lbry-sdk/lbrynet/lbrynet_daemon/LBRYDaemon.py

2791 lines
104 KiB
Python

import binascii
import logging.handlers
import mimetypes
import os
import platform
import random
import re
import socket
import string
import subprocess
import sys
import base58
import requests
import simplejson as json
from urllib2 import urlopen
from appdirs import user_data_dir
from datetime import datetime
from decimal import Decimal
from twisted.web import server
from twisted.internet import defer, threads, error, reactor, task
from twisted.internet.task import LoopingCall
from txjsonrpc import jsonrpclib
from txjsonrpc.web import jsonrpc
from txjsonrpc.web.jsonrpc import Handler
from lbrynet import __version__ as lbrynet_version
from lbryum.version import LBRYUM_VERSION as lbryum_version
from lbrynet import analytics
from lbrynet.core.PaymentRateManager import PaymentRateManager
from lbrynet.core.server.BlobAvailabilityHandler import BlobAvailabilityHandlerFactory
from lbrynet.core.server.BlobRequestHandler import BlobRequestHandlerFactory
from lbrynet.core.server.ServerProtocol import ServerProtocolFactory
from lbrynet.core.Error import UnknownNameError, InsufficientFundsError, InvalidNameError
from lbrynet.lbryfile.StreamDescriptor import LBRYFileStreamType
from lbrynet.lbryfile.client.LBRYFileDownloader import LBRYFileSaverFactory, LBRYFileOpenerFactory
from lbrynet.lbryfile.client.LBRYFileOptions import add_lbry_file_to_sd_identifier
from lbrynet.lbrynet_daemon.LBRYUIManager import LBRYUIManager
from lbrynet.lbrynet_daemon.LBRYDownloader import GetStream
from lbrynet.lbrynet_daemon.LBRYPublisher import Publisher
from lbrynet.lbrynet_daemon.LBRYExchangeRateManager import ExchangeRateManager
from lbrynet.lbrynet_daemon.Lighthouse import LighthouseClient
from lbrynet.metadata.LBRYMetadata import Metadata, verify_name_characters
from lbrynet.core import log_support
from lbrynet.core import utils
from lbrynet.core.utils import generate_id
from lbrynet.lbrynet_console.LBRYSettings import LBRYSettings
from lbrynet.conf import MIN_BLOB_DATA_PAYMENT_RATE, DEFAULT_MAX_SEARCH_RESULTS, \
KNOWN_DHT_NODES, DEFAULT_MAX_KEY_FEE, DEFAULT_WALLET, \
DEFAULT_SEARCH_TIMEOUT, DEFAULT_CACHE_TIME, DEFAULT_UI_BRANCH, \
LOG_POST_URL, LOG_FILE_NAME, REFLECTOR_SERVERS, SEARCH_SERVERS
from lbrynet.conf import DEFAULT_SD_DOWNLOAD_TIMEOUT
from lbrynet.conf import DEFAULT_TIMEOUT
from lbrynet.core.StreamDescriptor import StreamDescriptorIdentifier, download_sd_blob, BlobStreamDescriptorReader
from lbrynet.core.Session import LBRYSession
from lbrynet.core.PTCWallet import PTCWallet
from lbrynet.core.LBRYWallet import LBRYcrdWallet, LBRYumWallet
from lbrynet.lbryfilemanager.LBRYFileManager import LBRYFileManager
from lbrynet.lbryfile.LBRYFileMetadataManager import DBLBRYFileMetadataManager, TempLBRYFileMetadataManager
from lbrynet import reflector
# TODO: this code snippet is everywhere. Make it go away
if sys.platform != "darwin":
log_dir = os.path.join(os.path.expanduser("~"), ".lbrynet")
else:
log_dir = user_data_dir("LBRY")
if not os.path.isdir(log_dir):
os.mkdir(log_dir)
lbrynet_log = os.path.join(log_dir, LOG_FILE_NAME)
log = logging.getLogger(__name__)
if os.path.isfile(lbrynet_log):
with open(lbrynet_log, 'r') as f:
PREVIOUS_LBRYNET_LOG = len(f.read())
else:
PREVIOUS_LBRYNET_LOG = 0
INITIALIZING_CODE = 'initializing'
LOADING_DB_CODE = 'loading_db'
LOADING_WALLET_CODE = 'loading_wallet'
LOADING_FILE_MANAGER_CODE = 'loading_file_manager'
LOADING_SERVER_CODE = 'loading_server'
STARTED_CODE = 'started'
WAITING_FOR_FIRST_RUN_CREDITS = 'waiting_for_credits'
STARTUP_STAGES = [
(INITIALIZING_CODE, 'Initializing...'),
(LOADING_DB_CODE, 'Loading databases...'),
(LOADING_WALLET_CODE, 'Catching up with the blockchain... %s'),
(LOADING_FILE_MANAGER_CODE, 'Setting up file manager'),
(LOADING_SERVER_CODE, 'Starting lbrynet'),
(STARTED_CODE, 'Started lbrynet'),
(WAITING_FOR_FIRST_RUN_CREDITS, 'Waiting for first run credits...')
]
DOWNLOAD_METADATA_CODE = 'downloading_metadata'
DOWNLOAD_TIMEOUT_CODE = 'timeout'
DOWNLOAD_RUNNING_CODE = 'running'
DOWNLOAD_STOPPED_CODE = 'stopped'
STREAM_STAGES = [
(INITIALIZING_CODE, 'Initializing...'),
(DOWNLOAD_METADATA_CODE, 'Downloading metadata'),
(DOWNLOAD_RUNNING_CODE, 'Started %s, got %s/%s blobs, stream status: %s'),
(DOWNLOAD_STOPPED_CODE, 'Paused stream'),
(DOWNLOAD_TIMEOUT_CODE, 'Stream timed out')
]
CONNECT_CODE_VERSION_CHECK = 'version_check'
CONNECT_CODE_NETWORK = 'network_connection'
CONNECT_CODE_WALLET = 'wallet_catchup_lag'
CONNECTION_PROBLEM_CODES = [
(CONNECT_CODE_VERSION_CHECK, "There was a problem checking for updates on github"),
(CONNECT_CODE_NETWORK, "Your internet connection appears to have been interrupted"),
(CONNECT_CODE_WALLET, "Synchronization with the blockchain is lagging... if this continues try restarting LBRY")
]
ALLOWED_DURING_STARTUP = ['is_running', 'is_first_run',
'get_time_behind_blockchain', 'stop',
'daemon_status', 'get_start_notice',
'version', 'get_search_servers']
BAD_REQUEST = 400
NOT_FOUND = 404
OK_CODE = 200
# TODO add login credentials in a conf file
# TODO alert if your copy of a lbry file is out of date with the name record
REMOTE_SERVER = "www.google.com"
class Parameters(object):
def __init__(self, **kwargs):
self.__dict__.update(kwargs)
class LBRYDaemon(jsonrpc.JSONRPC):
"""
LBRYnet daemon, a jsonrpc interface to lbry functions
"""
isLeaf = True
def __init__(self, root, wallet_type=None):
jsonrpc.JSONRPC.__init__(self)
reactor.addSystemEventTrigger('before', 'shutdown', self._shutdown)
self.startup_status = STARTUP_STAGES[0]
self.startup_message = None
self.announced_startup = False
self.connected_to_internet = True
self.connection_problem = None
self.query_handlers = {}
self.git_lbrynet_version = None
self.git_lbryum_version = None
self.ui_version = None
self.ip = None
# TODO: this is confusing to set here, and then to be reset below.
self.wallet_type = wallet_type
self.first_run = None
self.log_file = lbrynet_log
self.current_db_revision = 1
self.run_server = True
self.session = None
self.exchange_rate_manager = ExchangeRateManager()
self.lighthouse_client = LighthouseClient()
self.waiting_on = {}
self.streams = {}
self.pending_claims = {}
self.known_dht_nodes = KNOWN_DHT_NODES
self.first_run_after_update = False
self.uploaded_temp_files = []
self._session_id = base58.b58encode(generate_id())
if os.name == "nt":
from lbrynet.winhelpers.knownpaths import get_path, FOLDERID, UserHandle
default_download_directory = get_path(FOLDERID.Downloads, UserHandle.current)
self.db_dir = os.path.join(get_path(FOLDERID.RoamingAppData, UserHandle.current), "lbrynet")
try:
os.makedirs(self.db_dir)
except OSError:
if not os.path.isdir(self.db_dir):
raise
elif sys.platform == "darwin":
default_download_directory = os.path.join(os.path.expanduser("~"), 'Downloads')
self.db_dir = user_data_dir("LBRY")
else:
default_download_directory = os.path.join(os.path.expanduser("~"), 'Downloads')
self.db_dir = os.path.join(os.path.expanduser("~"), ".lbrynet")
try:
if not os.path.isdir(default_download_directory):
os.mkdir(default_download_directory)
except:
log.info("Couldn't make download directory, using home")
default_download_directory = os.path.expanduser("~")
self.daemon_conf = os.path.join(self.db_dir, 'daemon_settings.json')
self.default_settings = {
'run_on_startup': False,
'data_rate': MIN_BLOB_DATA_PAYMENT_RATE,
'max_key_fee': DEFAULT_MAX_KEY_FEE,
'download_directory': default_download_directory,
'max_upload': 0.0,
'max_download': 0.0,
'upload_log': True,
'search_timeout': DEFAULT_SEARCH_TIMEOUT,
'download_timeout': DEFAULT_TIMEOUT,
'max_search_results': DEFAULT_MAX_SEARCH_RESULTS,
'wallet_type': DEFAULT_WALLET,
'delete_blobs_on_remove': True,
'peer_port': 3333,
'dht_node_port': 4444,
'reflector_port': 5566,
'use_upnp': True,
'start_lbrycrdd': True,
'requested_first_run_credits': False,
'run_reflector_server': False,
'cache_time': DEFAULT_CACHE_TIME,
'startup_scripts': [],
'last_version': {'lbrynet': lbrynet_version, 'lbryum': lbryum_version}
}
if os.path.isfile(self.daemon_conf):
f = open(self.daemon_conf, "r")
loaded_settings = json.loads(f.read())
f.close()
missing_settings = {}
removed_settings = {}
for k in self.default_settings.keys():
if k not in loaded_settings.keys():
missing_settings[k] = self.default_settings[k]
for k in loaded_settings.keys():
if not k in self.default_settings.keys():
log.info("Removing unused setting: " + k + " with value: " + str(loaded_settings[k]))
removed_settings[k] = loaded_settings[k]
del loaded_settings[k]
for k in missing_settings.keys():
log.info("Adding missing setting: " + k + " with default value: " + str(missing_settings[k]))
loaded_settings[k] = missing_settings[k]
if loaded_settings['wallet_type'] != self.wallet_type and self.wallet_type:
loaded_settings['wallet_type'] = self.wallet_type
if missing_settings or removed_settings:
log.info("Updated and loaded lbrynet-daemon configuration")
else:
log.info("Loaded lbrynet-daemon configuration")
self.session_settings = loaded_settings
else:
missing_settings = self.default_settings
log.info("Writing default settings : " + json.dumps(self.default_settings) + " --> " + str(self.daemon_conf))
self.session_settings = self.default_settings
if 'last_version' in missing_settings.keys():
self.session_settings['last_version'] = None
if self.session_settings['last_version'] != self.default_settings['last_version']:
self.session_settings['last_version'] = self.default_settings['last_version']
self.first_run_after_update = True
log.info("First run after update")
log.info("lbrynet %s --> %s" % (self.session_settings['last_version']['lbrynet'], self.default_settings['last_version']['lbrynet']))
log.info("lbryum %s --> %s" % (self.session_settings['last_version']['lbryum'], self.default_settings['last_version']['lbryum']))
if "0.4.5" == self.default_settings['last_version']['lbrynet']:
log.info("Lowering name cache time")
self.session_settings['cache_time'] = DEFAULT_CACHE_TIME
f = open(self.daemon_conf, "w")
f.write(json.dumps(self.session_settings))
f.close()
self.run_on_startup = self.session_settings['run_on_startup']
self.data_rate = self.session_settings['data_rate']
self.max_key_fee = self.session_settings['max_key_fee']
self.download_directory = self.session_settings['download_directory']
self.max_upload = self.session_settings['max_upload']
self.max_download = self.session_settings['max_download']
self.upload_log = self.session_settings['upload_log']
self.search_timeout = self.session_settings['search_timeout']
self.download_timeout = self.session_settings['download_timeout']
self.max_search_results = self.session_settings['max_search_results']
self.run_reflector_server = self.session_settings['run_reflector_server']
####
#
# Ignore the saved wallet type. Some users will have their wallet type
# saved as lbrycrd and we want wallets to be lbryum unless explicitly
# set on the command line to be lbrycrd.
#
# if self.session_settings['wallet_type'] in WALLET_TYPES and not wallet_type:
# self.wallet_type = self.session_settings['wallet_type']
# log.info("Using wallet type %s from config" % self.wallet_type)
# else:
# self.wallet_type = wallet_type
# self.session_settings['wallet_type'] = wallet_type
# log.info("Using wallet type %s specified from command line" % self.wallet_type)
#
# Instead, if wallet is not set on the command line, default to the default wallet
#
if wallet_type:
log.info("Using wallet type %s specified from command line", wallet_type)
self.wallet_type = wallet_type
else:
log.info("Using the default wallet type %s", DEFAULT_WALLET)
self.wallet_type = DEFAULT_WALLET
#
####
self.delete_blobs_on_remove = self.session_settings['delete_blobs_on_remove']
self.peer_port = self.session_settings['peer_port']
self.reflector_port = self.session_settings['reflector_port']
self.dht_node_port = self.session_settings['dht_node_port']
self.use_upnp = self.session_settings['use_upnp']
self.start_lbrycrdd = self.session_settings['start_lbrycrdd']
self.requested_first_run_credits = self.session_settings['requested_first_run_credits']
self.cache_time = self.session_settings['cache_time']
self.startup_scripts = self.session_settings['startup_scripts']
if os.path.isfile(os.path.join(self.db_dir, "stream_info_cache.json")):
f = open(os.path.join(self.db_dir, "stream_info_cache.json"), "r")
self.name_cache = json.loads(f.read())
f.close()
log.info("Loaded claim info cache")
else:
self.name_cache = {}
if os.name == "nt":
from lbrynet.winhelpers.knownpaths import get_path, FOLDERID, UserHandle
self.lbrycrdd_path = "lbrycrdd.exe"
if self.wallet_type == "lbrycrd":
self.wallet_dir = os.path.join(get_path(FOLDERID.RoamingAppData, UserHandle.current), "lbrycrd")
else:
self.wallet_dir = os.path.join(get_path(FOLDERID.RoamingAppData, UserHandle.current), "lbryum")
elif sys.platform == "darwin":
self.lbrycrdd_path = get_darwin_lbrycrdd_path()
if self.wallet_type == "lbrycrd":
self.wallet_dir = user_data_dir("lbrycrd")
else:
self.wallet_dir = user_data_dir("LBRY")
else:
self.lbrycrdd_path = "lbrycrdd"
if self.wallet_type == "lbrycrd":
self.wallet_dir = os.path.join(os.path.expanduser("~"), ".lbrycrd")
else:
self.wallet_dir = os.path.join(os.path.expanduser("~"), ".lbryum")
if os.name != 'nt':
lbrycrdd_path_conf = os.path.join(os.path.expanduser("~"), ".lbrycrddpath.conf")
if not os.path.isfile(lbrycrdd_path_conf):
f = open(lbrycrdd_path_conf, "w")
f.write(str(self.lbrycrdd_path))
f.close()
self.created_data_dir = False
if not os.path.exists(self.db_dir):
os.mkdir(self.db_dir)
self.created_data_dir = True
self.blobfile_dir = os.path.join(self.db_dir, "blobfiles")
self.lbrycrd_conf = os.path.join(self.wallet_dir, "lbrycrd.conf")
self.autofetcher_conf = os.path.join(self.wallet_dir, "autofetcher.conf")
self.wallet_conf = os.path.join(self.wallet_dir, "lbrycrd.conf")
self.wallet_user = None
self.wallet_password = None
self.internet_connection_checker = LoopingCall(self._check_network_connection)
self.version_checker = LoopingCall(self._check_remote_versions)
self.connection_problem_checker = LoopingCall(self._check_connection_problems)
self.pending_claim_checker = LoopingCall(self._check_pending_claims)
self.send_heartbeat = LoopingCall(self._send_heartbeat)
# self.lbrynet_connection_checker = LoopingCall(self._check_lbrynet_connection)
self.sd_identifier = StreamDescriptorIdentifier()
self.stream_info_manager = TempLBRYFileMetadataManager()
self.settings = LBRYSettings(self.db_dir)
self.lbry_ui_manager = LBRYUIManager(root)
self.blob_request_payment_rate_manager = None
self.lbry_file_metadata_manager = None
self.lbry_file_manager = None
if self.wallet_type == "lbrycrd":
if os.path.isfile(self.lbrycrd_conf):
log.info("Using lbrycrd.conf found at " + self.lbrycrd_conf)
else:
log.info("No lbrycrd.conf found at " + self.lbrycrd_conf + ". Generating now...")
password = "".join(random.SystemRandom().choice(string.ascii_letters + string.digits + "_") for i in range(20))
with open(self.lbrycrd_conf, 'w') as f:
f.write("rpcuser=rpcuser\n")
f.write("rpcpassword=" + password)
log.info("Done writing lbrycrd.conf")
def _responseFailed(self, err, call):
log.debug(err.getTraceback())
def render(self, request):
origin = request.getHeader("Origin")
if origin not in [None, 'http://localhost:5279']:
log.warning("Attempted api call from %s", origin)
return server.failure
request.content.seek(0, 0)
# Unmarshal the JSON-RPC data.
content = request.content.read()
parsed = jsonrpclib.loads(content)
functionPath = parsed.get("method")
args = parsed.get('params')
#TODO convert args to correct types if possible
id = parsed.get('id')
version = parsed.get('jsonrpc')
if version:
version = int(float(version))
elif id and not version:
version = jsonrpclib.VERSION_1
else:
version = jsonrpclib.VERSION_PRE1
# XXX this all needs to be re-worked to support logic for multiple
# versions...
if not self.announced_startup:
if functionPath not in ALLOWED_DURING_STARTUP:
return server.failure
if self.wallet_type == "lbryum" and functionPath in ['set_miner', 'get_miner_status']:
return server.failure
try:
function = self._getFunction(functionPath)
except jsonrpclib.Fault, f:
self._cbRender(f, request, id, version)
else:
request.setHeader("Access-Control-Allow-Origin", "localhost")
request.setHeader("content-type", "text/json")
if args == [{}]:
d = defer.maybeDeferred(function)
else:
d = defer.maybeDeferred(function, *args)
# cancel the response if the connection is broken
notify_finish = request.notifyFinish()
notify_finish.addErrback(self._responseFailed, d)
d.addErrback(self._ebRender, id)
d.addCallback(self._cbRender, request, id, version)
d.addErrback(notify_finish.errback)
return server.NOT_DONE_YET
def _cbRender(self, result, request, id, version):
def default_decimal(obj):
if isinstance(obj, Decimal):
return float(obj)
if isinstance(result, Handler):
result = result.result
if isinstance(result, dict):
result = result['result']
if version == jsonrpclib.VERSION_PRE1:
if not isinstance(result, jsonrpclib.Fault):
result = (result,)
# Convert the result (python) to JSON-RPC
try:
s = jsonrpclib.dumps(result, version=version, default=default_decimal)
except:
f = jsonrpclib.Fault(self.FAILURE, "can't serialize output")
s = jsonrpclib.dumps(f, version=version)
request.setHeader("content-length", str(len(s)))
request.write(s)
request.finish()
def _ebRender(self, failure, id):
if isinstance(failure.value, jsonrpclib.Fault):
return failure.value
log.error(failure)
return jsonrpclib.Fault(self.FAILURE, "error")
def setup(self, branch=DEFAULT_UI_BRANCH, user_specified=False, branch_specified=False, host_ui=True):
def _log_starting_vals():
log.info("Starting balance: " + str(self.session.wallet.wallet_balance))
return defer.succeed(None)
def _announce_startup():
def _wait_for_credits():
if float(self.session.wallet.wallet_balance) == 0.0:
self.startup_status = STARTUP_STAGES[6]
return reactor.callLater(1, _wait_for_credits)
else:
return _announce()
def _announce():
self.announced_startup = True
self.startup_status = STARTUP_STAGES[5]
log.info("Started lbrynet-daemon")
if len(self.startup_scripts):
log.info("Scheduling scripts")
reactor.callLater(3, self._run_scripts)
if self.first_run:
d = self._upload_log(log_type="first_run")
elif self.upload_log:
d = self._upload_log(exclude_previous=True, log_type="start")
else:
d = defer.succeed(None)
d.addCallback(lambda _: _announce())
return d
log.info("Starting lbrynet-daemon")
self.internet_connection_checker.start(3600)
self.version_checker.start(3600 * 12)
self.connection_problem_checker.start(1)
self.exchange_rate_manager.start()
if host_ui:
self.lbry_ui_manager.update_checker.start(1800, now=False)
d = defer.Deferred()
if host_ui:
d.addCallback(lambda _: self.lbry_ui_manager.setup(branch=branch,
user_specified=user_specified,
branch_specified=branch_specified))
d.addCallback(lambda _: self._initial_setup())
d.addCallback(lambda _: threads.deferToThread(self._setup_data_directory))
d.addCallback(lambda _: self._check_db_migration())
d.addCallback(lambda _: self._get_settings())
d.addCallback(lambda _: self._set_events())
d.addCallback(lambda _: self._get_session())
d.addCallback(lambda _: add_lbry_file_to_sd_identifier(self.sd_identifier))
d.addCallback(lambda _: self._setup_stream_identifier())
d.addCallback(lambda _: self._setup_lbry_file_manager())
d.addCallback(lambda _: self._setup_lbry_file_opener())
d.addCallback(lambda _: self._setup_query_handlers())
d.addCallback(lambda _: self._setup_server())
d.addCallback(lambda _: _log_starting_vals())
d.addCallback(lambda _: _announce_startup())
d.addCallback(lambda _: self._load_analytics_api())
# TODO: handle errors here
d.callback(None)
return defer.succeed(None)
def _load_analytics_api(self):
self.analytics_api = analytics.Api.load()
self.send_heartbeat.start(60)
def _send_heartbeat(self):
log.debug('Sending heartbeat')
heartbeat = self._events.heartbeat()
self.analytics_api.track(heartbeat)
def _get_platform(self):
r = {
"processor": platform.processor(),
"python_version": platform.python_version(),
"platform": platform.platform(),
"os_release": platform.release(),
"os_system": platform.system(),
"lbrynet_version": lbrynet_version,
"lbryum_version": lbryum_version,
"ui_version": self.lbry_ui_manager.loaded_git_version,
}
if not self.ip:
try:
r['ip'] = json.load(urlopen('http://jsonip.com'))['ip']
self.ip = r['ip']
except:
r['ip'] = "Could not determine"
return r
def _initial_setup(self):
def _log_platform():
log.info("Platform: %s", json.dumps(self._get_platform()))
return defer.succeed(None)
d = _log_platform()
return d
def _set_events(self):
context = analytics.make_context(self._get_platform(), self.wallet_type)
self._events = analytics.Events(context, base58.b58encode(self.lbryid), self._session_id)
def _check_network_connection(self):
try:
host = socket.gethostbyname(REMOTE_SERVER)
s = socket.create_connection((host, 80), 2)
self.connected_to_internet = True
except:
log.info("Internet connection not working")
self.connected_to_internet = False
def _check_lbrynet_connection(self):
def _log_success():
log.info("lbrynet connectivity test passed")
def _log_failure():
log.info("lbrynet connectivity test failed")
wonderfullife_sh = "6f3af0fa3924be98a54766aa2715d22c6c1509c3f7fa32566df4899a41f3530a9f97b2ecb817fa1dcbf1b30553aefaa7"
d = download_sd_blob(self.session, wonderfullife_sh, self.session.base_payment_rate_manager)
d.addCallbacks(lambda _: _log_success, lambda _: _log_failure)
def _check_remote_versions(self):
def _get_lbryum_version():
try:
r = urlopen("https://raw.githubusercontent.com/lbryio/lbryum/master/lib/version.py").read().split('\n')
version = next(line.split("=")[1].split("#")[0].replace(" ", "")
for line in r if "LBRYUM_VERSION" in line)
version = version.replace("'", "")
log.info(
"remote lbryum %s > local lbryum %s = %s",
version, lbryum_version,
utils.version_is_greater_than(version, lbryum_version)
)
self.git_lbryum_version = version
return defer.succeed(None)
except:
log.info("Failed to get lbryum version from git")
self.git_lbryum_version = None
return defer.fail(None)
def _get_lbrynet_version():
try:
version = get_lbrynet_version_from_github()
log.info(
"remote lbrynet %s > local lbrynet %s = %s",
version, lbrynet_version,
utils.version_is_greater_than(version, lbrynet_version)
)
self.git_lbrynet_version = version
return defer.succeed(None)
except:
log.info("Failed to get lbrynet version from git")
self.git_lbrynet_version = None
return defer.fail(None)
d = _get_lbrynet_version()
d.addCallback(lambda _: _get_lbryum_version())
def _check_connection_problems(self):
if not self.git_lbrynet_version or not self.git_lbryum_version:
self.connection_problem = CONNECTION_PROBLEM_CODES[0]
elif self.startup_status[0] == 'loading_wallet':
if self.session.wallet.is_lagging:
self.connection_problem = CONNECTION_PROBLEM_CODES[2]
else:
self.connection_problem = None
if not self.connected_to_internet:
self.connection_problem = CONNECTION_PROBLEM_CODES[1]
def _add_to_pending_claims(self, name, txid):
log.info("Adding lbry://%s to pending claims, txid %s" % (name, txid))
self.pending_claims[name] = txid
return txid
def _check_pending_claims(self):
# TODO: this was blatantly copied from jsonrpc_start_lbry_file. Be DRY.
def _start_file(f):
d = self.lbry_file_manager.toggle_lbry_file_running(f)
d.addCallback(lambda _: self.lighthouse_client.announce_sd(f.sd_hash))
return defer.succeed("Started LBRY file")
def _get_and_start_file(name):
d = defer.succeed(self.pending_claims.pop(name))
d.addCallback(lambda _: self._get_lbry_file("name", name, return_json=False))
d.addCallback(lambda l: _start_file(l) if l.stopped else "LBRY file was already running")
def re_add_to_pending_claims(name):
txid = self.pending_claims.pop(name)
self._add_to_pending_claims(name, txid)
def _process_lbry_file(name, lbry_file):
# lbry_file is an instance of ManagedLBRYFileDownloader or None
# TODO: check for sd_hash in addition to txid
ready_to_start = (
lbry_file and
self.pending_claims[name] == lbry_file.txid
)
if ready_to_start:
_get_and_start_file(name)
else:
re_add_to_pending_claims(name)
for name in self.pending_claims:
log.info("Checking if new claim for lbry://%s is confirmed" % name)
d = self._resolve_name(name, force_refresh=True)
d.addCallback(lambda _: self._get_lbry_file_by_uri(name))
d.addCallbacks(
lambda lbry_file: _process_lbry_file(name, lbry_file),
lambda _: re_add_to_pending_claims(name)
)
def _start_server(self):
if self.peer_port is not None:
server_factory = ServerProtocolFactory(self.session.rate_limiter,
self.query_handlers,
self.session.peer_manager)
try:
self.lbry_server_port = reactor.listenTCP(self.peer_port, server_factory)
except error.CannotListenError as e:
import traceback
log.error("Couldn't bind to port %d. %s", self.peer_port, traceback.format_exc())
raise ValueError("%s lbrynet may already be running on your computer.", str(e))
return defer.succeed(True)
def _start_reflector(self):
if self.run_reflector_server:
log.info("Starting reflector server")
if self.reflector_port is not None:
reflector_factory = reflector.ServerFactory(
self.session.peer_manager,
self.session.blob_manager
)
try:
self.reflector_server_port = reactor.listenTCP(self.reflector_port, reflector_factory)
log.info('Started reflector on port %s', self.reflector_port)
except error.CannotListenError as e:
log.exception("Couldn't bind reflector to port %d", self.reflector_port)
raise ValueError("{} lbrynet may already be running on your computer.".format(e))
return defer.succeed(True)
def _stop_reflector(self):
if self.run_reflector_server:
log.info("Stopping reflector server")
try:
if self.reflector_server_port is not None:
self.reflector_server_port, p = None, self.reflector_server_port
return defer.maybeDeferred(p.stopListening)
except AttributeError:
return defer.succeed(True)
return defer.succeed(True)
def _stop_server(self):
try:
if self.lbry_server_port is not None:
self.lbry_server_port, p = None, self.lbry_server_port
return defer.maybeDeferred(p.stopListening)
else:
return defer.succeed(True)
except AttributeError:
return defer.succeed(True)
def _setup_server(self):
def restore_running_status(running):
if running is True:
d = self._start_server()
d.addCallback(lambda _: self._start_reflector())
return defer.succeed(True)
self.startup_status = STARTUP_STAGES[4]
dl = self.settings.get_server_running_status()
dl.addCallback(restore_running_status)
return dl
def _setup_query_handlers(self):
handlers = [
# CryptBlobInfoQueryHandlerFactory(self.lbry_file_metadata_manager, self.session.wallet,
# self._server_payment_rate_manager),
BlobAvailabilityHandlerFactory(self.session.blob_manager),
# BlobRequestHandlerFactory(self.session.blob_manager, self.session.wallet,
# self._server_payment_rate_manager),
self.session.wallet.get_wallet_info_query_handler_factory(),
]
def get_blob_request_handler_factory(rate):
self.blob_request_payment_rate_manager = PaymentRateManager(
self.session.base_payment_rate_manager, rate
)
handlers.append(BlobRequestHandlerFactory(self.session.blob_manager, self.session.wallet,
self.blob_request_payment_rate_manager))
d1 = self.settings.get_server_data_payment_rate()
d1.addCallback(get_blob_request_handler_factory)
dl = defer.DeferredList([d1])
dl.addCallback(lambda _: self._add_query_handlers(handlers))
return dl
def _add_query_handlers(self, query_handlers):
def _set_query_handlers(statuses):
from future_builtins import zip
for handler, (success, status) in zip(query_handlers, statuses):
if success is True:
self.query_handlers[handler] = status
ds = []
for handler in query_handlers:
ds.append(self.settings.get_query_handler_status(handler.get_primary_query_identifier()))
dl = defer.DeferredList(ds)
dl.addCallback(_set_query_handlers)
return dl
def _upload_log(self, log_type=None, exclude_previous=False, force=False):
if self.upload_log or force:
for lm, lp in [('lbrynet', lbrynet_log)]: #, ('lbryum', lbryum_log)]:
if os.path.isfile(lp):
if exclude_previous:
f = open(lp, "r")
f.seek(PREVIOUS_LBRYNET_LOG) # if lm == 'lbrynet' else PREVIOUS_LBRYUM_LOG)
log_contents = f.read()
f.close()
else:
f = open(lp, "r")
log_contents = f.read()
f.close()
params = {
'date': datetime.utcnow().strftime('%Y%m%d-%H%M%S'),
'hash': base58.b58encode(self.lbryid)[:20],
'sys': platform.system(),
'type': "%s-%s" % (lm, log_type) if log_type else lm,
'log': log_contents
}
requests.post(LOG_POST_URL, params)
return defer.succeed(None)
else:
return defer.succeed(None)
def _clean_up_temp_files(self):
for path in self.uploaded_temp_files:
try:
os.remove(path)
except OSError:
pass
def _shutdown(self):
log.info("Closing lbrynet session")
log.info("Status at time of shutdown: " + self.startup_status[0])
if self.internet_connection_checker.running:
self.internet_connection_checker.stop()
if self.version_checker.running:
self.version_checker.stop()
if self.connection_problem_checker.running:
self.connection_problem_checker.stop()
if self.lbry_ui_manager.update_checker.running:
self.lbry_ui_manager.update_checker.stop()
if self.pending_claim_checker.running:
self.pending_claim_checker.stop()
if self.send_heartbeat.running:
self.send_heartbeat.stop()
self._clean_up_temp_files()
d = self._upload_log(log_type="close", exclude_previous=False if self.first_run else True)
d.addCallback(lambda _: self._stop_server())
d.addCallback(lambda _: self._stop_reflector())
d.addErrback(lambda err: True)
d.addCallback(lambda _: self.lbry_file_manager.stop())
d.addErrback(lambda err: True)
if self.session is not None:
d.addCallback(lambda _: self.session.shut_down())
d.addErrback(lambda err: True)
return d
def _update_settings(self, settings):
for k in settings.keys():
if k == 'run_on_startup':
if type(settings['run_on_startup']) is bool:
self.session_settings['run_on_startup'] = settings['run_on_startup']
else:
return defer.fail()
elif k == 'data_rate':
if type(settings['data_rate']) is float:
self.session_settings['data_rate'] = settings['data_rate']
elif type(settings['data_rate']) is int:
self.session_settings['data_rate'] = float(settings['data_rate'])
else:
return defer.fail()
elif k == 'max_key_fee':
if type(settings['max_key_fee']) is float:
self.session_settings['max_key_fee'] = settings['max_key_fee']
elif type(settings['max_key_fee']) is int:
self.session_settings['max_key_fee'] = float(settings['max_key_fee'])
else:
return defer.fail()
elif k == 'download_directory':
if type(settings['download_directory']) is unicode:
if os.path.isdir(settings['download_directory']):
self.session_settings['download_directory'] = settings['download_directory']
else:
pass
else:
return defer.fail()
elif k == 'max_upload':
if type(settings['max_upload']) is float:
self.session_settings['max_upload'] = settings['max_upload']
elif type(settings['max_upload']) is int:
self.session_settings['max_upload'] = float(settings['max_upload'])
else:
return defer.fail()
elif k == 'max_download':
if type(settings['max_download']) is float:
self.session_settings['max_download'] = settings['max_download']
if type(settings['max_download']) is int:
self.session_settings['max_download'] = float(settings['max_download'])
else:
return defer.fail()
elif k == 'upload_log':
if type(settings['upload_log']) is bool:
self.session_settings['upload_log'] = settings['upload_log']
else:
return defer.fail()
elif k == 'download_timeout':
if type(settings['download_timeout']) is int:
self.session_settings['download_timeout'] = settings['download_timeout']
elif type(settings['download_timeout']) is float:
self.session_settings['download_timeout'] = int(settings['download_timeout'])
else:
return defer.fail()
elif k == 'search_timeout':
if type(settings['search_timeout']) is float:
self.session_settings['search_timeout'] = settings['search_timeout']
elif type(settings['search_timeout']) is int:
self.session_settings['search_timeout'] = float(settings['search_timeout'])
else:
return defer.fail()
elif k == 'cache_time':
if type(settings['cache_time']) is int:
self.session_settings['cache_time'] = settings['cache_time']
elif type(settings['cache_time']) is float:
self.session_settings['cache_time'] = int(settings['cache_time'])
else:
return defer.fail()
self.run_on_startup = self.session_settings['run_on_startup']
self.data_rate = self.session_settings['data_rate']
self.max_key_fee = self.session_settings['max_key_fee']
self.download_directory = self.session_settings['download_directory']
self.max_upload = self.session_settings['max_upload']
self.max_download = self.session_settings['max_download']
self.upload_log = self.session_settings['upload_log']
self.download_timeout = self.session_settings['download_timeout']
self.search_timeout = self.session_settings['search_timeout']
self.cache_time = self.session_settings['cache_time']
f = open(self.daemon_conf, "w")
f.write(json.dumps(self.session_settings))
f.close()
return defer.succeed(True)
def _setup_data_directory(self):
self.startup_status = STARTUP_STAGES[1]
log.info("Loading databases...")
if self.created_data_dir:
db_revision = open(os.path.join(self.db_dir, "db_revision"), mode='w')
db_revision.write(str(self.current_db_revision))
db_revision.close()
log.debug("Created the db revision file: %s", str(os.path.join(self.db_dir, "db_revision")))
if not os.path.exists(self.blobfile_dir):
os.mkdir(self.blobfile_dir)
log.debug("Created the blobfile directory: %s", str(self.blobfile_dir))
def _check_db_migration(self):
old_revision = 1
db_revision_file = os.path.join(self.db_dir, "db_revision")
if os.path.exists(db_revision_file):
old_revision = int(open(db_revision_file).read().strip())
if old_revision < self.current_db_revision:
from lbrynet.db_migrator import dbmigrator
log.info("Upgrading your databases...")
d = threads.deferToThread(dbmigrator.migrate_db, self.db_dir, old_revision, self.current_db_revision)
def print_success(old_dirs):
success_string = "Finished upgrading the databases. It is now safe to delete the"
success_string += " following directories, if you feel like it. It won't make any"
success_string += " difference.\nAnyway here they are: "
for i, old_dir in enumerate(old_dirs):
success_string += old_dir
if i + 1 < len(old_dir):
success_string += ", "
log.info(success_string)
d.addCallback(print_success)
return d
return defer.succeed(True)
def _get_settings(self):
d = self.settings.start()
d.addCallback(lambda _: self.settings.get_lbryid())
d.addCallback(self._set_lbryid)
d.addCallback(lambda _: self._modify_loggly_formatter())
return d
def _set_lbryid(self, lbryid):
if lbryid is None:
return self._make_lbryid()
else:
log.info("LBRY ID: " + base58.b58encode(lbryid))
self.lbryid = lbryid
def _make_lbryid(self):
self.lbryid = generate_id()
log.info("Generated new LBRY ID: " + base58.b58encode(self.lbryid))
d = self.settings.save_lbryid(self.lbryid)
return d
def _modify_loggly_formatter(self):
log_support.configure_loggly_handler(
lbry_id=base58.b58encode(self.lbryid),
session_id=self._session_id
)
def _setup_lbry_file_manager(self):
self.startup_status = STARTUP_STAGES[3]
self.lbry_file_metadata_manager = DBLBRYFileMetadataManager(self.db_dir)
d = self.lbry_file_metadata_manager.setup()
def set_lbry_file_manager():
self.lbry_file_manager = LBRYFileManager(self.session,
self.lbry_file_metadata_manager,
self.sd_identifier,
download_directory=self.download_directory)
return self.lbry_file_manager.setup()
d.addCallback(lambda _: set_lbry_file_manager())
return d
def _get_session(self):
def get_default_data_rate():
d = self.settings.get_default_data_payment_rate()
d.addCallback(lambda rate: {"default_data_payment_rate": rate if rate is not None else
MIN_BLOB_DATA_PAYMENT_RATE})
return d
def get_wallet():
if self.wallet_type == "lbrycrd":
log.info("Using lbrycrd wallet")
d = defer.succeed(LBRYcrdWallet(self.db_dir, wallet_dir=self.wallet_dir, wallet_conf=self.lbrycrd_conf,
lbrycrdd_path=self.lbrycrdd_path))
elif self.wallet_type == "lbryum":
log.info("Using lbryum wallet")
d = defer.succeed(LBRYumWallet(self.db_dir))
elif self.wallet_type == "ptc":
log.info("Using PTC wallet")
d = defer.succeed(PTCWallet(self.db_dir))
else:
# TODO: should fail here. Can't switch to lbrycrd because the wallet_dir, conf and path won't be set
log.info("Requested unknown wallet '%s', using default lbryum", self.wallet_type)
d = defer.succeed(LBRYumWallet(self.db_dir))
d.addCallback(lambda wallet: {"wallet": wallet})
return d
d1 = get_default_data_rate()
d2 = get_wallet()
def combine_results(results):
r = {}
for success, result in results:
if success is True:
r.update(result)
return r
def create_session(results):
self.session = LBRYSession(results['default_data_payment_rate'], db_dir=self.db_dir, lbryid=self.lbryid,
blob_dir=self.blobfile_dir, dht_node_port=self.dht_node_port,
known_dht_nodes=self.known_dht_nodes, peer_port=self.peer_port,
use_upnp=self.use_upnp, wallet=results['wallet'])
self.startup_status = STARTUP_STAGES[2]
dl = defer.DeferredList([d1, d2], fireOnOneErrback=True)
dl.addCallback(combine_results)
dl.addCallback(create_session)
dl.addCallback(lambda _: self.session.setup())
return dl
def _setup_stream_identifier(self):
file_saver_factory = LBRYFileSaverFactory(self.session.peer_finder, self.session.rate_limiter,
self.session.blob_manager, self.stream_info_manager,
self.session.wallet, self.download_directory)
self.sd_identifier.add_stream_downloader_factory(LBRYFileStreamType, file_saver_factory)
file_opener_factory = LBRYFileOpenerFactory(self.session.peer_finder, self.session.rate_limiter,
self.session.blob_manager, self.stream_info_manager,
self.session.wallet)
self.sd_identifier.add_stream_downloader_factory(LBRYFileStreamType, file_opener_factory)
return defer.succeed(None)
def _setup_lbry_file_opener(self):
downloader_factory = LBRYFileOpenerFactory(self.session.peer_finder, self.session.rate_limiter,
self.session.blob_manager, self.stream_info_manager,
self.session.wallet)
self.sd_identifier.add_stream_downloader_factory(LBRYFileStreamType, downloader_factory)
return defer.succeed(True)
def _download_sd_blob(self, sd_hash, timeout=DEFAULT_SD_DOWNLOAD_TIMEOUT):
def cb(result):
if not r.called:
r.callback(result)
def eb():
if not r.called:
r.errback(Exception("sd timeout"))
r = defer.Deferred(None)
reactor.callLater(timeout, eb)
d = download_sd_blob(self.session, sd_hash, PaymentRateManager(self.session.base_payment_rate_manager))
d.addCallback(BlobStreamDescriptorReader)
d.addCallback(lambda blob: blob.get_info())
d.addCallback(cb)
return r
def _download_name(self, name, timeout=DEFAULT_TIMEOUT, download_directory=None,
file_name=None, stream_info=None, wait_for_write=True):
"""
Add a lbry file to the file manager, start the download, and return the new lbry file.
If it already exists in the file manager, return the existing lbry file
"""
helper = _DownloadNameHelper(
self, name, timeout, download_directory, file_name, wait_for_write)
if not stream_info:
self.waiting_on[name] = True
d = self._resolve_name(name)
else:
d = defer.succeed(stream_info)
d.addCallback(helper._setup_stream)
d.addCallback(helper.wait_or_get_stream)
if not stream_info:
d.addCallback(helper._remove_from_wait)
return d
def add_stream(self, name, timeout, download_directory, file_name, stream_info):
"""Makes, adds and starts a stream"""
self.streams[name] = GetStream(self.sd_identifier,
self.session,
self.session.wallet,
self.lbry_file_manager,
self.exchange_rate_manager,
max_key_fee=self.max_key_fee,
data_rate=self.data_rate,
timeout=timeout,
download_directory=download_directory,
file_name=file_name)
d = self.streams[name].start(stream_info, name)
return d
def _get_long_count_timestamp(self):
return int((datetime.utcnow() - (datetime(year=2012, month=12, day=21))).total_seconds())
def _update_claim_cache(self):
f = open(os.path.join(self.db_dir, "stream_info_cache.json"), "w")
f.write(json.dumps(self.name_cache))
f.close()
return defer.succeed(True)
def _resolve_name(self, name, force_refresh=False):
"""Resolves a name. Checks the cache first before going out to the blockchain.
Args:
name: the lbry://<name> to resolve
force_refresh: if True, always go out to the blockchain to resolve.
"""
if name.startswith('lbry://'):
raise ValueError('name %s should not start with lbry://')
helper = _ResolveNameHelper(self, name, force_refresh)
return helper.get_deferred()
def _delete_lbry_file(self, lbry_file, delete_file=True):
d = self.lbry_file_manager.delete_lbry_file(lbry_file)
def finish_deletion(lbry_file):
d = lbry_file.delete_data()
d.addCallback(lambda _: _delete_stream_data(lbry_file))
return d
def _delete_stream_data(lbry_file):
s_h = lbry_file.stream_hash
d = self.lbry_file_manager.get_count_for_stream_hash(s_h)
# TODO: could possibly be a timing issue here
d.addCallback(lambda c: self.stream_info_manager.delete_stream(s_h) if c == 0 else True)
if delete_file:
d.addCallback(lambda _: os.remove(os.path.join(self.download_directory, lbry_file.file_name)) if
os.path.isfile(os.path.join(self.download_directory, lbry_file.file_name)) else defer.succeed(None))
return d
d.addCallback(lambda _: finish_deletion(lbry_file))
d.addCallback(lambda _: log.info("Delete lbry file"))
return d
def _get_est_cost(self, name):
def _check_est(d, name):
try:
if isinstance(d.result, float):
log.info("Cost est for lbry://" + name + ": " + str(d.result) + "LBC")
return defer.succeed(None)
except AttributeError:
pass
log.info("Timeout estimating cost for lbry://" + name + ", using key fee")
d.cancel()
return defer.succeed(None)
def _add_key_fee(data_cost):
d = self._resolve_name(name)
d.addCallback(lambda info: self.exchange_rate_manager.to_lbc(info.get('fee', None)))
d.addCallback(lambda fee: data_cost if fee is None else data_cost + fee.amount)
return d
d = self._resolve_name(name)
d.addCallback(lambda info: info['sources']['lbry_sd_hash'])
d.addCallback(lambda sd_hash: download_sd_blob(self.session, sd_hash,
self.blob_request_payment_rate_manager))
d.addCallback(self.sd_identifier.get_metadata_for_sd_blob)
d.addCallback(lambda metadata: metadata.validator.info_to_show())
d.addCallback(lambda info: int(dict(info)['stream_size']) / 1000000 * self.data_rate)
d.addCallbacks(_add_key_fee, lambda _: _add_key_fee(0.0))
reactor.callLater(self.search_timeout, _check_est, d, name)
return d
def _get_lbry_file_by_uri(self, name):
def _get_file(stream_info):
sd = stream_info['sources']['lbry_sd_hash']
for l in self.lbry_file_manager.lbry_files:
if l.sd_hash == sd:
return defer.succeed(l)
return defer.succeed(None)
d = self._resolve_name(name)
d.addCallback(_get_file)
return d
def _get_lbry_file_by_sd_hash(self, sd_hash):
for l in self.lbry_file_manager.lbry_files:
if l.sd_hash == sd_hash:
return defer.succeed(l)
return defer.succeed(None)
def _get_lbry_file_by_file_name(self, file_name):
for l in self.lbry_file_manager.lbry_files:
if l.file_name == file_name:
return defer.succeed(l)
return defer.succeed(None)
def _get_lbry_file(self, search_by, val, return_json=True):
def _log_get_lbry_file(f):
if f and val:
log.info("Found LBRY file for " + search_by + ": " + val)
elif val:
log.info("Did not find LBRY file for " + search_by + ": " + val)
return f
def _get_json_for_return(f):
def _get_file_status(file_status):
message = STREAM_STAGES[2][1] % (file_status.name, file_status.num_completed, file_status.num_known, file_status.running_status)
return defer.succeed(message)
def _generate_reply(size):
if f.key:
key = binascii.b2a_hex(f.key)
else:
key = None
if os.path.isfile(os.path.join(self.download_directory, f.file_name)):
written_file = file(os.path.join(self.download_directory, f.file_name))
written_file.seek(0, os.SEEK_END)
written_bytes = written_file.tell()
written_file.close()
else:
written_bytes = False
if search_by == "name":
if val in self.streams.keys():
status = self.streams[val].code
elif f in self.lbry_file_manager.lbry_files:
# if f.stopped:
# status = STREAM_STAGES[3]
# else:
status = STREAM_STAGES[2]
else:
status = [False, False]
else:
status = [False, False]
if status[0] == DOWNLOAD_RUNNING_CODE:
d = f.status()
d.addCallback(_get_file_status)
d.addCallback(lambda message: {'completed': f.completed, 'file_name': f.file_name,
'download_directory': f.download_directory,
'download_path': os.path.join(f.download_directory, f.file_name),
'mime_type': mimetypes.guess_type(os.path.join(f.download_directory, f.file_name))[0],
'key': key,
'points_paid': f.points_paid, 'stopped': f.stopped,
'stream_hash': f.stream_hash,
'stream_name': f.stream_name,
'suggested_file_name': f.suggested_file_name,
'upload_allowed': f.upload_allowed, 'sd_hash': f.sd_hash,
'lbry_uri': f.uri, 'txid': f.txid, 'claim_id': f.claim_id,
'total_bytes': size,
'written_bytes': written_bytes, 'code': status[0],
'message': message})
else:
d = defer.succeed({'completed': f.completed, 'file_name': f.file_name, 'key': key,
'download_directory': f.download_directory,
'download_path': os.path.join(f.download_directory, f.file_name),
'mime_type': mimetypes.guess_type(os.path.join(f.download_directory, f.file_name))[0],
'points_paid': f.points_paid, 'stopped': f.stopped, 'stream_hash': f.stream_hash,
'stream_name': f.stream_name, 'suggested_file_name': f.suggested_file_name,
'upload_allowed': f.upload_allowed, 'sd_hash': f.sd_hash, 'total_bytes': size,
'written_bytes': written_bytes, 'lbry_uri': f.uri, 'txid': f.txid, 'claim_id': f.claim_id,
'code': status[0], 'message': status[1]})
return d
def _add_metadata(message):
def _add_to_dict(metadata):
message['metadata'] = metadata
return defer.succeed(message)
if f.txid:
d = self._resolve_name(f.uri)
d.addCallbacks(_add_to_dict, lambda _: _add_to_dict("Pending confirmation"))
else:
d = defer.succeed(message)
return d
if f:
d = f.get_total_bytes()
d.addCallback(_generate_reply)
d.addCallback(_add_metadata)
return d
else:
return False
if search_by == "name":
d = self._get_lbry_file_by_uri(val)
elif search_by == "sd_hash":
d = self._get_lbry_file_by_sd_hash(val)
elif search_by == "file_name":
d = self._get_lbry_file_by_file_name(val)
# d.addCallback(_log_get_lbry_file)
if return_json:
d.addCallback(_get_json_for_return)
return d
def _get_lbry_files(self):
d = defer.DeferredList([self._get_lbry_file('sd_hash', l.sd_hash) for l in self.lbry_file_manager.lbry_files])
return d
def _reflect(self, lbry_file):
if not lbry_file:
return defer.fail(Exception("no lbry file given to reflect"))
stream_hash = lbry_file.stream_hash
if stream_hash is None:
return defer.fail(Exception("no stream hash"))
log.info("Reflecting stream: %s" % stream_hash)
reflector_server = random.choice(REFLECTOR_SERVERS)
reflector_address, reflector_port = reflector_server[0], reflector_server[1]
log.info("Start reflector client")
factory = reflector.ClientFactory(
self.session.blob_manager,
self.lbry_file_manager.stream_info_manager,
stream_hash
)
d = reactor.resolve(reflector_address)
d.addCallback(lambda ip: reactor.connectTCP(ip, reflector_port, factory))
d.addCallback(lambda _: factory.finished_deferred)
return d
def _reflect_blobs(self, blob_hashes):
if not blob_hashes:
return defer.fail(Exception("no lbry file given to reflect"))
log.info("Reflecting %i blobs" % len(blob_hashes))
reflector_server = random.choice(REFLECTOR_SERVERS)
reflector_address, reflector_port = reflector_server[0], reflector_server[1]
log.info("Start reflector client")
factory = reflector.BlobClientFactory(
self.session.blob_manager,
blob_hashes
)
d = reactor.resolve(reflector_address)
d.addCallback(lambda ip: reactor.connectTCP(ip, reflector_port, factory))
d.addCallback(lambda _: factory.finished_deferred)
return d
def _log_to_slack(self, msg):
URL = "https://hooks.slack.com/services/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"
msg = platform.platform() + ": " + base58.b58encode(self.lbryid)[:20] + ", " + msg
requests.post(URL, json.dumps({"text": msg}))
return defer.succeed(None)
def _run_scripts(self):
if len([k for k in self.startup_scripts if 'run_once' in k.keys()]):
log.info("Removing one time startup scripts")
remaining_scripts = [s for s in self.startup_scripts if 'run_once' not in s.keys()]
startup_scripts = self.startup_scripts
self.startup_scripts = self.session_settings['startup_scripts'] = remaining_scripts
f = open(self.daemon_conf, "w")
f.write(json.dumps(self.session_settings))
f.close()
for script in startup_scripts:
if script['script_name'] == 'migrateto025':
log.info("Running migrator to 0.2.5")
from lbrynet.lbrynet_daemon.daemon_scripts.migrateto025 import run as run_migrate
run_migrate(self)
if script['script_name'] == 'Autofetcher':
log.info("Starting autofetcher script")
from lbrynet.lbrynet_daemon.daemon_scripts.Autofetcher import run as run_autofetcher
run_autofetcher(self)
return defer.succeed(None)
def _search(self, search):
return self.lighthouse_client.search(search)
def _render_response(self, result, code):
return defer.succeed({'result': result, 'code': code})
def jsonrpc_is_running(self):
"""
Check if lbrynet daemon is running
Args:
None
Returns: true if daemon completed startup, otherwise false
"""
log.info("is_running: " + str(self.announced_startup))
if self.announced_startup:
return self._render_response(True, OK_CODE)
else:
return self._render_response(False, OK_CODE)
def jsonrpc_daemon_status(self):
"""
Get lbrynet daemon status information
Args:
None
Returns:
'message': startup status message
'code': status_code
'progress': progress, only used in loading_wallet
'is_lagging': flag set to indicate lag, if set message will contain relevant message
"""
r = {'code': self.startup_status[0], 'message': self.startup_status[1],
'progress': None, 'is_lagging': None, 'problem_code': None}
if self.connection_problem:
r['problem_code'] = self.connection_problem[0]
r['message'] = self.connection_problem[1]
r['is_lagging'] = True
elif self.startup_status[0] == LOADING_WALLET_CODE:
if self.wallet_type == 'lbryum':
if self.session.wallet.blocks_behind_alert != 0:
r['message'] = r['message'] % (str(self.session.wallet.blocks_behind_alert) + " blocks behind")
r['progress'] = self.session.wallet.catchup_progress
else:
r['message'] = "Catching up with the blockchain"
r['progress'] = 0
else:
r['message'] = "Catching up with the blockchain"
r['progress'] = 0
log.info("daemon status: " + str(r))
return self._render_response(r, OK_CODE)
def jsonrpc_is_first_run(self):
"""
Check if this is the first time lbrynet daemon has been run
Args:
None
Returns:
True if first run, otherwise False
"""
log.info("Check if is first run")
try:
d = self.session.wallet.is_first_run()
except:
d = defer.fail(None)
d.addCallbacks(lambda r: self._render_response(r, OK_CODE), lambda _: self._render_response(None, OK_CODE))
return d
def jsonrpc_get_start_notice(self):
"""
Get special message to be displayed at startup
Args:
None
Returns:
Startup message, such as first run notification
"""
log.info("Get startup notice")
if self.first_run and not self.session.wallet.wallet_balance:
return self._render_response(self.startup_message, OK_CODE)
elif self.first_run:
return self._render_response(None, OK_CODE)
else:
self._render_response(self.startup_message, OK_CODE)
def jsonrpc_version(self):
"""
Get lbry version information
Args:
None
Returns:
"platform": platform string
"os_release": os release string
"os_system": os name
"lbrynet_version: ": lbrynet_version,
"lbryum_version: ": lbryum_version,
"ui_version": commit hash of ui version being used
"remote_lbrynet": most recent lbrynet version available from github
"remote_lbryum": most recent lbryum version available from github
"""
platform_info = self._get_platform()
msg = {
'platform': platform_info['platform'],
'os_release': platform_info['os_release'],
'os_system': platform_info['os_system'],
'lbrynet_version': lbrynet_version,
'lbryum_version': lbryum_version,
'ui_version': self.ui_version,
'remote_lbrynet': self.git_lbrynet_version,
'remote_lbryum': self.git_lbryum_version,
'lbrynet_update_available': utils.version_is_greater_than(self.git_lbrynet_version, lbrynet_version),
'lbryum_update_available': utils.version_is_greater_than(self.git_lbryum_version, lbryum_version),
}
log.info("Get version info: " + json.dumps(msg))
return self._render_response(msg, OK_CODE)
def jsonrpc_get_settings(self):
"""
Get lbrynet daemon settings
Args:
None
Returns:
'run_on_startup': bool,
'data_rate': float,
'max_key_fee': float,
'download_directory': string,
'max_upload': float, 0.0 for unlimited
'max_download': float, 0.0 for unlimited
'upload_log': bool,
'search_timeout': float,
'download_timeout': int
'max_search_results': int,
'wallet_type': string,
'delete_blobs_on_remove': bool,
'peer_port': int,
'dht_node_port': int,
'use_upnp': bool,
'start_lbrycrdd': bool,
"""
log.info("Get daemon settings")
return self._render_response(self.session_settings, OK_CODE)
def jsonrpc_set_settings(self, p):
"""
Set lbrynet daemon settings
Args:
'run_on_startup': bool,
'data_rate': float,
'max_key_fee': float,
'download_directory': string,
'max_upload': float, 0.0 for unlimited
'max_download': float, 0.0 for unlimited
'upload_log': bool,
'download_timeout': int
Returns:
settings dict
"""
def _log_settings_change():
log.info("Set daemon settings to " + json.dumps(self.session_settings))
d = self._update_settings(p)
d.addErrback(lambda err: log.info(err.getTraceback()))
d.addCallback(lambda _: _log_settings_change())
d.addCallback(lambda _: self._render_response(self.session_settings, OK_CODE))
return d
def jsonrpc_help(self, p=None):
"""
Function to retrieve docstring for API function
Args:
optional 'function': function to retrieve documentation for
optional 'callable_during_startup':
Returns:
if given a function, returns given documentation
if given callable_during_startup flag, returns list of functions callable during the startup sequence
if no params are given, returns the list of callable functions
"""
if not p:
return self._render_response(self._listFunctions(), OK_CODE)
elif 'callable_during_start' in p.keys():
return self._render_response(ALLOWED_DURING_STARTUP, OK_CODE)
elif 'function' in p.keys():
func_path = p['function']
function = self._getFunction(func_path)
return self._render_response(function.__doc__, OK_CODE)
else:
return self._render_response(self.jsonrpc_help.__doc__, OK_CODE)
def jsonrpc_get_balance(self):
"""
Get balance
Args:
None
Returns:
balance, float
"""
log.info("Get balance")
return self._render_response(float(self.session.wallet.wallet_balance), OK_CODE)
def jsonrpc_stop(self):
"""
Stop lbrynet-daemon
Args:
None
Returns:
shutdown message
"""
def _disp_shutdown():
log.info("Shutting down lbrynet daemon")
d = self._shutdown()
d.addCallback(lambda _: _disp_shutdown())
d.addCallback(lambda _: reactor.callLater(0.0, reactor.stop))
return self._render_response("Shutting down", OK_CODE)
def jsonrpc_get_lbry_files(self):
"""
Get LBRY files
Args:
None
Returns:
List of lbry files:
'completed': bool
'file_name': string
'key': hex string
'points_paid': float
'stopped': bool
'stream_hash': base 58 string
'stream_name': string
'suggested_file_name': string
'upload_allowed': bool
'sd_hash': string
"""
d = self._get_lbry_files()
d.addCallback(lambda r: [d[1] for d in r])
d.addCallback(lambda r: self._render_response(r, OK_CODE) if len(r) else self._render_response(False, OK_CODE))
return d
def jsonrpc_get_lbry_file(self, p):
"""
Get lbry file
Args:
'name': get file by lbry uri,
'sd_hash': get file by the hash in the name claim,
'file_name': get file by its name in the downloads folder,
Returns:
'completed': bool
'file_name': string
'key': hex string
'points_paid': float
'stopped': bool
'stream_hash': base 58 string
'stream_name': string
'suggested_file_name': string
'upload_allowed': bool
'sd_hash': string
"""
if p.keys()[0] in ['name', 'sd_hash', 'file_name']:
search_type = p.keys()[0]
d = self._get_lbry_file(search_type, p[search_type])
else:
d = defer.fail()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_resolve_name(self, p):
"""
Resolve stream info from a LBRY uri
Args:
'name': name to look up, string, do not include lbry:// prefix
Returns:
metadata from name claim
"""
force = p.get('force', False)
if 'name' in p:
name = p['name']
else:
return self._render_response(None, BAD_REQUEST)
d = self._resolve_name(name, force_refresh=force)
d.addCallbacks(lambda info: self._render_response(info, OK_CODE), lambda _: server.failure)
return d
def jsonrpc_get_my_claim(self, p):
"""
Return existing claim for a given name
Args:
'name': name to look up
Returns:
claim info, False if no such claim exists
"""
name = p['name']
d = self.session.wallet.get_my_claim(name)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_claim_info(self, p):
"""
Resolve claim info from a LBRY uri
Args:
'name': name to look up, string, do not include lbry:// prefix
Returns:
txid, amount, value, n, height
"""
def _convert_amount_to_float(r):
if not r:
return False
else:
r['amount'] = float(r['amount']) / 10**8
return r
name = p['name']
txid = p.get('txid', None)
d = self.session.wallet.get_claim_info(name, txid)
d.addCallback(_convert_amount_to_float)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def _process_get_parameters(self, p):
"""Extract info from input parameters and fill in default values for `get` call."""
# TODO: this process can be abstracted s.t. each method
# can spec what parameters it expects and how to set default values
timeout = p.get('timeout', self.download_timeout)
download_directory = p.get('download_directory', self.download_directory)
file_name = p.get('file_name')
stream_info = p.get('stream_info')
sd_hash = get_sd_hash(stream_info)
wait_for_write = p.get('wait_for_write', True)
name = p.get('name')
return Parameters(
timeout=timeout,
download_directory=download_directory,
file_name=file_name,
stream_info=stream_info,
sd_hash=sd_hash,
wait_for_write=wait_for_write,
name=name
)
def jsonrpc_get(self, p):
"""Download stream from a LBRY uri.
Args:
'name': name to download, string
'download_directory': optional, path to directory where file will be saved, string
'file_name': optional, a user specified name for the downloaded file
'stream_info': optional, specified stream info overrides name
'timeout': optional
'wait_for_write': optional, defaults to True
Returns:
'stream_hash': hex string
'path': path of download
"""
params = self._process_get_parameters(p)
if not params.name:
return server.failure
if params.name in self.waiting_on:
return server.failure
d = self._download_name(name=params.name,
timeout=params.timeout,
download_directory=params.download_directory,
stream_info=params.stream_info,
file_name=params.file_name,
wait_for_write=params.wait_for_write)
# TODO: downloading can timeout. Not sure what to do when that happens
d.addCallbacks(get_output_callback(params), lambda err: str(err))
d.addCallback(lambda message: self._render_response(message, OK_CODE))
return d
def jsonrpc_stop_lbry_file(self, p):
"""
Stop lbry file
Args:
'name': stop file by lbry uri,
'sd_hash': stop file by the hash in the name claim,
'file_name': stop file by its name in the downloads folder,
Returns:
confirmation message
"""
def _stop_file(f):
d = self.lbry_file_manager.toggle_lbry_file_running(f)
d.addCallback(lambda _: "Stopped LBRY file")
return d
if p.keys()[0] in ['name', 'sd_hash', 'file_name']:
search_type = p.keys()[0]
d = self._get_lbry_file(search_type, p[search_type], return_json=False)
d.addCallback(lambda l: _stop_file(l) if not l.stopped else "LBRY file wasn't running")
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_start_lbry_file(self, p):
"""
Stop lbry file
Args:
'name': stop file by lbry uri,
'sd_hash': stop file by the hash in the name claim,
'file_name': stop file by its name in the downloads folder,
Returns:
confirmation message
"""
def _start_file(f):
d = self.lbry_file_manager.toggle_lbry_file_running(f)
return defer.succeed("Started LBRY file")
if p.keys()[0] in ['name', 'sd_hash', 'file_name']:
search_type = p.keys()[0]
d = self._get_lbry_file(search_type, p[search_type], return_json=False)
d.addCallback(lambda l: _start_file(l) if l.stopped else "LBRY file was already running")
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_est_cost(self, p):
"""
Get estimated cost for a lbry uri
Args:
'name': lbry uri
Returns:
estimated cost
"""
name = p['name']
force = p.get('force', False)
if force:
d = self._get_est_cost(name)
else:
d = self._search(name)
d.addCallback(lambda r: [i['cost'] for i in r][0])
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_search_nametrie(self, p):
"""
Search the nametrie for claims
Args:
'search': search query, string
Returns:
List of search results
"""
# TODO: change this function to "search"
if 'search' in p.keys():
search = p['search']
else:
return self._render_response(None, BAD_REQUEST)
# TODO: have ui accept the actual outputs
def _clean(n):
t = []
for i in n:
td = {k: i['value'][k] for k in i['value']}
td['cost_est'] = float(i['cost'])
td['thumbnail'] = i['value'].get('thumbnail', "img/Free-speech-flag.svg")
td['name'] = i['name']
t.append(td)
return t
log.info('Search: %s' % search)
d = self._search(search)
d.addCallback(_clean)
d.addCallback(lambda results: self._render_response(results, OK_CODE))
return d
def jsonrpc_delete_lbry_file(self, p):
"""
Delete a lbry file
Args:
'file_name': downloaded file name, string
Returns:
confirmation message
"""
if 'delete_target_file' in p.keys():
delete_file = p['delete_target_file']
else:
delete_file = True
def _delete_file(f):
file_name = f.file_name
d = self._delete_lbry_file(f, delete_file=delete_file)
d.addCallback(lambda _: "Deleted LBRY file" + file_name)
return d
if 'name' in p.keys() or 'sd_hash' in p.keys() or 'file_name' in p.keys():
search_type = [k for k in p.keys() if k != 'delete_target_file'][0]
d = self._get_lbry_file(search_type, p[search_type], return_json=False)
d.addCallback(lambda l: _delete_file(l) if l else False)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_publish(self, p):
"""
Make a new name claim and publish associated data to lbrynet
Args:
'name': name to be claimed, string
'file_path': path to file to be associated with name, string
'bid': amount of credits to commit in this claim, float
'metadata': metadata dictionary
optional 'fee'
Returns:
Claim txid
"""
def _set_address(address, currency, m):
log.info("Generated new address for key fee: " + str(address))
m['fee'][currency]['address'] = address
return m
def _reflect_if_possible(sd_hash, txid):
d = self._get_lbry_file('sd_hash', sd_hash, return_json=False)
d.addCallback(self._reflect)
d.addCallback(lambda _: txid)
return d
name = p['name']
log.info("Publish: ")
log.info(p)
try:
verify_name_characters(name)
except AssertionError:
log.error("Bad name")
return defer.fail(InvalidNameError("Bad name"))
bid = p['bid']
try:
metadata = Metadata(p['metadata'])
make_lbry_file = False
sd_hash = metadata['sources']['lbry_sd_hash']
except AssertionError:
make_lbry_file = True
sd_hash = None
metadata = p['metadata']
file_path = p['file_path']
if not self.pending_claim_checker.running:
self.pending_claim_checker.start(30)
d = self._resolve_name(name, force_refresh=True)
d.addErrback(lambda _: None)
if 'fee' in p:
metadata['fee'] = p['fee']
assert len(metadata['fee']) == 1, "Too many fees"
for c in metadata['fee']:
if 'address' not in metadata['fee'][c]:
d.addCallback(lambda _: self.session.wallet.get_new_address())
d.addCallback(lambda addr: _set_address(addr, c, metadata))
else:
d.addCallback(lambda _: metadata)
if make_lbry_file:
pub = Publisher(self.session, self.lbry_file_manager, self.session.wallet)
d.addCallback(lambda meta: pub.start(name, file_path, bid, meta))
else:
d.addCallback(lambda meta: self.session.wallet.claim_name(name, bid, meta))
if sd_hash:
d.addCallback(lambda txid: _reflect_if_possible(sd_hash, txid))
d.addCallback(lambda txid: self._add_to_pending_claims(name, txid))
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_abandon_claim(self, p):
"""
Abandon a name and reclaim credits from the claim
Args:
'txid': txid of claim, string
Return:
txid
"""
if 'txid' in p.keys():
txid = p['txid']
else:
return server.failure
def _disp(x):
log.info("Abandoned name claim tx " + str(x))
return self._render_response(x, OK_CODE)
d = defer.Deferred()
d.addCallback(lambda _: self.session.wallet.abandon_name(txid))
d.addCallback(_disp)
d.callback(None)
return d
def jsonrpc_abandon_name(self, p):
"""
DEPRECIATED, use abandon_claim
Args:
'txid': txid of claim, string
Return:
txid
"""
return self.jsonrpc_abandon_claim(p)
def jsonrpc_support_claim(self, p):
"""
Support a name claim
Args:
'name': name
'claim_id': claim id of claim to support
'amount': amount to support by
Return:
txid
"""
name = p['name']
claim_id = p['claim_id']
amount = p['amount']
d = self.session.wallet.support_claim(name, claim_id, amount)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_name_claims(self):
"""
Get my name claims
Args:
None
Returns
list of name claims
"""
def _clean(claims):
for c in claims:
for k in c.keys():
if isinstance(c[k], Decimal):
c[k] = float(c[k])
return defer.succeed(claims)
d = self.session.wallet.get_name_claims()
d.addCallback(_clean)
d.addCallback(lambda claims: self._render_response(claims, OK_CODE))
return d
def jsonrpc_get_claims_for_name(self, p):
"""
Get claims for a name
Args:
'name': name
Returns
list of name claims
"""
name = p['name']
d = self.session.wallet.get_claims_for_name(name)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_transaction_history(self):
"""
Get transaction history
Args:
None
Returns:
list of transactions
"""
d = self.session.wallet.get_history()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_transaction(self, p):
"""
Get a decoded transaction from a txid
Args:
txid: txid hex string
Returns:
JSON formatted transaction
"""
txid = p['txid']
d = self.session.wallet.get_tx_json(txid)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_address_is_mine(self, p):
"""
Checks if an address is associated with the current wallet.
Args:
address: string
Returns:
is_mine: bool
"""
address = p['address']
d = self.session.wallet.address_is_mine(address)
d.addCallback(lambda is_mine: self._render_response(is_mine, OK_CODE))
return d
def jsonrpc_get_public_key_from_wallet(self, p):
"""
Get public key from wallet address
Args:
wallet: wallet address, base58
Returns:
public key
"""
wallet = p['wallet']
d = self.session.wallet.get_pub_keys(wallet)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
def jsonrpc_get_time_behind_blockchain(self):
"""
Get number of blocks behind the blockchain
Args:
None
Returns:
number of blocks behind blockchain, int
"""
def _get_time_behind():
try:
local_height = self.session.wallet.network.get_local_height()
remote_height = self.session.wallet.network.get_server_height()
return defer.succeed(remote_height - local_height)
except:
return defer.fail()
d = _get_time_behind()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_new_address(self):
"""
Generate a new wallet address
Args:
None
Returns:
new wallet address, base 58 string
"""
def _disp(address):
log.info("Got new wallet address: " + address)
return defer.succeed(address)
d = self.session.wallet.get_new_address()
d.addCallback(_disp)
d.addCallback(lambda address: self._render_response(address, OK_CODE))
return d
def jsonrpc_send_amount_to_address(self, p):
"""
Send credits to an address
Args:
amount: the amount to send
address: the address of the recipient
Returns:
True if payment successfully scheduled
"""
if 'amount' in p.keys() and 'address' in p.keys():
amount = p['amount']
address = p['address']
else:
return server.failure
reserved_points = self.session.wallet.reserve_points(address, amount)
if reserved_points is None:
return defer.fail(InsufficientFundsError())
d = self.session.wallet.send_points_to_address(reserved_points, amount)
d.addCallback(lambda _: self._render_response(True, OK_CODE))
return d
def jsonrpc_get_best_blockhash(self):
"""
Get hash of most recent block
Args:
None
Returns:
Hash of most recent block
"""
d = self.session.wallet.get_best_blockhash()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_block(self, p):
"""
Get contents of a block
Args:
blockhash: hash of the block to look up
Returns:
requested block
"""
if 'blockhash' in p.keys():
blockhash = p['blockhash']
d = self.session.wallet.get_block(blockhash)
elif 'height' in p.keys():
height = p['height']
d = self.session.wallet.get_block_info(height)
d.addCallback(lambda blockhash: self.session.wallet.get_block(blockhash))
else:
return server.failure
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_claims_for_tx(self, p):
"""
Get claims for tx
Args:
txid: txid of a name claim transaction
Returns:
any claims contained in the requested tx
"""
if 'txid' in p.keys():
txid = p['txid']
else:
return server.failure
d = self.session.wallet.get_claims_from_tx(txid)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_download_descriptor(self, p):
"""
Download and return a sd blob
Args:
sd_hash
Returns
sd blob, dict
"""
sd_hash = p['sd_hash']
timeout = p.get('timeout', DEFAULT_SD_DOWNLOAD_TIMEOUT)
d = self._download_sd_blob(sd_hash, timeout)
d.addCallbacks(lambda r: self._render_response(r, OK_CODE), lambda _: self._render_response(False, OK_CODE))
return d
def jsonrpc_get_nametrie(self):
"""
Get the nametrie
Args:
None
Returns:
Name claim trie
"""
d = self.session.wallet.get_nametrie()
d.addCallback(lambda r: [i for i in r if 'txid' in i.keys()])
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_set_miner(self, p):
"""
Start of stop the miner, function only available when lbrycrd is set as the wallet
Args:
run: True/False
Returns:
miner status, True/False
"""
stat = p['run']
if stat:
d = self.session.wallet.start_miner()
else:
d = self.session.wallet.stop_miner()
d.addCallback(lambda _: self.session.wallet.get_miner_status())
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_miner_status(self):
"""
Get status of miner
Args:
None
Returns:
True/False
"""
d = self.session.wallet.get_miner_status()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_log(self, p):
"""
Log message
Args:
'message': message to be logged
Returns:
True
"""
message = p['message']
log.info("API client log request: %s" % message)
return self._render_response(True, OK_CODE)
def jsonrpc_upload_log(self, p=None):
"""
Upload log
Args, optional:
'name_prefix': prefix to indicate what is requesting the log upload
'exclude_previous': true/false, whether or not to exclude previous sessions from upload, defaults on true
Returns:
True
"""
if p:
if 'name_prefix' in p.keys():
log_type = p['name_prefix'] + '_api'
elif 'log_type' in p.keys():
log_type = p['log_type'] + '_api'
else:
log_type = None
if 'exclude_previous' in p.keys():
exclude_previous = p['exclude_previous']
else:
exclude_previous = True
if 'message' in p.keys():
log.info("Upload log message: " + str(p['message']))
if 'force' in p.keys():
force = p['force']
else:
force = False
else:
log_type = "api"
exclude_previous = True
d = self._upload_log(log_type=log_type, exclude_previous=exclude_previous, force=force)
if 'message' in p.keys():
d.addCallback(lambda _: self._log_to_slack(p['message']))
d.addCallback(lambda _: self._render_response(True, OK_CODE))
return d
def jsonrpc_configure_ui(self, p):
"""
Configure the UI being hosted
Args, optional:
'branch': a branch name on lbryio/lbry-web-ui
'path': path to a ui folder
"""
if 'check_requirements' in p:
check_require = p['check_requirements']
else:
check_require = True
if 'path' in p:
d = self.lbry_ui_manager.setup(user_specified=p['path'], check_requirements=check_require)
elif 'branch' in p:
d = self.lbry_ui_manager.setup(branch=p['branch'], check_requirements=check_require)
else:
d = self.lbry_ui_manager.setup(check_requirements=check_require)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_reveal(self, p):
"""
Reveal a file or directory in file browser
Args:
'path': path to be selected in file browser
Returns:
True, opens file browser
"""
path = p['path']
if sys.platform == "darwin":
d = threads.deferToThread(subprocess.Popen, ['open', '-R', path])
else:
# No easy way to reveal specific files on Linux, so just open the containing directory
d = threads.deferToThread(subprocess.Popen, ['xdg-open', os.path.dirname(path)])
d.addCallback(lambda _: self._render_response(True, OK_CODE))
return d
def jsonrpc_get_peers_for_hash(self, p):
"""
Get peers for blob hash
Args:
'blob_hash': blob hash
Returns:
List of contacts
"""
blob_hash = p['blob_hash']
d = self.session.peer_finder.find_peers_for_blob(blob_hash)
d.addCallback(lambda r: [[c.host, c.port, c.is_available()] for c in r])
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_announce_all_blobs_to_dht(self):
"""
Announce all blobs to the dht
Args:
None
Returns:
"""
d = self.session.blob_manager.immediate_announce_all_blobs()
d.addCallback(lambda _: self._render_response("Announced", OK_CODE))
return d
def jsonrpc_reflect(self, p):
"""
Reflect a stream
Args:
sd_hash: sd_hash of lbry file
Returns:
True or traceback
"""
sd_hash = p['sd_hash']
d = self._get_lbry_file('sd_hash', sd_hash, return_json=False)
d.addCallback(self._reflect)
d.addCallbacks(lambda _: self._render_response(True, OK_CODE), lambda err: self._render_response(err.getTraceback(), OK_CODE))
return d
def jsonrpc_get_blob_hashes(self):
"""
Returns all blob hashes
Args:
None
Returns:
list of blob hashes
"""
d = self.session.blob_manager.get_all_verified_blobs()
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_reflect_all_blobs(self):
"""
Reflects all saved blobs
Args:
None
Returns:
True
"""
d = self.session.blob_manager.get_all_verified_blobs()
d.addCallback(self._reflect_blobs)
d.addCallback(lambda r: self._render_response(r, OK_CODE))
return d
def jsonrpc_get_search_servers(self):
"""
Get list of lighthouse servers
Args:
None
Returns:
List of address:port
"""
d = self._render_response(SEARCH_SERVERS, OK_CODE)
return d
def get_lbrynet_version_from_github():
"""Return the latest released version from github."""
response = requests.get('https://api.github.com/repos/lbryio/lbry/releases/latest')
release = response.json()
tag = release['tag_name']
# githubs documentation claims this should never happen, but we'll check just in case
if release['prerelease']:
raise Exception('Release {} is a pre-release'.format(tag))
return get_version_from_tag(tag)
def get_version_from_tag(tag):
match = re.match('v([\d.]+)', tag)
if match:
return match.group(1)
else:
raise Exception('Failed to parse version from tag {}'.format(tag))
def get_sd_hash(stream_info):
if not stream_info:
return None
try:
return stream_info['sources']['lbry_sd_hash']
except KeyError:
return stream_info.get('stream_hash')
def get_output_callback(params):
def callback(l):
return {
'stream_hash': params.sd_hash if params.stream_info else l.sd_hash,
'path': os.path.join(params.download_directory, l.file_name)
}
return callback
def get_darwin_lbrycrdd_path():
# use the path from the bundle if its available.
default = "./lbrycrdd"
try:
import Foundation
except ImportError:
log.warning('Foundation module not installed, falling back to default lbrycrdd path')
return default
else:
try:
bundle = Foundation.NSBundle.mainBundle()
return bundle.pathForResource_ofType_('lbrycrdd', None)
except Exception:
log.exception('Failed to get path from bundle, falling back to default')
return default
class _DownloadNameHelper(object):
def __init__(self, daemon, name, timeout=DEFAULT_TIMEOUT, download_directory=None,
file_name=None, wait_for_write=True):
self.daemon = daemon
self.name = name
self.timeout = timeout
if not download_directory or not os.path.isdir(download_directory):
self.download_directory = daemon.download_directory
else:
self.download_directory = download_directory
self.file_name = file_name
self.wait_for_write = wait_for_write
def _setup_stream(self, stream_info):
stream_hash = get_sd_hash(stream_info)
d = self.daemon._get_lbry_file_by_sd_hash(stream_hash)
d.addCallback(self._add_results_callback(stream_info))
return d
def _add_results_callback(self, stream_info):
def add_results(l):
if l:
if os.path.isfile(os.path.join(self.download_directory, l.file_name)):
return defer.succeed((stream_info, l))
return defer.succeed((stream_info, None))
return add_results
def wait_or_get_stream(self, args):
stream_info, lbry_file = args
if lbry_file:
log.debug('Wait on lbry_file')
return self._wait_on_lbry_file(lbry_file)
else:
log.debug('No lbry_file, need to get stream')
return self._get_stream(stream_info)
def _get_stream(self, stream_info):
d = self.daemon.add_stream(
self.name, self.timeout, self.download_directory, self.file_name, stream_info)
def _handle_timeout(args):
was_successful, _, _ = args
if not was_successful:
log.warning("lbry://%s timed out, removing from streams", self.name)
del self.daemon.streams[self.name]
d.addCallback(_handle_timeout)
if self.wait_for_write:
d.addCallback(lambda _: self._wait_for_write())
def _get_stream_for_return():
stream = self.daemon.streams.get(self.name, None)
if stream:
return stream.downloader
else:
self._remove_from_wait("Timed out")
return defer.fail(Exception("Timed out"))
d.addCallback(lambda _: _get_stream_for_return())
return d
def _wait_for_write(self):
d = defer.succeed(None)
if not self.has_downloader_wrote():
d.addCallback(lambda _: reactor.callLater(1, self._wait_for_write))
return d
def has_downloader_wrote(self):
stream = self.daemon.streams.get(self.name, False)
if stream:
downloader = stream.downloader
else:
downloader = False
if not downloader:
return False
return self.get_written_bytes(downloader.file_name)
def _wait_on_lbry_file(self, f):
written_bytes = self.get_written_bytes(f.file_name)
if written_bytes:
return defer.succeed(self._disp_file(f))
return task.deferLater(reactor, 1, self._wait_on_lbry_file, f)
def get_written_bytes(self, file_name):
"""Returns the number of bytes written to `file_name`.
Returns False if there were issues reading `file_name`.
"""
try:
file_path = os.path.join(self.download_directory, file_name)
if os.path.isfile(file_path):
written_file = file(file_path)
written_file.seek(0, os.SEEK_END)
written_bytes = written_file.tell()
written_file.close()
else:
written_bytes = False
except Exception:
writen_bytes = False
return written_bytes
def _disp_file(self, f):
file_path = os.path.join(self.download_directory, f.file_name)
log.info("Already downloaded: %s --> %s", f.sd_hash, file_path)
return f
def _remove_from_wait(self, r):
if self.name in self.daemon.waiting_on:
del self.daemon.waiting_on[self.name]
return r
class _ResolveNameHelper(object):
def __init__(self, daemon, name, force_refresh):
self.daemon = daemon
self.name = name
self.force_refresh = force_refresh
def get_deferred(self):
if self.need_fresh_stream():
log.info("Resolving stream info for lbry://%s", self.name)
d = self.wallet.get_stream_info_for_name(self.name)
d.addCallbacks(self._cache_stream_info, lambda _: defer.fail(UnknownNameError))
else:
log.debug("Returning cached stream info for lbry://%s", self.name)
d = defer.succeed(self.name_data['claim_metadata'])
return d
@property
def name_data(self):
return self.daemon.name_cache[self.name]
@property
def wallet(self):
return self.daemon.session.wallet
def now(self):
return self.daemon._get_long_count_timestamp()
def _add_txid(self, txid):
self.name_data['txid'] = txid
return defer.succeed(None)
def _cache_stream_info(self, stream_info):
self.daemon.name_cache[self.name] = {
'claim_metadata': stream_info,
'timestamp': self.now()
}
d = self.wallet.get_txid_for_name(self.name)
d.addCallback(self._add_txid)
d.addCallback(lambda _: self.daemon._update_claim_cache())
d.addCallback(lambda _: self.name_data['claim_metadata'])
return d
def need_fresh_stream(self):
return self.force_refresh or not self.is_in_cache() or self.is_cached_name_expired()
def is_in_cache(self):
return self.name in self.daemon.name_cache
def is_cached_name_expired(self):
time_in_cache = self.now() - self.name_data['timestamp']
return time_in_cache >= self.daemon.cache_time