tahoe-lafs/src/allmydata/scripts/consolidate.py

400 lines
18 KiB
Python

import os, pickle, time
import sqlite3 as sqlite
import urllib
import simplejson
from allmydata.scripts.common_http import do_http
from allmydata.scripts.tahoe_backup import parse_old_timestamp, readonly, \
raiseHTTPError, HTTPError
from allmydata.util import hashutil, base32
from allmydata.util.netstring import netstring
from allmydata.scripts.common import get_alias, DEFAULT_ALIAS
TAG = "consolidator_dirhash_v1"
class CycleDetected(Exception):
pass
class Consolidator:
def __init__(self, options):
self.options = options
self.rootcap, path = get_alias(options.aliases, options.where,
DEFAULT_ALIAS)
assert path == ""
self.dbfile = options["dbfile"]
assert self.dbfile, "--dbfile is required"
self.backupfile = options["backupfile"]
assert self.backupfile, "--backupfile is required"
self.nodeurl = options["node-url"]
if not self.nodeurl.endswith("/"):
self.nodeurl += "/"
self.must_rescan_readonly_snapshots = not os.path.exists(self.dbfile)
self.db = sqlite.connect(self.dbfile)
self.cursor = self.db.cursor()
try:
self.cursor.execute("CREATE TABLE dirhashes"
"("
" dirhash TEXT PRIMARY KEY,"
" dircap TEXT"
")")
except sqlite.OperationalError, e:
if "table dirhashes already exists" not in str(e):
raise
def read_directory_json(self, dircap):
url = self.nodeurl + "uri/%s?t=json" % urllib.quote(dircap)
resp = do_http("GET", url)
if resp.status != 200:
raiseHTTPError("Error during directory GET", resp)
jd = simplejson.load(resp)
ntype, ndata = jd
if ntype != "dirnode":
return None
return ndata
def msg(self, text):
print >>self.options.stdout, text
self.options.stdout.flush()
def err(self, text):
print >>self.options.stderr, text
self.options.stderr.flush()
def consolidate(self):
try:
data = self.read_directory_json(self.rootcap + "/Backups")
except HTTPError:
self.err("Unable to list /Backups, maybe this account has none?")
return 1
kids = data["children"]
potential_systems = {}
for (childname, (childtype, childdata)) in kids.items():
if childtype != "dirnode":
continue
potential_systems[childname] = str(childdata["rw_uri"])
backup_data = {"Backups": data, "systems": {}, "archives": {}}
systems = {}
for name, sdircap in potential_systems.items():
sdata = self.read_directory_json(sdircap)
kids = sdata["children"]
if not u"Archives" in kids and not u"Latest Backup" in kids:
self.msg("%s: not a backupdir, no 'Archives' and 'Latest'" % name)
continue
self.msg("%s is a system" % name)
backup_data["systems"][name] = sdata
archives_dircap = kids[u"Archives"][1]["rw_uri"]
archives_data = self.read_directory_json(archives_dircap)
backup_data["archives"][name] = archives_data
systems[name] = archives_dircap
if not systems:
self.msg("No systems under /Backups, nothing to consolidate")
return 0
backupfile = self.backupfile
counter = 0
while os.path.exists(backupfile):
backupfile = self.backupfile + "." + str(counter)
counter += 1
f = open(backupfile, "wb")
pickle.dump(backup_data, f)
f.close()
for name, archives_dircap in sorted(systems.items()):
self.do_system(name, archives_dircap)
return 0
def do_system(self, system_name, archives_dircap):
# first we walk through the Archives list, looking for the existing
# snapshots. Each one will have a $NAME like "2008-11-16 10.34 PM"
# (in various forms: we use tahoe_backup.parse_old_timestamp to
# interpret it). At first, they'll all have $NAME and be writecaps.
# As we run, we will create $NAME-readonly (with a readcap) for each
# one (the first one will just be the readonly equivalent of the
# oldest snapshot: the others will be constructed out of shared
# directories). When we're done we'll have a $NAME-readonly for
# everything except the latest snapshot (to avoid any danger of
# modifying a backup that's already in progress). The very last step,
# which won't be enabled until we're sure that everything is working
# correctly, will replace each $NAME with $NAME-readonly.
# We maintain a table that maps dirhash (hash of directory contents)
# to a directory readcap which contains those contents. We use this
# to decide if we can link to an existing directory, or if we must
# create a brand new one. Usually we add to this table in two places:
# when we scan the oldest snapshot (which we've just converted to
# readonly form), and when we must create a brand new one. If the
# table doesn't exist (probably because we've manually deleted it),
# we will scan *all* the existing readonly snapshots, and repopulate
# the table. We keep this table in a SQLite database (rather than a
# pickle) because we want to update it persistently after every
# directory creation, and writing out a 10k entry pickle takes about
# 250ms
# 'snapshots' maps timestamp to [rwname, writecap, roname, readcap].
# The possibilities are:
# [$NAME, writecap, None, None] : haven't touched it
# [$NAME, writecap, $NAME-readonly, readcap] : processed, not replaced
# [None, None, $NAME, readcap] : processed and replaced
self.msg("consolidating system %s" % system_name)
self.directories_reused = 0
self.directories_used_as_is = 0
self.directories_created = 0
self.directories_seen = set()
self.directories_used = set()
data = self.read_directory_json(archives_dircap)
snapshots = {}
children = sorted(data["children"].items())
for i, (childname, (childtype, childdata)) in enumerate(children):
if childtype != "dirnode":
self.msg("non-dirnode %s in Archives/" % childname)
continue
timename = childname
if childname.endswith("-readonly"):
timename = childname[:-len("-readonly")]
timestamp = parse_old_timestamp(timename, self.options)
assert timestamp is not None, timename
snapshots.setdefault(timestamp, [None, None, None, None])
# if the snapshot is readonly (i.e. it has no rw_uri), we might
# need to re-scan it
is_readonly = not childdata.has_key("rw_uri")
if is_readonly:
readcap = str(childdata["ro_uri"])
if self.must_rescan_readonly_snapshots:
self.msg(" scanning old %s (%d/%d)" %
(childname, i+1, len(children)))
self.scan_old_directory(str(childdata["ro_uri"]))
snapshots[timestamp][2] = childname
snapshots[timestamp][3] = readcap
else:
writecap = str(childdata["rw_uri"])
snapshots[timestamp][0] = childname
snapshots[timestamp][1] = writecap
snapshots = [ [timestamp] + values
for (timestamp, values) in snapshots.items() ]
# now 'snapshots' is [timestamp, rwname, writecap, roname, readcap],
# which makes it easier to process in temporal order
snapshots.sort()
self.msg(" %d snapshots" % len(snapshots))
# we always ignore the last one, for safety
snapshots = snapshots[:-1]
first_snapshot = True
for i,(timestamp, rwname, writecap, roname, readcap) in enumerate(snapshots):
eta = "?"
start_created = self.directories_created
start_used_as_is = self.directories_used_as_is
start_reused = self.directories_reused
# [None, None, $NAME, readcap] : processed and replaced
# [$NAME, writecap, $NAME-readonly, readcap] : processed, not replaced
# [$NAME, writecap, None, None] : haven't touched it
if readcap and not writecap:
# skip past anything we've already processed and replaced
assert roname
assert not rwname
first_snapshot = False
self.msg(" %s already readonly" % roname)
continue
if readcap and writecap:
# we've processed it, creating a -readonly version, but we
# haven't replaced it.
assert roname
assert rwname
first_snapshot = False
self.msg(" %s processed but not yet replaced" % roname)
if self.options["really"]:
self.msg(" replacing %s with %s" % (rwname, roname))
self.put_child(archives_dircap, rwname, readcap)
self.delete_child(archives_dircap, roname)
continue
assert writecap
assert rwname
assert not readcap
assert not roname
roname = rwname + "-readonly"
# for the oldest snapshot, we can do a simple readonly conversion
if first_snapshot:
first_snapshot = False
readcap = readonly(writecap)
self.directories_used_as_is += 1
self.msg(" %s: oldest snapshot, using as-is" % rwname)
self.scan_old_directory(readcap)
else:
# for the others, we must scan their contents and build up a new
# readonly directory (which shares common subdirs with previous
# backups)
self.msg(" %s: processing (%d/%d)" % (rwname, i+1, len(snapshots)))
started = time.time()
readcap = self.process_directory(readonly(writecap), (rwname,))
elapsed = time.time() - started
eta = "%ds" % (elapsed * (len(snapshots) - i-1))
if self.options["really"]:
self.msg(" replaced %s" % rwname)
self.put_child(archives_dircap, rwname, readcap)
else:
self.msg(" created %s" % roname)
self.put_child(archives_dircap, roname, readcap)
snapshot_created = self.directories_created - start_created
snapshot_used_as_is = self.directories_used_as_is - start_used_as_is
snapshot_reused = self.directories_reused - start_reused
self.msg(" %s: done: %d dirs created, %d used as-is, %d reused, eta %s"
% (rwname,
snapshot_created, snapshot_used_as_is, snapshot_reused,
eta))
# done!
self.msg(" system done, dircounts: %d/%d seen/used, %d created, %d as-is, %d reused" \
% (len(self.directories_seen), len(self.directories_used),
self.directories_created, self.directories_used_as_is,
self.directories_reused))
def process_directory(self, readcap, path):
# I walk all my children (recursing over any subdirectories), build
# up a table of my contents, then see if I can re-use an old
# directory with the same contents. If not, I create a new directory
# for my contents. In all cases I return a directory readcap that
# points to my contents.
assert isinstance(readcap, str)
self.directories_seen.add(readcap)
# build up contents to pass to mkdir() (which uses t=set_children)
contents = {} # childname -> (type, rocap, metadata)
data = self.read_directory_json(readcap)
assert data is not None
hashkids = []
children_modified = False
for (childname, (childtype, childdata)) in sorted(data["children"].items()):
if childtype == "dirnode":
childpath = path + (childname,)
old_childcap = str(childdata["ro_uri"])
childcap = self.process_directory(old_childcap, childpath)
if childcap != old_childcap:
children_modified = True
contents[childname] = ("dirnode", childcap, None)
else:
childcap = str(childdata["ro_uri"])
contents[childname] = (childtype, childcap, None)
hashkids.append( (childname, childcap) )
dirhash = self.hash_directory_contents(hashkids)
old_dircap = self.get_old_dirhash(dirhash)
if old_dircap:
if self.options["verbose"]:
self.msg(" %r: reused" % "/".join(path))
assert isinstance(old_dircap, str)
self.directories_reused += 1
self.directories_used.add(old_dircap)
return old_dircap
if not children_modified:
# we're allowed to use this directory as-is
if self.options["verbose"]:
self.msg(" %r: used as-is" % "/".join(path))
new_dircap = readonly(readcap)
assert isinstance(new_dircap, str)
self.store_dirhash(dirhash, new_dircap)
self.directories_used_as_is += 1
self.directories_used.add(new_dircap)
return new_dircap
# otherwise, we need to create a new directory
if self.options["verbose"]:
self.msg(" %r: created" % "/".join(path))
new_dircap = readonly(self.mkdir(contents))
assert isinstance(new_dircap, str)
self.store_dirhash(dirhash, new_dircap)
self.directories_created += 1
self.directories_used.add(new_dircap)
return new_dircap
def put_child(self, dircap, childname, childcap):
url = self.nodeurl + "uri/%s/%s?t=uri" % (urllib.quote(dircap),
urllib.quote(childname))
resp = do_http("PUT", url, childcap)
if resp.status not in (200, 201):
raiseHTTPError("error during put_child", resp)
def delete_child(self, dircap, childname):
url = self.nodeurl + "uri/%s/%s" % (urllib.quote(dircap),
urllib.quote(childname))
resp = do_http("DELETE", url)
if resp.status not in (200, 201):
raiseHTTPError("error during delete_child", resp)
def mkdir(self, contents):
url = self.nodeurl + "uri?t=mkdir"
resp = do_http("POST", url)
if resp.status < 200 or resp.status >= 300:
raiseHTTPError("error during mkdir", resp)
dircap = str(resp.read().strip())
url = self.nodeurl + "uri/%s?t=set_children" % urllib.quote(dircap)
body = dict([ (childname, (contents[childname][0],
{"ro_uri": contents[childname][1],
"metadata": contents[childname][2],
}))
for childname in contents
])
resp = do_http("POST", url, simplejson.dumps(body))
if resp.status != 200:
raiseHTTPError("error during set_children", resp)
return dircap
def scan_old_directory(self, dircap, ancestors=()):
# scan this directory (recursively) and stash a hash of its contents
# in the DB. This assumes that all subdirs can be used as-is (i.e.
# they've already been declared immutable)
dircap = readonly(dircap)
if dircap in ancestors:
raise CycleDetected
ancestors = ancestors + (dircap,)
#self.visited.add(dircap)
# TODO: we could use self.visited as a mapping from dircap to dirhash,
# to avoid re-scanning old shared directories multiple times
self.directories_seen.add(dircap)
self.directories_used.add(dircap)
data = self.read_directory_json(dircap)
kids = []
for (childname, (childtype, childdata)) in data["children"].items():
childcap = str(childdata["ro_uri"])
if childtype == "dirnode":
self.scan_old_directory(childcap, ancestors)
kids.append( (childname, childcap) )
dirhash = self.hash_directory_contents(kids)
self.store_dirhash(dirhash, dircap)
return dirhash
def hash_directory_contents(self, kids):
kids.sort()
s = "".join([netstring(childname.encode("utf-8"))+netstring(childcap)
for (childname, childcap) in kids])
return hashutil.tagged_hash(TAG, s)
def store_dirhash(self, dirhash, dircap):
assert isinstance(dircap, str)
# existing items should prevail
try:
c = self.cursor
c.execute("INSERT INTO dirhashes (dirhash, dircap) VALUES (?,?)",
(base32.b2a(dirhash), dircap))
self.db.commit()
except sqlite.IntegrityError:
# already present
pass
def get_old_dirhash(self, dirhash):
self.cursor.execute("SELECT dircap FROM dirhashes WHERE dirhash=?",
(base32.b2a(dirhash),))
row = self.cursor.fetchone()
if not row:
return None
(dircap,) = row
return str(dircap)
def main(options):
c = Consolidator(options)
return c.consolidate()