]> git.rkrishnan.org Git - tahoe-lafs/tahoe-lafs.git/commitdiff
add sqlite-based backupdb, for #598 and others (including 'tahoe cp'). Not enabled...
authorBrian Warner <warner@allmydata.com>
Fri, 6 Feb 2009 01:17:56 +0000 (18:17 -0700)
committerBrian Warner <warner@allmydata.com>
Fri, 6 Feb 2009 01:17:56 +0000 (18:17 -0700)
src/allmydata/scripts/backupdb.py [new file with mode: 0644]
src/allmydata/test/test_backupdb.py [new file with mode: 0644]

diff --git a/src/allmydata/scripts/backupdb.py b/src/allmydata/scripts/backupdb.py
new file mode 100644 (file)
index 0000000..b93270c
--- /dev/null
@@ -0,0 +1,246 @@
+
+# the backupdb is only available if sqlite3 is available. Python-2.5.x and
+# beyond include sqlite3 in the standard library. For python-2.4, the
+# "pysqlite2" package (which, despite the confusing name, uses sqlite3) must
+# be installed. On debian, install python-pysqlite2
+
+import os.path, sys, time, random, stat
+
+DAY = 24*60*60
+MONTH = 30*DAY
+
+SCHEMA_v1 = """
+CREATE TABLE version
+(
+ version INTEGER  -- contains one row, set to 1
+);
+
+CREATE TABLE local_files
+(
+ path  VARCHAR(1024) PRIMARY KEY, -- index, this is os.path.abspath(fn)
+ size  INTEGER,       -- os.stat(fn)[stat.ST_SIZE]
+ mtime NUMBER,        -- os.stat(fn)[stat.ST_MTIME]
+ ctime NUMBER,        -- os.stat(fn)[stat.ST_CTIME]
+ fileid INTEGER
+);
+
+CREATE TABLE caps
+(
+ fileid INTEGER PRIMARY KEY AUTOINCREMENT,
+ filecap VARCHAR(256) UNIQUE       -- URI:CHK:...
+);
+
+CREATE TABLE last_upload
+(
+ fileid INTEGER PRIMARY KEY,
+ last_uploaded TIMESTAMP,
+ last_checked TIMESTAMP
+);
+
+"""
+
+def get_backupdb(dbfile, stderr=sys.stderr):
+    # open or create the given backupdb file. The parent directory must
+    # exist.
+    try:
+        import sqlite3
+        sqlite = sqlite3 # pyflakes whines about 'import sqlite3 as sqlite' ..
+    except ImportError:
+        try:
+            from pysqlite2 import dbapi2
+            sqlite = dbapi2 # .. when this clause does it too
+        except ImportError:
+            print >>stderr, "sqlite unavailable, not using backupdb"
+            return None
+
+    must_create = not os.path.exists(dbfile)
+    try:
+        db = sqlite.connect(dbfile)
+    except (EnvironmentError, sqlite.OperationalError), e:
+        print >>stderr, "Unable to create/open backupdb file %s: %s" % (dbfile, e)
+        return None
+
+    c = db.cursor()
+    if must_create:
+        c.executescript(SCHEMA_v1)
+        c.execute("INSERT INTO version (version) VALUES (1)")
+        db.commit()
+
+    try:
+        c.execute("SELECT version FROM version")
+        version = c.fetchone()[0]
+    except sqlite.DatabaseError, e:
+        # this indicates that the file is not a compatible database format.
+        # Perhaps it was created with an old version, or it might be junk.
+        print >>stderr, "backupdb file is unusable: %s" % e
+        return None
+
+    if version == 1:
+        return BackupDB_v1(sqlite, db)
+    print >>stderr, "Unable to handle backupdb version %s" % version
+    return None
+
+MUST_UPLOAD, ALREADY_UPLOADED = range(2)
+class Result:
+    def __init__(self, bdb, filecap, should_check,
+                 path, mtime, ctime, size):
+        self.bdb = bdb
+        self.filecap = filecap
+        self.should_check_p = should_check
+
+        self.path = path
+        self.mtime = mtime
+        self.ctime = ctime
+        self.size = size
+
+    def was_uploaded(self):
+        if self.filecap:
+            return self.filecap
+        return False
+
+    def did_upload(self, filecap):
+        self.bdb.did_upload(filecap,
+                            self.path,
+                            self.mtime, self.ctime, self.size)
+
+    def should_check(self):
+        return self.should_check_p
+
+    def did_check_healthy(self, results):
+        self.bdb.did_check_healthy(self.filecap, results)
+
+class BackupDB_v1:
+    VERSION = 1
+    NO_CHECK_BEFORE = 1*MONTH
+    ALWAYS_CHECK_AFTER = 2*MONTH
+
+    def __init__(self, sqlite_module, connection):
+        self.sqlite_module = sqlite_module
+        self.connection = connection
+        self.cursor = connection.cursor()
+
+    def check_file(self, path, use_timestamps=True):
+        """I will tell you if a given local file needs to be uploaded or not,
+        by looking in a database and seeing if I have a record of this file
+        having been uploaded earlier.
+
+        I return a Results object, synchronously. If r.was_uploaded() returns
+        False, you should upload the file. When you are finished uploading
+        it, call r.did_upload(filecap), so I can update my database.
+
+        If was_uploaded() returns a filecap, you might be able to avoid an
+        upload. Call r.must_check(), and if it says False, you can skip the
+        upload and use the filecap returned by was_uploaded().
+
+        If should_check() returns True, you should perform a filecheck on the
+        filecap returned by was_uploaded(). If the check indicates the file
+        is healthy, please call r.did_check_healthy(checker_results) so I can
+        update the database. If the check indicates the file is not healthy,
+        please upload the file and call r.did_upload(filecap) when you're
+        done.
+
+        I use_timestamps=True (the default), I will compare ctime and mtime
+        of the local file against an entry in my database, and consider the
+        file to be unchanged if ctime, mtime, and filesize are all the same
+        as the earlier version. If use_timestamps=False, I will not trust the
+        timestamps, so more files (perhaps all) will be marked as needing
+        upload. A future version of this database may hash the file to make
+        equality decisions, in which case use_timestamps=False will not
+        always imply r.must_upload()==True.
+
+        'path' points to a local file on disk, possibly relative to the
+        current working directory. The database stores absolute pathnames.
+        """
+
+        path = os.path.abspath(path)
+        s = os.stat(path)
+        size = s[stat.ST_SIZE]
+        ctime = s[stat.ST_CTIME]
+        mtime = s[stat.ST_MTIME]
+
+        now = time.time()
+        c = self.cursor
+
+        c.execute("SELECT size,mtime,ctime,fileid"
+                  " FROM local_files"
+                  " WHERE path=?",
+                  (path,))
+        row = self.cursor.fetchone()
+        if not row:
+            return Result(self, None, False, path, mtime, ctime, size)
+        (last_size,last_mtime,last_ctime,last_fileid) = row
+
+        c.execute("SELECT caps.filecap, last_upload.last_checked"
+                  " FROM caps,last_upload"
+                  " WHERE caps.fileid=? AND last_upload.fileid=?",
+                  (last_fileid, last_fileid))
+        row2 = c.fetchone()
+
+        if ((last_size != size
+             or not use_timestamps
+             or last_mtime != mtime
+             or last_ctime != ctime) # the file has been changed
+            or (not row2) # we somehow forgot where we put the file last time
+            ):
+            c.execute("DELETE FROM local_files WHERE path=?", (path,))
+            self.connection.commit()
+            return Result(self, None, False, path, mtime, ctime, size)
+
+        # at this point, we're allowed to assume the file hasn't been changed
+        (filecap, last_checked) = row2
+        age = now - last_checked
+
+        probability = ((age - self.NO_CHECK_BEFORE) /
+                       (self.ALWAYS_CHECK_AFTER - self.NO_CHECK_BEFORE))
+        probability = min(max(probability, 0.0), 1.0)
+        should_check = bool(random.random() < probability)
+
+        return Result(self, filecap, should_check, path, mtime, ctime, size)
+
+    def get_or_allocate_fileid_for_cap(self, filecap):
+        # find an existing fileid for this filecap, or insert a new one. The
+        # caller is required to commit() afterwards.
+
+        # mysql has "INSERT ... ON DUPLICATE KEY UPDATE", but not sqlite
+        # sqlite has "INSERT ON CONFLICT REPLACE", but not mysql
+        # So we use INSERT, ignore any error, then a SELECT
+        c = self.cursor
+        try:
+            c.execute("INSERT INTO caps (filecap) VALUES (?)", (filecap,))
+        except self.sqlite_module.IntegrityError:
+            pass
+        c.execute("SELECT fileid FROM caps WHERE filecap=?", (filecap,))
+        foundrow = c.fetchone()
+        assert foundrow
+        fileid = foundrow[0]
+        return fileid
+
+    def did_upload(self, filecap, path, mtime, ctime, size):
+        now = time.time()
+        fileid = self.get_or_allocate_fileid_for_cap(filecap)
+        try:
+            self.cursor.execute("INSERT INTO last_upload VALUES (?,?,?)",
+                                (fileid, now, now))
+        except self.sqlite_module.IntegrityError:
+            self.cursor.execute("UPDATE last_upload"
+                                " SET last_uploaded=?, last_checked=?"
+                                " WHERE fileid=?",
+                                (now, now, fileid))
+        try:
+            self.cursor.execute("INSERT INTO local_files VALUES (?,?,?,?,?)",
+                                (path, size, mtime, ctime, fileid))
+        except self.sqlite_module.IntegrityError:
+            self.cursor.execute("UPDATE local_files"
+                                " SET size=?, mtime=?, ctime=?, fileid=?"
+                                " WHERE path=?",
+                                (size, mtime, ctime, fileid, path))
+        self.connection.commit()
+
+    def did_check_healthy(self, filecap, results):
+        now = time.time()
+        fileid = self.get_or_allocate_fileid_for_cap(filecap)
+        self.cursor.execute("UPDATE last_upload"
+                            " SET last_checked=?"
+                            " WHERE fileid=?",
+                            (now, fileid))
+        self.connection.commit()
diff --git a/src/allmydata/test/test_backupdb.py b/src/allmydata/test/test_backupdb.py
new file mode 100644 (file)
index 0000000..3af720b
--- /dev/null
@@ -0,0 +1,145 @@
+
+import os.path, time
+from StringIO import StringIO
+from twisted.trial import unittest
+
+from allmydata.util import fileutil
+from allmydata.scripts import backupdb
+
+class BackupDB(unittest.TestCase):
+    def create_or_skip(self, dbfile):
+        stderr = StringIO()
+        bdb = backupdb.get_backupdb(dbfile, stderr=stderr)
+        if not bdb:
+            if "sqlite unavailable" in stderr.getvalue():
+                raise unittest.SkipTest("sqlite unavailable, skipping test")
+        return bdb
+
+    def test_basic(self):
+        self.basedir = basedir = os.path.join("backupdb", "create")
+        fileutil.make_dirs(basedir)
+        dbfile = os.path.join(basedir, "dbfile")
+        bdb = self.create_or_skip(dbfile)
+        self.failUnless(bdb)
+        self.failUnlessEqual(bdb.VERSION, 1)
+
+    def test_fail(self):
+        self.basedir = basedir = os.path.join("backupdb", "fail")
+        fileutil.make_dirs(basedir)
+
+        # put a non-DB file in the way
+        self.writeto("not-a-database", "I do not look like a sqlite database")
+        stderr_f = StringIO()
+        bdb = backupdb.get_backupdb(os.path.join(basedir, "not-a-database"),
+                                    stderr_f)
+        self.failUnlessEqual(bdb, None)
+        stderr = stderr_f.getvalue()
+        if "sqlite unavailable" in stderr:
+            pass
+        else:
+            self.failUnless("backupdb file is unusable" in stderr)
+            self.failUnless("file is encrypted or is not a database" in stderr)
+
+        # put a directory in the way, to exercise a different error path
+        where = os.path.join(basedir, "roadblock-dir")
+        fileutil.make_dirs(where)
+        stderr_f = StringIO()
+        bdb = backupdb.get_backupdb(where, stderr_f)
+        self.failUnlessEqual(bdb, None)
+        stderr = stderr_f.getvalue()
+        if "sqlite unavailable" in stderr:
+            pass
+        else:
+            self.failUnless(("Unable to create/open backupdb file %s" % where)
+                            in stderr)
+            self.failUnless("unable to open database file" in stderr)
+
+
+    def writeto(self, filename, data):
+        fn = os.path.join(self.basedir, filename)
+        parentdir = os.path.dirname(fn)
+        fileutil.make_dirs(parentdir)
+        f = open(fn, "w")
+        f.write(data)
+        f.close()
+        return fn
+
+    def test_check(self):
+        self.basedir = basedir = os.path.join("backupdb", "check")
+        fileutil.make_dirs(basedir)
+        dbfile = os.path.join(basedir, "dbfile")
+        bdb = self.create_or_skip(dbfile)
+        self.failUnless(bdb)
+
+        foo_fn = self.writeto("foo.txt", "foo.txt")
+        blah_fn = self.writeto("bar/blah.txt", "blah.txt")
+
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), False)
+        r.did_upload("foo-cap")
+
+        r = bdb.check_file(blah_fn)
+        self.failUnlessEqual(r.was_uploaded(), False)
+        r.did_upload("blah-cap")
+
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), "foo-cap")
+        self.failUnlessEqual(r.should_check(), False)
+
+        time.sleep(1.0) # make sure the timestamp changes
+        self.writeto("foo.txt", "NEW")
+
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), False)
+        r.did_upload("new-cap")
+
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), "new-cap")
+        self.failUnlessEqual(r.should_check(), False)
+        # if we spontaneously decide to upload it anyways, nothing should
+        # break
+        r.did_upload("new-cap")
+
+        r = bdb.check_file(foo_fn, use_timestamps=False)
+        self.failUnlessEqual(r.was_uploaded(), False)
+        r.did_upload("new-cap")
+
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), "new-cap")
+        self.failUnlessEqual(r.should_check(), False)
+
+        bdb.NO_CHECK_BEFORE = 0
+        bdb.ALWAYS_CHECK_AFTER = 0.1
+
+        r = bdb.check_file(blah_fn)
+        self.failUnlessEqual(r.was_uploaded(), "blah-cap")
+        self.failUnlessEqual(r.should_check(), True)
+        r.did_check_healthy("results") # we know they're ignored for now
+
+        r = bdb.check_file(blah_fn)
+        self.failUnlessEqual(r.was_uploaded(), "blah-cap")
+        self.failUnlessEqual(r.should_check(), False)
+
+        os.unlink(os.path.join(basedir, "foo.txt"))
+        fileutil.make_dirs(os.path.join(basedir, "foo.txt")) # file becomes dir
+        r = bdb.check_file(foo_fn)
+        self.failUnlessEqual(r.was_uploaded(), False)
+
+    def test_wrong_version(self):
+        self.basedir = basedir = os.path.join("backupdb", "wrong_version")
+        fileutil.make_dirs(basedir)
+
+        where = os.path.join(basedir, "tooold.db")
+        bdb = self.create_or_skip(where)
+        # reach into the DB and make it old
+        bdb.cursor.execute("UPDATE version SET version=0")
+        bdb.connection.commit()
+
+        # now the next time we open the database, it should be an unusable
+        # version
+        stderr_f = StringIO()
+        bdb = backupdb.get_backupdb(where, stderr_f)
+        self.failUnlessEqual(bdb, None)
+        stderr = stderr_f.getvalue()
+        self.failUnlessEqual(stderr.strip(),
+                             "Unable to handle backupdb version 0")