
return self.needFile(*args, **kwargs)
def isFileDownloadAllowed(self, inner_path, file_info):
# Verify space for all site
if self.settings["size"] > self.getSizeLimit() * 1024 * 1024:
return False
# Verify space for file
if file_info.get("size", 0) > config.file_size_limit * 1024 * 1024:
self.log.debug(
"File size %s too large: %sMB > %sMB, skipping..." %
(inner_path, file_info.get("size", 0) / 1024 / 1024, config.file_size_limit)
)
return False
else:
return True
def needFileInfo(self, inner_path):
file_info = self.content_manager.getFileInfo(inner_path)
if not file_info:
# No info for file, download all content.json first
self.log.debug("No info for %s, waiting for all content.json" % inner_path)
success = self.downloadContent("content.json", download_files=False)
if not success:
return False
file_info = self.content_manager.getFileInfo(inner_path)
return file_info
# Check and download if file not exist
def needFile(self, inner_path, update=False, blocking=True, peer=None, priority=0):
if self.worker_manager.tasks.findTask(inner_path):
task = self.worker_manager.addTask(inner_path, peer, priority=priority)
if blocking:
return task["evt"].get()
else:
return task["evt"]
elif self.storage.isFile(inner_path) and not update: # File exist, no need to do anything
return True
elif not self.isServing(): # Site not serving
return False
else: # Wait until file downloaded
if not self.content_manager.contents.get("content.json"): # No content.json, download it first!
self.log.debug("Need content.json first (inner_path: %s, priority: %s)" % (inner_path, priority))
if priority > 0:
self.spawn(self.announce)
if inner_path != "content.json": # Prevent double download
task = self.worker_manager.addTask("content.json", peer)
task["evt"].get()
self.content_manager.loadContent()
if not self.content_manager.contents.get("content.json"):
return False # Content.json download failed
file_info = None
if not inner_path.endswith("content.json"):
file_info = self.needFileInfo(inner_path)
if not file_info:
return False
if "cert_signers" in file_info and not file_info["content_inner_path"] in self.content_manager.contents:
self.log.debug("Missing content.json for requested user file: %s" % inner_path)
if self.bad_files.get(file_info["content_inner_path"], 0) > 5:
self.log.debug("File %s not reachable: retry %s" % (
inner_path, self.bad_files.get(file_info["content_inner_path"], 0)
))
return False
self.downloadContent(file_info["content_inner_path"])
if not self.isFileDownloadAllowed(inner_path, file_info):
self.log.debug("%s: Download not allowed" % inner_path)
return False
self.bad_files[inner_path] = self.bad_files.get(inner_path, 1) # Mark as bad file
task = self.worker_manager.addTask(inner_path, peer, priority=priority, file_info=file_info)
if blocking:
return task["evt"].get()
else:
return task["evt"]
# Add or update a peer to site
# return_peer: Always return the peer even if it was already present
def addPeer(self, ip, port, return_peer=False, connection=None, source="other"):
if not ip or ip == "0.0.0.0":
return False
key = "%s:%s" % (ip, port)
peer = self.peers.get(key)
if peer: # Already has this ip
peer.found(source)
if return_peer: # Always return peer
return peer
else:
return False
else: # New peer
if (ip, port) in self.peer_blacklist:
return False # Ignore blacklist (eg. myself)
peer = Peer(ip, port, self)
self.peers[key] = peer
peer.found(source)
self.peer_connector.processReqs()
self.peer_connector.addPeer(peer)
return peer
# Called from peer.remove to erase links to peer
def deregisterPeer(self, peer):
self.peers.pop(peer.key, None)
try:
self.peers_recent.remove(peer)
except:
pass
self.peer_connector.deregisterPeer(peer)
def announce(self, *args, **kwargs):
if self.isServing():
self.announcer.announce(*args, **kwargs)
def getActivityRating(self, force_safe=False):
age = time.time() - self.settings.get("modified", 0)
if age < 60 * 60:
rating = 1.0
elif age < 60 * 60 * 5:
rating = 0.8
elif age < 60 * 60 * 24:
rating = 0.6
elif age < 60 * 60 * 24 * 3:
rating = 0.4
elif age < 60 * 60 * 24 * 7:
rating = 0.2
else:
rating = 0.0
force_safe = force_safe or config.expose_no_ownership
if (not force_safe) and self.settings["own"]:
rating = min(rating, 0.6)
if self.content_updated is False: # Last check modification failed
rating += 0.1
elif self.bad_files:
rating += 0.1
if rating > 1.0:
rating = 1.0
return rating
def getAnnounceRating(self):
# rare frequent
# announces announces
# 0 ------------------- 1
# activity -------------> -- active site ==> frequent announces
# <---------------- peers -- many peers ==> rare announces
# trackers -------------> -- many trackers ==> frequent announces to iterate over more trackers
activity_rating = self.getActivityRating(force_safe=True)
peer_count = len(self.peers)
peer_rating = 1.0 - min(peer_count, 50) / 50.0
tracker_count = self.announcer.getSupportedTrackerCount()
tracker_count = max(tracker_count, 1)
tracker_rating = 1.0 - (1.0 / tracker_count)
v = [activity_rating, peer_rating, tracker_rating]
return sum(v) / float(len(v))
def getPreferableConnectablePeerCount(self):
if not self.isServing():
return 0
count = lerp(
config.site_connectable_peer_count_min,
config.site_connectable_peer_count_max,
self.getActivityRating(force_safe=True))
return count
# The engine tries to maintain the number of active connections:
# >= getPreferableActiveConnectionCount()
# and
# <= getActiveConnectionCountLimit()
def getPreferableActiveConnectionCount(self):
if not self.isServing():
return 0
age = time.time() - self.settings.get("modified", 0)
count = int(10 * self.getActivityRating(force_safe=True))
if len(self.peers) < 50:
count = max(count, 5)
count = min(count, config.connected_limit)
return count
def getActiveConnectionCountLimit(self):
count_above_preferable = 2
limit = self.getPreferableActiveConnectionCount() + count_above_preferable
limit = min(limit, config.connected_limit)
return limit
############################################################################
# Returns the maximum value of current reqs for connections
def waitingForConnections(self):
self.peer_connector.processReqs()
return self.peer_connector.need_nr_connected_peers
def needConnections(self, num=None, update_site_on_reconnect=False):
if not self.connection_server.allowsCreatingConnections():
return
if num is None:
num = self.getPreferableActiveConnectionCount()
num = min(len(self.peers), num)
req = self.peer_connector.newReq(0, num)
return req
# Wait for peers to be discovered and/or connected according to reqs
# and send updates to the UI
def waitForPeers(self, reqs, update_websocket=True):
if not reqs:
return 0
i = 0
nr_connected_peers = -1
while self.isServing():
ready_reqs = list(filter(lambda req: req.ready(), reqs))
if len(ready_reqs) == len(reqs):
if nr_connected_peers < 0:
nr_connected_peers = ready_reqs[0].nr_connected_peers
break
waiting_reqs = list(filter(lambda req: not req.ready(), reqs))
if not waiting_reqs:
break
waiting_req = waiting_reqs[0]
#self.log.debug("waiting_req: %s %s %s", waiting_req.need_nr_connected_peers, waiting_req.nr_connected_peers, waiting_req.expiration_interval)
waiting_req.waitHeartbeat(timeout=1.0)
if i > 0 and nr_connected_peers != waiting_req.nr_connected_peers:
nr_connected_peers = waiting_req.nr_connected_peers
if update_websocket:
self.updateWebsocket(connecting_to_peers=nr_connected_peers)
i += 1
if update_websocket:
self.updateWebsocket(connected_to_peers=max(nr_connected_peers, 0))
if i > 1:
# If we waited some time, pause now for displaying connected_to_peers message in the UI.
# This sleep is solely needed for site status updates on ZeroHello to be more cool-looking.
gevent.sleep(1)
return nr_connected_peers
############################################################################
# Return: Peers verified to be connectable recently, or if not enough, other peers as well
def getConnectablePeers(self, need_num=5, ignore=[], allow_private=True):
peers = list(self.peers.values())
random.shuffle(peers)
connectable_peers = []
reachable_peers = []
for peer in peers:
if peer.key in ignore:
continue
if not allow_private and helper.isPrivateIp(peer.ip):
continue
if peer.isConnectable():
connectable_peers.append(peer)
elif peer.isReachable():
reachable_peers.append(peer)
if len(connectable_peers) >= need_num:
break # Found requested number of peers
if len(connectable_peers) < need_num: # Return not that good peers
connectable_peers += reachable_peers[0:need_num - len(connectable_peers)]
return connectable_peers
# Return: Recently found peers
def getReachablePeers(self):
return [peer for peer in self.peers.values() if peer.isReachable()]
# Return: Recently found peers, sorted by reputation.
# If there not enough recently found peers, adds other known peers with highest reputation
def getRecentPeers(self, need_num):
need_num = int(need_num)
found = set(self.peers_recent)
self.log.debug(
"Recent peers %s of %s (need: %s)" %
(len(found), len(self.peers), need_num)
)
if len(found) < need_num and len(found) < len(self.peers):
# Add random peers
peers = self.getReachablePeers()
peers = sorted(
list(peers),
key=lambda peer: peer.reputation,
reverse=True
)
while len(found) < need_num and len(peers) > 0:
found.add(peers.pop())
return sorted(
list(found),
key=lambda peer: peer.reputation,
reverse=True
)[0:need_num]
return found[0:need_num]
# Returns the list of connected peers
# By default the result may contain peers chosen optimistically:
# If the connection is being established and 20 seconds have not yet passed
# since the connection start time, those peers are included in the result.
# Set only_fully_connected=True for restricting only by fully connected peers.
def getConnectedPeers(self, only_fully_connected=False):
back = []
if not self.connection_server:
return []
tor_manager = self.connection_server.tor_manager
for connection in self.connection_server.connections:
if len(back) >= len(self.peers): # short cut for breaking early; no peers to check left
break
if not connection.connected and time.time() - connection.start_time > 20: # Still not connected after 20s
continue
if not connection.connected and only_fully_connected: # Only fully connected peers
continue
peer = self.peers.get("%s:%s" % (connection.ip, connection.port))
if peer:
if connection.ip.endswith(".onion") and connection.target_onion and tor_manager.start_onions:
# Check if the connection is made with the onion address created for the site
valid_target_onions = (tor_manager.getOnion(self.address), tor_manager.getOnion("global"))
if connection.target_onion not in valid_target_onions:
continue
if not peer.connection:
peer.connect(connection)
back.append(peer)
return back
def removeDeadPeers(self):
peers = list(self.peers.values())
if len(peers) <= 20:
return
removed = 0
if len(peers) > 10000:
ttl = 60 * 2
elif len(peers) > 1000:
ttl = 60 * 60 * 1
elif len(peers) > 100:
ttl = 60 * 60 * 4
else:
ttl = 60 * 60 * 8
for peer in peers:
if peer.isConnected() or peer.isProtected():
continue
if peer.isTtlExpired(ttl):
peer.remove("TTL expired")
removed += 1
if removed > len(peers) * 0.1: # Don't remove too much at once
break
if removed:
self.log.debug("Cleanup peers result: Removed %s, left: %s" % (removed, len(self.peers)))
# Cleanup probably dead peers and close connection if too much
def cleanupPeers(self):
self.removeDeadPeers()
limit = max(self.getActiveConnectionCountLimit(), self.waitingForConnections())
connected_peers = self.getConnectedPeers(only_fully_connected=True)
need_to_close = len(connected_peers) - limit
if need_to_close > 0:
closed = 0
for peer in sorted(connected_peers, key=lambda peer: min(peer.connection.sites, 5)):
if not peer.isConnected():
continue
if peer.isProtected():
continue
if peer.connection.sites > 5:
break
peer.disconnect("Cleanup peers")
closed += 1
if closed >= need_to_close:
break
self.log.debug("Connected: %s, Need to close: %s, Closed: %s" % (
len(connected_peers), need_to_close, closed))
def lookForConnectablePeers(self):
num_tries = 2
need_connectable_peers = self.getPreferableConnectablePeerCount()
connectable_peers = 0
reachable_peers = []
for peer in list(self.peers.values()):
if peer.isConnected() or peer.isConnectable():
connectable_peers += 1
elif peer.isReachable():
reachable_peers.append(peer)
if connectable_peers >= need_connectable_peers:
return True
random.shuffle(reachable_peers)
for peer in reachable_peers:
if peer.isConnected() or peer.isConnectable() or peer.removed:
continue
peer.ping()
if peer.isConnected():
peer.pex()
num_tries -= 1
if num_tries < 1:
break
@util.Noparallel(queue=True)
def runPeriodicMaintenance(self, startup=False, force=False):
if not self.isServing():
return False
self.log.debug("runPeriodicMaintenance: startup=%s, force=%s" % (startup, force))
result = False
for handler in self.periodic_maintenance_handlers:
result = result | bool(handler.run(startup=startup, force=force))
return result
def periodicMaintenanceHandler_general(self, startup=False, force=False):
if not self.isServing():
return False
if not self.peers:
return False
self.log.debug("periodicMaintenanceHandler_general: startup=%s, force=%s" % (startup, force))
#self.persistent_peer_req = self.needConnections(update_site_on_reconnect=True)
#self.persistent_peer_req.result_connected.wait(timeout=2.0)
#self.announcer.announcePex()
self.processBackgroundPublishers()
self.update()
return True
def periodicMaintenanceHandler_peer_check(self, startup=False, force=False):
if not self.isServing():
return False
if not self.peers:
return False
self.log.debug("periodicMaintenanceHandler_peer_check: startup=%s, force=%s" % (startup, force))
if not startup:
self.cleanupPeers()
self.lookForConnectablePeers()
return True
def periodicMaintenanceHandler_announce(self, startup=False, force=False):
if not self.isServing():
return False
self.log.debug("periodicMaintenanceHandler_announce: startup=%s, force=%s" % (startup, force))
if startup and len(self.peers) < 10:
self.announce(mode="startup")
else:
self.announce(mode="update", pex=False)
return True
# Send hashfield to peers
def sendMyHashfield(self, limit=5):
if not self.isServing():
return False
if not self.content_manager.hashfield: # No optional files
return False
sent = 0
connected_peers = self.getConnectedPeers(only_fully_connected=True)
for peer in connected_peers:
if peer.sendMyHashfield():
sent += 1
if sent >= limit:
break
if sent:
my_hashfield_changed = self.content_manager.hashfield.time_changed
self.log.debug("Sent my hashfield (chaged %.3fs ago) to %s peers" % (time.time() - my_hashfield_changed, sent))
return sent
# Update hashfield
def updateHashfield(self, limit=5):
if not self.isServing():
return False
# Return if no optional files
if not self.content_manager.hashfield and not self.content_manager.has_optional_files:
return False
s = time.time()
queried = 0
connected_peers = self.getConnectedPeers(only_fully_connected=True)
for peer in connected_peers:
if peer.time_hashfield:
continue
if peer.updateHashfield():
queried += 1
if queried >= limit:
break
if queried:
self.log.debug("Queried hashfield from %s peers in %.3fs" % (queried, time.time() - s))
return queried
# Returns if the optional file is need to be downloaded or not
def isDownloadable(self, inner_path):
return self.settings.get("autodownloadoptional")
def delete(self):
self.log.info("Deleting site...")
s = time.time()
self.settings["serving"] = False
self.settings["deleting"] = True
self.saveSettings()
num_greenlets = self.greenlet_manager.stopGreenlets("Site %s deleted" % self.address)
self.worker_manager.running = False
num_workers = self.worker_manager.stopWorkers()
SiteManager.site_manager.delete(self.address)
self.content_manager.contents.db.deleteSite(self)
self.updateWebsocket(deleted=True)
self.storage.deleteFiles()
self.log.info(
"Deleted site in %.3fs (greenlets: %s, workers: %s)" %
(time.time() - s, num_greenlets, num_workers)
)
# - Events -
# Add event listeners
def addEventListeners(self):
self.onFileStart = util.Event() # If WorkerManager added new task
self.onFileDone = util.Event() # If WorkerManager successfully downloaded a file
self.onFileFail = util.Event() # If WorkerManager failed to download a file
self.onComplete = util.Event() # All files finished
self.onFileStart.append(lambda inner_path: self.fileStarted()) # No parameters to make Noparallel batching working
self.onFileDone.append(lambda inner_path: self.fileDone(inner_path))
self.onFileFail.append(lambda inner_path: self.fileFailed(inner_path))
# Send site status update to websocket clients
def updateWebsocket(self, **kwargs):
if kwargs:
param = {"event": list(kwargs.items())[0]}
else:
param = None
for ws in self.websockets:
ws.event("siteChanged", self, param)
time.sleep(0.001)
def messageWebsocket(self, message, type="info", progress=None):
for ws in self.websockets:
if progress is None:
ws.cmd("notification", [type, message])
else:
ws.cmd("progress", [type, message, progress])
# File download started
@util.Noparallel(blocking=False)
def fileStarted(self):
time.sleep(0.001) # Wait for other files adds
self.updateWebsocket(file_started=True)
# File downloaded successfully
def fileDone(self, inner_path):
# File downloaded, remove it from bad files
if inner_path in self.bad_files:
if config.verbose:
self.log.debug("Bad file solved: %s" % inner_path)
del(self.bad_files[inner_path])
# Update content.json last downlad time
if inner_path == "content.json":
if not self.settings.get("downloaded"):
self.settings["downloaded"] = int(time.time())
self.content_updated = time.time()
self.updateWebsocket(file_done=inner_path)
# File download failed
def fileFailed(self, inner_path):
if inner_path == "content.json":
self.content_updated = False
self.log.debug("Can't update content.json")
if inner_path in self.bad_files and self.connection_server.has_internet:
self.bad_files[inner_path] = self.bad_files.get(inner_path, 0) + 1
self.updateWebsocket(file_failed=inner_path)
if self.bad_files.get(inner_path, 0) > 30:
self.fileForgot(inner_path)
def fileForgot(self, inner_path):
self.log.debug("Giving up on %s" % inner_path)
del self.bad_files[inner_path] # Give up after 30 tries