1 """High-level interface for working with LBS archives.
3 This module provides an easy interface for reading from and manipulating
4 various parts of an LBS archive:
5 - listing the snapshots and segments present
6 - reading segment contents
7 - parsing snapshot descriptors and snapshot metadata logs
8 - reading and maintaining the local object database
11 from __future__ import division
12 import os, re, sha, tarfile, tempfile, thread
13 from pysqlite2 import dbapi2 as sqlite3
15 # Maximum number of nested indirect references allowed in a snapshot.
16 MAX_RECURSION_DEPTH = 3
19 """A class which merely acts as a data container.
21 Instances of this class (or its subclasses) are merely used to store data
22 in various attributes. No methods are provided.
26 return "<%s %s>" % (self.__class__, self.__dict__)
28 CHECKSUM_ALGORITHMS = {
32 class ChecksumCreator:
33 """Compute an LBS checksum for provided data.
35 The algorithm used is selectable, but currently defaults to sha1.
38 def __init__(self, algorithm='sha1'):
39 self.algorithm = algorithm
40 self.hash = CHECKSUM_ALGORITHMS[algorithm]()
42 def update(self, data):
43 self.hash.update(data)
47 return "%s=%s" % (self.algorithm, self.hash.hexdigest())
49 class ChecksumVerifier:
50 """Verify whether a checksum from a snapshot matches the supplied data."""
52 def __init__(self, checksumstr):
53 """Create an object to check the supplied checksum."""
55 (algo, checksum) = checksumstr.split("=", 1)
56 self.checksum = checksum
57 self.hash = CHECKSUM_ALGORITHMS[algo]()
59 def update(self, data):
60 self.hash.update(data)
63 """Return a boolean indicating whether the checksum matches."""
65 result = self.hash.hexdigest()
66 return result == self.checksum
68 class LowlevelDataStore:
69 """Access to the backup store containing segments and snapshot descriptors.
71 Instances of this class are used to get direct filesystem-level access to
72 the backup data. To read a backup, a caller will ordinarily not care about
73 direct access to backup segments, but will instead merely need to access
74 objects from those segments. The ObjectStore class provides a suitable
75 wrapper around a DataStore to give this high-level access.
78 def __init__(self, path):
81 # Low-level filesystem access. These methods could be overwritten to
82 # provide access to remote data stores.
83 def lowlevel_list(self):
84 """Get a listing of files stored."""
86 return os.listdir(self.path)
88 def lowlevel_open(self, filename):
89 """Return a file-like object for reading data from the given file."""
91 return open(os.path.join(self.path, filename), 'rb')
93 def lowlevel_stat(self, filename):
94 """Return a dictionary of information about the given file.
96 Currently, the only defined field is 'size', giving the size of the
100 stat = os.stat(os.path.join(self.path, filename))
101 return {'size': stat.st_size}
103 # Slightly higher-level list methods.
104 def list_snapshots(self):
105 for f in self.lowlevel_list():
106 m = re.match(r"^snapshot-(.*)\.lbs$", f)
110 def list_segments(self):
111 for f in self.lowlevel_list():
112 m = re.match(r"^([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})(\.\S+)?$", f)
117 def __init__(self, data_store):
118 self.store = data_store
123 def get_cachedir(self):
124 if self.cachedir is None:
125 self.cachedir = tempfile.mkdtemp(".lbs")
129 if self.cachedir is not None:
130 # TODO: Avoid use of system, make this safer
131 os.system("rm -rf " + self.cachedir)
135 def parse_ref(refstr):
136 m = re.match(r"^([-0-9a-f]+)\/([0-9a-f]+)(\(\S+\))?(\[(\d+)\+(\d+)\])?$", refstr)
141 checksum = m.group(3)
144 if checksum is not None:
145 checksum = checksum.lstrip("(").rstrip(")")
147 if slice is not None:
148 slice = (int(m.group(5)), int(m.group(6)))
150 return (segment, object, checksum, slice)
152 def get_segment(self, segment):
153 raw = self.store.lowlevel_open(segment + ".tar.gpg")
155 (input, output) = os.popen2("lbs-filter-gpg --decrypt")
156 def copy_thread(src, dst):
159 block = src.read(BLOCK_SIZE)
160 if len(block) == 0: break
164 thread.start_new_thread(copy_thread, (raw, input))
167 def load_segment(self, segment):
168 seg = tarfile.open(segment, 'r|', self.get_segment(segment))
170 data_obj = seg.extractfile(item)
171 path = item.name.split('/')
172 if len(path) == 2 and path[0] == segment:
173 yield (path[1], data_obj.read())
175 def load_snapshot(self, snapshot):
176 file = self.store.lowlevel_open("snapshot-" + snapshot + ".lbs")
177 return file.read().splitlines(True)
179 def extract_segment(self, segment):
180 segdir = os.path.join(self.get_cachedir(), segment)
182 for (object, data) in self.load_segment(segment):
183 f = open(os.path.join(segdir, object), 'wb')
187 def load_object(self, segment, object):
188 path = os.path.join(self.get_cachedir(), segment, object)
189 if not os.access(path, os.R_OK):
190 self.extract_segment(segment)
191 if segment in self.lru_list: self.lru_list.remove(segment)
192 self.lru_list.append(segment)
193 while len(self.lru_list) > self.CACHE_SIZE:
194 os.system("rm -rf " + os.path.join(self.cachedir, self.lru_list[0]))
195 self.lru_list = self.lru_list[1:]
196 return open(path, 'rb').read()
198 def get(self, refstr):
199 """Fetch the given object and return it.
201 The input should be an object reference, in string form.
204 (segment, object, checksum, slice) = self.parse_ref(refstr)
206 data = self.load_object(segment, object)
208 if checksum is not None:
209 verifier = ChecksumVerifier(checksum)
210 verifier.update(data)
211 if not verifier.valid():
214 if slice is not None:
215 (start, length) = slice
216 data = data[start:start+length]
217 if len(data) != length: raise IndexError
221 def parse(lines, terminate=None):
222 """Generic parser for RFC822-style "Key: Value" data streams.
224 This parser can be used to read metadata logs and snapshot root descriptor
227 lines must be an iterable object which yields a sequence of lines of input.
229 If terminate is specified, it is used as a predicate to determine when to
230 stop reading input lines.
237 # Strip off a trailing newline, if present
238 if len(l) > 0 and l[-1] == "\n":
241 if terminate is not None and terminate(l):
242 if len(dict) > 0: yield dict
247 m = re.match(r"^(\w+):\s*(.*)$", l)
249 dict[m.group(1)] = m.group(2)
250 last_key = m.group(1)
251 elif len(l) > 0 and l[0].isspace() and last_key is not None:
256 if len(dict) > 0: yield dict
258 def parse_full(lines):
260 return parse(lines).next()
261 except StopIteration:
264 def read_metadata(object_store, root):
265 """Iterate through all lines in the metadata log, following references."""
267 # Stack for keeping track of recursion when following references to
268 # portions of the log. The last entry in the stack corresponds to the
269 # object currently being parsed. Each entry is a list of lines which have
270 # been reversed, so that popping successive lines from the end of each list
271 # will return lines of the metadata log in order.
274 def follow_ref(refstr):
275 if len(stack) >= MAX_RECURSION_DEPTH: raise OverflowError
276 lines = object_store.get(refstr).splitlines(True)
282 while len(stack) > 0:
289 # An indirect reference which we must follow?
290 if len(line) > 0 and line[0] == '@':
298 """Metadata for a single file (or directory or...) from a snapshot."""
300 # Functions for parsing various datatypes that can appear in a metadata log
304 """Decode an integer, expressed in decimal, octal, or hexadecimal."""
305 if s.startswith("0x"):
307 elif s.startswith("0"):
314 """Decode a URI-encoded (%xx escapes) string."""
315 def hex_decode(m): return chr(int(m.group(1), 16))
316 return re.sub(r"%([0-9a-f]{2})", hex_decode, s)
320 """An unecoded string."""
325 """Decode a user/group to a tuple of uid/gid followed by name."""
327 uid = MetadataItem.decode_int(items[0])
330 if items[1].startswith("(") and items[1].endswith(")"):
331 name = MetadataItem.decode_str(items[1][1:-1])
335 def decode_device(s):
336 """Decode a device major/minor number."""
337 (major, minor) = map(MetadataItem.decode_int, s.split("/"))
338 return (major, minor)
342 def __init__(self, fields, object_store):
343 """Initialize from a dictionary of key/value pairs from metadata log."""
346 self.object_store = object_store
348 self.items = self.Items()
349 for (k, v) in fields.items():
350 if k in self.field_types:
351 decoder = self.field_types[k]
352 setattr(self.items, k, decoder(v))
356 """Return an iterator for the data blocks that make up a file."""
358 # This traverses the list of blocks that make up a file, following
359 # indirect references. It is implemented in much the same way as
360 # read_metadata, so see that function for details of the technique.
362 objects = self.fields['data'].split()
366 def follow_ref(refstr):
367 if len(stack) >= MAX_RECURSION_DEPTH: raise OverflowError
368 objects = self.object_store.get(refstr).split()
370 stack.append(objects)
372 while len(stack) > 0:
379 # An indirect reference which we must follow?
380 if len(ref) > 0 and ref[0] == '@':
385 # Description of fields that might appear, and how they should be parsed.
386 MetadataItem.field_types = {
387 'name': MetadataItem.decode_str,
388 'type': MetadataItem.raw_str,
389 'mode': MetadataItem.decode_int,
390 'device': MetadataItem.decode_device,
391 'user': MetadataItem.decode_user,
392 'group': MetadataItem.decode_user,
393 'mtime': MetadataItem.decode_int,
394 'links': MetadataItem.decode_int,
395 'inode': MetadataItem.raw_str,
396 'checksum': MetadataItem.decode_str,
397 'size': MetadataItem.decode_int,
398 'contents': MetadataItem.decode_str,
401 def iterate_metadata(object_store, root):
402 for d in parse(read_metadata(object_store, root), lambda l: len(l) == 0):
403 yield MetadataItem(d, object_store)
406 """Access to the local database of snapshot contents and object checksums.
408 The local database is consulted when creating a snapshot to determine what
409 data can be re-used from old snapshots. Segment cleaning is performed by
410 manipulating the data in the local database; the local database also
411 includes enough data to guide the segment cleaning process.
414 def __init__(self, path, dbname="localdb.sqlite"):
415 self.db_connection = sqlite3.connect(path + "/" + dbname)
417 # Low-level database access. Use these methods when there isn't a
418 # higher-level interface available. Exception: do, however, remember to
419 # use the commit() method after making changes to make sure they are
420 # actually saved, even when going through higher-level interfaces.
422 "Commit any pending changes to the local database."
423 self.db_connection.commit()
426 "Roll back any pending changes to the local database."
427 self.db_connection.rollback()
430 "Return a DB-API cursor for directly accessing the local database."
431 return self.db_connection.cursor()
433 def garbage_collect(self):
434 """Delete entries from old snapshots from the database."""
438 # Delete old snapshots.
439 cur.execute("""delete from snapshots
440 where snapshotid < (select max(snapshotid)
443 # Delete entries in the snapshot_contents table which are for
444 # non-existent snapshots.
445 cur.execute("""delete from snapshot_contents
446 where snapshotid not in
447 (select snapshotid from snapshots)""")
449 # Find segments which contain no objects used by any current snapshots,
450 # and delete them from the segment table.
451 cur.execute("""delete from segments where segmentid not in
452 (select distinct segmentid from snapshot_contents
453 natural join block_index)""")
455 # Finally, delete objects contained in non-existent segments. We can't
456 # simply delete unused objects, since we use the set of unused objects
457 # to determine the used/free ratio of segments.
458 cur.execute("""delete from block_index
459 where segmentid not in
460 (select segmentid from segments)""")
463 class SegmentInfo(Struct): pass
465 def get_segment_cleaning_list(self, age_boost=0.0):
466 """Return a list of all current segments with information for cleaning.
468 Return all segments which are currently known in the local database
469 (there might be other, older segments in the archive itself), and
470 return usage statistics for each to help decide which segments to
473 The returned list will be sorted by estimated cleaning benefit, with
474 segments that are best to clean at the start of the list.
476 If specified, the age_boost parameter (measured in days) will added to
477 the age of each segment, as a way of adjusting the benefit computation
478 before a long-lived snapshot is taken (for example, age_boost might be
479 set to 7 when cleaning prior to taking a weekly snapshot).
484 cur.execute("""select segmentid, used, size, mtime,
485 julianday('now') - mtime as age from segment_info""")
487 info = self.SegmentInfo()
489 info.used_bytes = row[1]
490 info.size_bytes = row[2]
492 info.age_days = row[4]
494 # Benefit calculation: u is the estimated fraction of each segment
495 # which is utilized (bytes belonging to objects still in use
496 # divided by total size; this doesn't take compression or storage
497 # overhead into account, but should give a reasonable estimate).
499 # The total benefit is a heuristic that combines several factors:
500 # the amount of space that can be reclaimed (1 - u), an ageing
501 # factor (info.age_days) that favors cleaning old segments to young
502 # ones and also is more likely to clean segments that will be
503 # rewritten for long-lived snapshots (age_boost), and finally a
504 # penalty factor for the cost of re-uploading data (u + 0.1).
505 u = info.used_bytes / info.size_bytes
506 info.cleaning_benefit \
507 = (1 - u) * (info.age_days + age_boost) / (u + 0.1)
509 segments.append(info)
511 segments.sort(cmp, key=lambda s: s.cleaning_benefit, reverse=True)
514 def mark_segment_expired(self, segment):
515 """Mark a segment for cleaning in the local database.
517 The segment parameter should be either a SegmentInfo object or an
518 integer segment id. Objects in the given segment will be marked as
519 expired, which means that any future snapshots that would re-use those
520 objects will instead write out a new copy of the object, and thus no
521 future snapshots will depend upon the given segment.
524 if isinstance(segment, int):
526 elif isinstance(segment, self.SegmentInfo):
529 raise TypeError("Invalid segment: %s, must be of type int or SegmentInfo, not %s" % (segment, type(segment)))
532 cur.execute("update block_index set expired = 1 where segmentid = ?",
535 def balance_expired_objects(self):
536 """Analyze expired objects in segments to be cleaned and group by age.
538 Update the block_index table of the local database to group expired
539 objects by age. The exact number of buckets and the cutoffs for each
540 are dynamically determined. Calling this function after marking
541 segments expired will help in the segment cleaning process, by ensuring
542 that when active objects from clean segments are rewritten, they will
543 be placed into new segments roughly grouped by age.
546 # The expired column of the block_index table is used when generating a
547 # new LBS snapshot. A null value indicates that an object may be
548 # re-used. Otherwise, an object must be written into a new segment if
549 # needed. Objects with distinct expired values will be written into
550 # distinct segments, to allow for some grouping by age. The value 0 is
551 # somewhat special in that it indicates any rewritten objects can be
552 # placed in the same segment as completely new objects; this can be
553 # used for very young objects which have been expired, or objects not
554 # expected to be encountered.
556 # In the balancing process, all objects which are not used in any
557 # current snapshots will have expired set to 0. Objects which have
558 # been seen will be sorted by age and will have expired values set to
559 # 0, 1, 2, and so on based on age (with younger objects being assigned
560 # lower values). The number of buckets and the age cutoffs is
561 # determined by looking at the distribution of block ages.
565 # First step: Mark all unused-and-expired objects with expired = -1,
566 # which will cause us to mostly ignore these objects when rebalancing.
567 # At the end, we will set these objects to be in group expired = 0.
568 # Mark expired objects which still seem to be in use with expired = 0;
569 # these objects will later have values set to indicate groupings of
570 # objects when repacking.
571 cur.execute("""update block_index set expired = -1
572 where expired is not null""")
574 cur.execute("""update block_index set expired = 0
575 where expired is not null and blockid in
576 (select blockid from snapshot_contents)""")
578 # We will want to aim for at least one full segment for each bucket
579 # that we eventually create, but don't know how many bytes that should
580 # be due to compression. So compute the average number of bytes in
581 # each expired segment as a rough estimate for the minimum size of each
582 # bucket. (This estimate could be thrown off by many not-fully-packed
583 # segments, but for now don't worry too much about that.) If we can't
584 # compute an average, it's probably because there are no expired
585 # segments, so we have no more work to do.
586 cur.execute("""select avg(size) from segment_info
588 (select distinct segmentid from block_index
589 where expired is not null)""")
590 segment_size_estimate = cur.fetchone()[0]
591 if not segment_size_estimate:
594 # Next, extract distribution of expired objects (number and size) by
595 # age. Save the timestamp for "now" so that the classification of
596 # blocks into age buckets will not change later in the function, after
597 # time has passed. Set any timestamps in the future to now, so we are
598 # guaranteed that for the rest of this function, age is always
600 cur.execute("select julianday('now')")
601 now = cur.fetchone()[0]
603 cur.execute("""update block_index set timestamp = ?
604 where timestamp > ? and expired is not null""",
607 cur.execute("""select round(? - timestamp) as age, count(*), sum(size)
608 from block_index where expired = 0
609 group by age order by age""", (now,))
610 distribution = cur.fetchall()
612 # Start to determine the buckets for expired objects. Heuristics used:
613 # - An upper bound on the number of buckets is given by the number of
614 # segments we estimate it will take to store all data. In fact,
615 # aim for a couple of segments per bucket.
616 # - Place very young objects in bucket 0 (place with new objects)
617 # unless there are enough of them to warrant a separate bucket.
618 # - Try not to create unnecessarily many buckets, since fewer buckets
619 # will allow repacked data to be grouped based on spatial locality
620 # (while more buckets will group by temporal locality). We want a
623 total_bytes = sum([i[2] for i in distribution])
624 target_buckets = 2 * (total_bytes / segment_size_estimate) ** 0.4
625 min_size = 1.5 * segment_size_estimate
626 target_size = max(2 * segment_size_estimate,
627 total_bytes / target_buckets)
629 print "segment_size:", segment_size_estimate
630 print "distribution:", distribution
631 print "total_bytes:", total_bytes
632 print "target_buckets:", target_buckets
633 print "min, target size:", min_size, target_size
635 # Chosen cutoffs. Each bucket consists of objects with age greater
636 # than one cutoff value, but not greater than the next largest cutoff.
639 # Starting with the oldest objects, begin grouping together into
640 # buckets of size at least target_size bytes.
641 distribution.reverse()
643 min_age_bucket = False
644 for (age, items, size) in distribution:
645 if bucket_size >= target_size \
646 or (age < MIN_AGE and not min_age_bucket):
647 if bucket_size < target_size and len(cutoffs) > 0:
654 min_age_bucket = True
656 # The last (youngest) bucket will be group 0, unless it has enough data
657 # to be of size min_size by itself, or there happen to be no objects
658 # less than MIN_AGE at all.
659 if bucket_size >= min_size or not min_age_bucket:
663 print "cutoffs:", cutoffs
665 # Update the database to assign each object to the appropriate bucket.
667 for i in range(len(cutoffs)):
668 cur.execute("""update block_index set expired = ?
669 where round(? - timestamp) > ? and expired >= 0""",
670 (i, now, cutoffs[i]))
671 cur.execute("update block_index set expired = 0 where expired = -1")