3 I contain the client-side code which speaks to storage servers, in particular
4 the foolscap-based server implemented in src/allmydata/storage/*.py .
9 # 1: implement StorageFarmBroker (i.e. "storage broker"), change Client to
10 # create it, change uploader/servermap to get rrefs from it. ServerFarm calls
11 # IntroducerClient.subscribe_to . ServerFarm hides descriptors, passes rrefs
12 # to clients. webapi status pages call broker.get_info_about_serverid.
14 # 2: move get_info methods to the descriptor, webapi status pages call
15 # broker.get_descriptor_for_serverid().get_info
17 # 3?later?: store descriptors in UploadResults/etc instead of serverids,
18 # webapi status pages call descriptor.get_info and don't use storage_broker
21 # 4: enable static config: tahoe.cfg can add descriptors. Make the introducer
22 # optional. This closes #467
24 # 5: implement NativeStorageClient, pass it to Tahoe2PeerSelector and other
25 # clients. Clients stop doing callRemote(), use NativeStorageClient methods
26 # instead (which might do something else, i.e. http or whatever). The
27 # introducer and tahoe.cfg only create NativeStorageClients for now.
29 # 6: implement other sorts of IStorageClient classes: S3, etc
33 from zope.interface import implements, Interface
34 from foolscap.api import eventually
35 from allmydata.interfaces import IStorageBroker
36 from allmydata.util import log, base32
37 from allmydata.util.assertutil import precondition
38 from allmydata.util.rrefutil import add_version_to_remote_reference
39 from allmydata.util.hashutil import sha1
41 # who is responsible for de-duplication?
43 # IC remembers the unpacked announcements it receives, to provide for late
44 # subscribers and to remove duplicates
46 # if a client subscribes after startup, will they receive old announcements?
49 # who will be responsible for signature checking?
50 # make it be IntroducerClient, so they can push the filter outwards and
51 # reduce inbound network traffic
53 # what should the interface between StorageFarmBroker and IntroducerClient
55 # don't pass signatures: only pass validated blessed-objects
57 class StorageFarmBroker:
58 implements(IStorageBroker)
59 """I live on the client, and know about storage servers. For each server
60 that is participating in a grid, I either maintain a connection to it or
61 remember enough information to establish a connection to it on demand.
62 I'm also responsible for subscribing to the IntroducerClient to find out
63 about new servers as they are announced by the Introducer.
65 def __init__(self, tub, permute_peers):
67 assert permute_peers # False not implemented yet
68 self.permute_peers = permute_peers
69 # self.servers maps serverid -> IServer, and keeps track of all the
70 # storage servers that we've heard about. Each descriptor manages its
71 # own Reconnector, and will give us a RemoteReference when we ask
74 self.introducer_client = None
76 # these two are used in unit tests
77 def test_add_rref(self, serverid, rref, ann):
78 s = NativeStorageServer(serverid, ann.copy())
80 self.servers[serverid] = s
82 def test_add_server(self, serverid, s):
83 self.servers[serverid] = s
85 def use_introducer(self, introducer_client):
86 self.introducer_client = ic = introducer_client
87 ic.subscribe_to("storage", self._got_announcement)
89 def _got_announcement(self, key_s, ann):
91 precondition(isinstance(key_s, str), key_s)
92 precondition(key_s.startswith("v0-"), key_s)
93 assert ann["service-name"] == "storage"
94 s = NativeStorageServer(key_s, ann)
95 serverid = s.get_serverid()
96 old = self.servers.get(serverid)
98 if old.get_announcement() == ann:
101 del self.servers[serverid]
102 old.stop_connecting()
103 # now we forget about them and start using the new one
104 self.servers[serverid] = s
105 s.start_connecting(self.tub, self._trigger_connections)
106 # the descriptor will manage their own Reconnector, and each time we
107 # need servers, we'll ask them if they're connected or not.
109 def _trigger_connections(self):
110 # when one connection is established, reset the timers on all others,
111 # to trigger a reconnection attempt in one second. This is intended
112 # to accelerate server connections when we've been offline for a
113 # while. The goal is to avoid hanging out for a long time with
114 # connections to only a subset of the servers, which would increase
115 # the chances that we'll put shares in weird places (and not update
116 # existing shares of mutable files). See #374 for more details.
117 for dsc in self.servers.values():
120 def get_servers_for_psi(self, peer_selection_index):
121 # return a list of server objects (IServers)
122 assert self.permute_peers == True
123 def _permuted(server):
124 seed = server.get_permutation_seed()
125 return sha1(peer_selection_index + seed).digest()
126 return sorted(self.get_connected_servers(), key=_permuted)
128 def get_all_serverids(self):
129 return frozenset(self.servers.keys())
131 def get_connected_servers(self):
132 return frozenset([s for s in self.servers.values() if s.get_rref()])
134 def get_known_servers(self):
135 return frozenset(self.servers.values())
137 def get_nickname_for_serverid(self, serverid):
138 if serverid in self.servers:
139 return self.servers[serverid].get_nickname()
142 class IServer(Interface):
143 """I live in the client, and represent a single server."""
144 def start_connecting(tub, trigger_cb):
151 class NativeStorageServer:
152 """I hold information about a storage server that we want to connect to.
153 If we are connected, I hold the RemoteReference, their host address, and
154 the their version information. I remember information about when we were
155 last connected too, even if we aren't currently connected.
157 @ivar announcement_time: when we first heard about this service
158 @ivar last_connect_time: when we last established a connection
159 @ivar last_loss_time: when we last lost a connection
161 @ivar version: the server's versiondict, from the most recent announcement
162 @ivar nickname: the server's self-reported nickname (unicode), same
164 @ivar rref: the RemoteReference, if connected, otherwise None
165 @ivar remote_host: the IAddress, if connected, otherwise None
170 "http://allmydata.org/tahoe/protocols/storage/v1" :
171 { "maximum-immutable-share-size": 2**32,
172 "tolerates-immutable-read-overrun": False,
173 "delete-mutable-shares-with-zero-length-writev": False,
175 "application-version": "unknown: no get_version()",
178 def __init__(self, key_s, ann, min_shares=1):
180 self.announcement = ann
181 self.min_shares = min_shares
183 assert "anonymous-storage-FURL" in ann, ann
184 furl = str(ann["anonymous-storage-FURL"])
185 m = re.match(r'pb://(\w+)@', furl)
187 tubid_s = m.group(1).lower()
188 self._tubid = base32.a2b(tubid_s)
189 assert "permutation-seed-base32" in ann, ann
190 ps = base32.a2b(str(ann["permutation-seed-base32"]))
191 self._permutation_seed = ps
193 name = key_s or tubid_s
194 self._long_description = name
195 self._short_description = name[:8] # TODO: decide who adds []
197 self.announcement_time = time.time()
198 self.last_connect_time = None
199 self.last_loss_time = None
200 self.remote_host = None
202 self._reconnector = None
203 self._trigger_cb = None
206 return "<NativeStorageServer for %s>" % self.get_name()
207 def get_serverid(self):
208 return self._tubid # XXX replace with self.key_s
209 def get_permutation_seed(self):
210 return self._permutation_seed
211 def get_version(self):
213 return self.rref.version
215 def get_name(self): # keep methodname short
216 return self._short_description
217 def get_longname(self):
218 return self._long_description
219 def get_lease_seed(self):
221 def get_foolscap_write_enabler_seed(self):
224 def get_nickname(self):
225 return self.announcement["nickname"].decode("utf-8")
226 def get_announcement(self):
227 return self.announcement
228 def get_remote_host(self):
229 return self.remote_host
230 def get_last_connect_time(self):
231 return self.last_connect_time
232 def get_last_loss_time(self):
233 return self.last_loss_time
234 def get_announcement_time(self):
235 return self.announcement_time
237 def start_connecting(self, tub, trigger_cb):
238 furl = str(self.announcement["anonymous-storage-FURL"])
239 self._trigger_cb = trigger_cb
240 self._reconnector = tub.connectTo(furl, self._got_connection)
242 def _got_connection(self, rref):
243 lp = log.msg(format="got connection to %(name)s, getting versions",
244 name=self.get_name(),
245 facility="tahoe.storage_broker", umid="coUECQ")
247 eventually(self._trigger_cb)
248 default = self.VERSION_DEFAULTS
249 d = add_version_to_remote_reference(rref, default)
250 d.addCallback(self._got_versioned_service, lp)
251 d.addErrback(log.err, format="storageclient._got_connection",
252 name=self.get_name(), umid="Sdq3pg")
254 def _got_versioned_service(self, rref, lp):
255 log.msg(format="%(name)s provided version info %(version)s",
256 name=self.get_name(), version=rref.version,
257 facility="tahoe.storage_broker", umid="SWmJYg",
258 level=log.NOISY, parent=lp)
260 self.last_connect_time = time.time()
261 self.remote_host = rref.getPeer()
263 rref.notifyOnDisconnect(self._lost)
269 log.msg(format="lost connection to %(name)s", name=self.get_name(),
270 facility="tahoe.storage_broker", umid="zbRllw")
271 self.last_loss_time = time.time()
273 self.remote_host = None
275 def stop_connecting(self):
276 # used when this descriptor has been superceded by another
277 self._reconnector.stopConnecting()
279 def try_to_connect(self):
280 # used when the broker wants us to hurry up
281 self._reconnector.reset()
283 class UnknownServerTypeError(Exception):