1 import os, re, weakref, struct, time
3 from foolscap import Referenceable
4 from twisted.application import service
6 from zope.interface import implements
7 from allmydata.interfaces import RIStorageServer, IStatsProducer
8 from allmydata.util import base32, fileutil, log, time_format
9 import allmydata # for __full_version__
11 from allmydata.storage.common import si_b2a, si_a2b, storage_index_to_dir
12 _pyflakes_hush = [si_b2a, si_a2b, storage_index_to_dir] # re-exported
13 from allmydata.storage.lease import LeaseInfo
14 from allmydata.storage.mutable import MutableShareFile, EmptyShare, \
15 create_mutable_sharefile
16 from allmydata.storage.immutable import ShareFile, BucketWriter, BucketReader
17 from allmydata.storage.crawler import BucketCountingCrawler
20 # storage/shares/incoming
21 # incoming/ holds temp dirs named $START/$STORAGEINDEX/$SHARENUM which will
22 # be moved to storage/shares/$START/$STORAGEINDEX/$SHARENUM upon success
23 # storage/shares/$START/$STORAGEINDEX
24 # storage/shares/$START/$STORAGEINDEX/$SHARENUM
26 # Where "$START" denotes the first 10 bits worth of $STORAGEINDEX (that's 2
29 # $SHARENUM matches this regex:
30 NUM_RE=re.compile("^[0-9]+$")
34 class StorageServer(service.MultiService, Referenceable):
35 implements(RIStorageServer, IStatsProducer)
38 def __init__(self, storedir, nodeid, reserved_space=0,
39 discard_storage=False, readonly_storage=False,
41 service.MultiService.__init__(self)
42 assert isinstance(nodeid, str)
43 assert len(nodeid) == 20
44 self.my_nodeid = nodeid
45 self.storedir = storedir
46 sharedir = os.path.join(storedir, "shares")
47 fileutil.make_dirs(sharedir)
48 self.sharedir = sharedir
49 # we don't actually create the corruption-advisory dir until necessary
50 self.corruption_advisory_dir = os.path.join(storedir,
51 "corruption-advisories")
52 self.reserved_space = int(reserved_space)
53 self.no_storage = discard_storage
54 self.readonly_storage = readonly_storage
55 self.stats_provider = stats_provider
56 if self.stats_provider:
57 self.stats_provider.register_producer(self)
58 self.incomingdir = os.path.join(sharedir, 'incoming')
59 self._clean_incomplete()
60 fileutil.make_dirs(self.incomingdir)
61 self._active_writers = weakref.WeakKeyDictionary()
62 lp = log.msg("StorageServer created", facility="tahoe.storage")
65 if self.get_available_space() is None:
66 log.msg("warning: [storage]reserved_space= is set, but this platform does not support statvfs(2), so this reservation cannot be honored",
67 umin="0wZ27w", level=log.UNUSUAL)
69 self.latencies = {"allocate": [], # immutable
74 "writev": [], # mutable
76 "add-lease": [], # both
80 self.add_bucket_counter()
82 def add_bucket_counter(self):
83 statefile = os.path.join(self.storedir, "bucket_counter.state")
84 self.bucket_counter = BucketCountingCrawler(self, statefile)
85 self.bucket_counter.setServiceParent(self)
87 def count(self, name, delta=1):
88 if self.stats_provider:
89 self.stats_provider.count("storage_server." + name, delta)
91 def add_latency(self, category, latency):
92 a = self.latencies[category]
95 self.latencies[category] = a[-1000:]
97 def get_latencies(self):
98 """Return a dict, indexed by category, that contains a dict of
99 latency numbers for each category. Each dict will contain the
100 following keys: mean, 01_0_percentile, 10_0_percentile,
101 50_0_percentile (median), 90_0_percentile, 95_0_percentile,
102 99_0_percentile, 99_9_percentile. If no samples have been collected
103 for the given category, then that category name will not be present
104 in the return value."""
105 # note that Amazon's Dynamo paper says they use 99.9% percentile.
107 for category in self.latencies:
108 if not self.latencies[category]:
111 samples = self.latencies[category][:]
114 stats["mean"] = sum(samples) / count
115 stats["01_0_percentile"] = samples[int(0.01 * count)]
116 stats["10_0_percentile"] = samples[int(0.1 * count)]
117 stats["50_0_percentile"] = samples[int(0.5 * count)]
118 stats["90_0_percentile"] = samples[int(0.9 * count)]
119 stats["95_0_percentile"] = samples[int(0.95 * count)]
120 stats["99_0_percentile"] = samples[int(0.99 * count)]
121 stats["99_9_percentile"] = samples[int(0.999 * count)]
122 output[category] = stats
125 def log(self, *args, **kwargs):
126 if "facility" not in kwargs:
127 kwargs["facility"] = "tahoe.storage"
128 return log.msg(*args, **kwargs)
130 def _clean_incomplete(self):
131 fileutil.rm_dir(self.incomingdir)
133 def do_statvfs(self):
134 return os.statvfs(self.storedir)
137 # remember: RIStatsProvider requires that our return dict
138 # contains numeric values.
139 stats = { 'storage_server.allocated': self.allocated_size(), }
140 stats["storage_server.reserved_space"] = self.reserved_space
141 for category,ld in self.get_latencies().items():
142 for name,v in ld.items():
143 stats['storage_server.latencies.%s.%s' % (category, name)] = v
145 if self.readonly_storage:
148 s = self.do_statvfs()
149 disk_total = s.f_bsize * s.f_blocks
150 disk_used = s.f_bsize * (s.f_blocks - s.f_bfree)
151 # spacetime predictors should look at the slope of disk_used.
152 disk_free_for_root = s.f_bsize * s.f_bfree
153 disk_free_for_nonroot = s.f_bsize * s.f_bavail
155 # include our local policy here: if we stop accepting shares when
156 # the available space drops below 1GB, then include that fact in
158 disk_avail = disk_free_for_nonroot - self.reserved_space
159 disk_avail = max(disk_avail, 0)
160 if self.readonly_storage:
165 # spacetime predictors should use disk_avail / (d(disk_used)/dt)
166 stats["storage_server.disk_total"] = disk_total
167 stats["storage_server.disk_used"] = disk_used
168 stats["storage_server.disk_free_for_root"] = disk_free_for_root
169 stats["storage_server.disk_free_for_nonroot"] = disk_free_for_nonroot
170 stats["storage_server.disk_avail"] = disk_avail
171 except AttributeError:
172 # os.statvfs is available only on unix
174 stats["storage_server.accepting_immutable_shares"] = int(writeable)
175 s = self.bucket_counter.get_state()
176 bucket_count = s.get("last-complete-bucket-count")
178 stats["storage_server.total_bucket_count"] = bucket_count
182 def stat_disk(self, d):
184 # s.f_bavail: available to non-root users
185 disk_avail = s.f_bsize * s.f_bavail
188 def get_available_space(self):
189 # returns None if it cannot be measured (windows)
191 disk_avail = self.stat_disk(self.storedir)
192 disk_avail -= self.reserved_space
193 except AttributeError:
195 if self.readonly_storage:
199 def allocated_size(self):
201 for bw in self._active_writers:
202 space += bw.allocated_size()
205 def remote_get_version(self):
206 remaining_space = self.get_available_space()
207 if remaining_space is None:
208 # we're on a platform that doesn't have 'df', so make a vague
210 remaining_space = 2**64
211 version = { "http://allmydata.org/tahoe/protocols/storage/v1" :
212 { "maximum-immutable-share-size": remaining_space,
213 "tolerates-immutable-read-overrun": True,
214 "delete-mutable-shares-with-zero-length-writev": True,
216 "application-version": str(allmydata.__full_version__),
220 def remote_allocate_buckets(self, storage_index,
221 renew_secret, cancel_secret,
222 sharenums, allocated_size,
223 canary, owner_num=0):
224 # owner_num is not for clients to set, but rather it should be
225 # curried into the PersonalStorageServer instance that is dedicated
226 # to a particular owner.
228 self.count("allocate")
230 bucketwriters = {} # k: shnum, v: BucketWriter
231 si_dir = storage_index_to_dir(storage_index)
232 si_s = si_b2a(storage_index)
234 log.msg("storage: allocate_buckets %s" % si_s)
236 # in this implementation, the lease information (including secrets)
237 # goes into the share files themselves. It could also be put into a
238 # separate database. Note that the lease should not be added until
239 # the BucketWriter has been closed.
240 expire_time = time.time() + 31*24*60*60
241 lease_info = LeaseInfo(owner_num,
242 renew_secret, cancel_secret,
243 expire_time, self.my_nodeid)
245 max_space_per_bucket = allocated_size
247 remaining_space = self.get_available_space()
248 limited = remaining_space is not None
250 # this is a bit conservative, since some of this allocated_size()
251 # has already been written to disk, where it will show up in
252 # get_available_space.
253 remaining_space -= self.allocated_size()
255 # fill alreadygot with all shares that we have, not just the ones
256 # they asked about: this will save them a lot of work. Add or update
257 # leases for all of them: if they want us to hold shares for this
258 # file, they'll want us to hold leases for this file.
259 for (shnum, fn) in self._get_bucket_shares(storage_index):
260 alreadygot.add(shnum)
262 sf.add_or_renew_lease(lease_info)
264 # self.readonly_storage causes remaining_space=0
266 for shnum in sharenums:
267 incominghome = os.path.join(self.incomingdir, si_dir, "%d" % shnum)
268 finalhome = os.path.join(self.sharedir, si_dir, "%d" % shnum)
269 if os.path.exists(finalhome):
270 # great! we already have it. easy.
272 elif os.path.exists(incominghome):
273 # Note that we don't create BucketWriters for shnums that
274 # have a partial share (in incoming/), so if a second upload
275 # occurs while the first is still in progress, the second
276 # uploader will use different storage servers.
278 elif (not limited) or (remaining_space >= max_space_per_bucket):
279 # ok! we need to create the new share file.
280 bw = BucketWriter(self, incominghome, finalhome,
281 max_space_per_bucket, lease_info, canary)
283 bw.throw_out_all_data = True
284 bucketwriters[shnum] = bw
285 self._active_writers[bw] = 1
287 remaining_space -= max_space_per_bucket
289 # bummer! not enough space to accept this bucket
293 fileutil.make_dirs(os.path.join(self.sharedir, si_dir))
295 self.add_latency("allocate", time.time() - start)
296 return alreadygot, bucketwriters
298 def _iter_share_files(self, storage_index):
299 for shnum, filename in self._get_bucket_shares(storage_index):
300 f = open(filename, 'rb')
303 if header[:32] == MutableShareFile.MAGIC:
304 sf = MutableShareFile(filename, self)
305 # note: if the share has been migrated, the renew_lease()
306 # call will throw an exception, with information to help the
307 # client update the lease.
308 elif header[:4] == struct.pack(">L", 1):
309 sf = ShareFile(filename)
311 continue # non-sharefile
314 def remote_add_lease(self, storage_index, renew_secret, cancel_secret,
317 self.count("add-lease")
318 new_expire_time = time.time() + 31*24*60*60
319 lease_info = LeaseInfo(owner_num,
320 renew_secret, cancel_secret,
321 new_expire_time, self.my_nodeid)
322 for sf in self._iter_share_files(storage_index):
323 sf.add_or_renew_lease(lease_info)
324 self.add_latency("add-lease", time.time() - start)
327 def remote_renew_lease(self, storage_index, renew_secret):
330 new_expire_time = time.time() + 31*24*60*60
331 found_buckets = False
332 for sf in self._iter_share_files(storage_index):
334 sf.renew_lease(renew_secret, new_expire_time)
335 self.add_latency("renew", time.time() - start)
336 if not found_buckets:
337 raise IndexError("no such lease to renew")
339 def remote_cancel_lease(self, storage_index, cancel_secret):
343 total_space_freed = 0
344 found_buckets = False
345 for sf in self._iter_share_files(storage_index):
346 # note: if we can't find a lease on one share, we won't bother
347 # looking in the others. Unless something broke internally
348 # (perhaps we ran out of disk space while adding a lease), the
349 # leases on all shares will be identical.
351 # this raises IndexError if the lease wasn't present XXXX
352 total_space_freed += sf.cancel_lease(cancel_secret)
355 storagedir = os.path.join(self.sharedir,
356 storage_index_to_dir(storage_index))
357 if not os.listdir(storagedir):
360 if self.stats_provider:
361 self.stats_provider.count('storage_server.bytes_freed',
363 self.add_latency("cancel", time.time() - start)
364 if not found_buckets:
365 raise IndexError("no such storage index")
367 def bucket_writer_closed(self, bw, consumed_size):
368 if self.stats_provider:
369 self.stats_provider.count('storage_server.bytes_added', consumed_size)
370 del self._active_writers[bw]
372 def _get_bucket_shares(self, storage_index):
373 """Return a list of (shnum, pathname) tuples for files that hold
374 shares for this storage_index. In each tuple, 'shnum' will always be
375 the integer form of the last component of 'pathname'."""
376 storagedir = os.path.join(self.sharedir, storage_index_to_dir(storage_index))
378 for f in os.listdir(storagedir):
380 filename = os.path.join(storagedir, f)
381 yield (int(f), filename)
383 # Commonly caused by there being no buckets at all.
386 def remote_get_buckets(self, storage_index):
389 si_s = si_b2a(storage_index)
390 log.msg("storage: get_buckets %s" % si_s)
391 bucketreaders = {} # k: sharenum, v: BucketReader
392 for shnum, filename in self._get_bucket_shares(storage_index):
393 bucketreaders[shnum] = BucketReader(self, filename,
394 storage_index, shnum)
395 self.add_latency("get", time.time() - start)
398 def get_leases(self, storage_index):
399 """Provide an iterator that yields all of the leases attached to this
400 bucket. Each lease is returned as a tuple of (owner_num,
401 renew_secret, cancel_secret, expiration_time).
403 This method is not for client use.
406 # since all shares get the same lease data, we just grab the leases
407 # from the first share
409 shnum, filename = self._get_bucket_shares(storage_index).next()
410 sf = ShareFile(filename)
411 return sf.iter_leases()
412 except StopIteration:
415 def remote_slot_testv_and_readv_and_writev(self, storage_index,
417 test_and_write_vectors,
421 si_s = si_b2a(storage_index)
422 lp = log.msg("storage: slot_writev %s" % si_s)
423 si_dir = storage_index_to_dir(storage_index)
424 (write_enabler, renew_secret, cancel_secret) = secrets
425 # shares exist if there is a file for them
426 bucketdir = os.path.join(self.sharedir, si_dir)
428 if os.path.isdir(bucketdir):
429 for sharenum_s in os.listdir(bucketdir):
431 sharenum = int(sharenum_s)
434 filename = os.path.join(bucketdir, sharenum_s)
435 msf = MutableShareFile(filename, self)
436 msf.check_write_enabler(write_enabler, si_s)
437 shares[sharenum] = msf
438 # write_enabler is good for all existing shares.
440 # Now evaluate test vectors.
442 for sharenum in test_and_write_vectors:
443 (testv, datav, new_length) = test_and_write_vectors[sharenum]
444 if sharenum in shares:
445 if not shares[sharenum].check_testv(testv):
446 self.log("testv failed: [%d]: %r" % (sharenum, testv))
447 testv_is_good = False
450 # compare the vectors against an empty share, in which all
451 # reads return empty strings.
452 if not EmptyShare().check_testv(testv):
453 self.log("testv failed (empty): [%d] %r" % (sharenum,
455 testv_is_good = False
458 # now gather the read vectors, before we do any writes
460 for sharenum, share in shares.items():
461 read_data[sharenum] = share.readv(read_vector)
464 expire_time = time.time() + 31*24*60*60 # one month
465 lease_info = LeaseInfo(ownerid,
466 renew_secret, cancel_secret,
467 expire_time, self.my_nodeid)
470 # now apply the write vectors
471 for sharenum in test_and_write_vectors:
472 (testv, datav, new_length) = test_and_write_vectors[sharenum]
474 if sharenum in shares:
475 shares[sharenum].unlink()
477 if sharenum not in shares:
478 # allocate a new share
479 allocated_size = 2000 # arbitrary, really
480 share = self._allocate_slot_share(bucketdir, secrets,
484 shares[sharenum] = share
485 shares[sharenum].writev(datav, new_length)
486 # and update the lease
487 shares[sharenum].add_or_renew_lease(lease_info)
490 # delete empty bucket directories
491 if not os.listdir(bucketdir):
496 self.add_latency("writev", time.time() - start)
497 return (testv_is_good, read_data)
499 def _allocate_slot_share(self, bucketdir, secrets, sharenum,
500 allocated_size, owner_num=0):
501 (write_enabler, renew_secret, cancel_secret) = secrets
502 my_nodeid = self.my_nodeid
503 fileutil.make_dirs(bucketdir)
504 filename = os.path.join(bucketdir, "%d" % sharenum)
505 share = create_mutable_sharefile(filename, my_nodeid, write_enabler,
509 def remote_slot_readv(self, storage_index, shares, readv):
512 si_s = si_b2a(storage_index)
513 lp = log.msg("storage: slot_readv %s %s" % (si_s, shares),
514 facility="tahoe.storage", level=log.OPERATIONAL)
515 si_dir = storage_index_to_dir(storage_index)
516 # shares exist if there is a file for them
517 bucketdir = os.path.join(self.sharedir, si_dir)
518 if not os.path.isdir(bucketdir):
519 self.add_latency("readv", time.time() - start)
522 for sharenum_s in os.listdir(bucketdir):
524 sharenum = int(sharenum_s)
527 if sharenum in shares or not shares:
528 filename = os.path.join(bucketdir, sharenum_s)
529 msf = MutableShareFile(filename, self)
530 datavs[sharenum] = msf.readv(readv)
531 log.msg("returning shares %s" % (datavs.keys(),),
532 facility="tahoe.storage", level=log.NOISY, parent=lp)
533 self.add_latency("readv", time.time() - start)
536 def remote_advise_corrupt_share(self, share_type, storage_index, shnum,
538 fileutil.make_dirs(self.corruption_advisory_dir)
539 now = time_format.iso_utc(sep="T")
540 si_s = si_b2a(storage_index)
541 # windows can't handle colons in the filename
542 fn = os.path.join(self.corruption_advisory_dir,
543 "%s--%s-%d" % (now, si_s, shnum)).replace(":","")
545 f.write("report: Share Corruption\n")
546 f.write("type: %s\n" % share_type)
547 f.write("storage_index: %s\n" % si_s)
548 f.write("share_number: %d\n" % shnum)
553 log.msg(format=("client claims corruption in (%(share_type)s) " +
554 "%(si)s-%(shnum)d: %(reason)s"),
555 share_type=share_type, si=si_s, shnum=shnum, reason=reason,
556 level=log.SCARY, umid="SGx2fA")