Reduce hardlink memory usage

git-svn-id: http://svn.savannah.nongnu.org/svn/rdiff-backup/trunk@719 2b77aa54-bcbc-44c9-a7ec-4f6cf2b41109

5 files changed, 44 insertions, 30 deletions

diff --git a/rdiff-backup/CHANGELOG b/rdiff-backup/CHANGELOG
index 8c9aa16..4766486 100644
--- a/rdiff-backup/CHANGELOG
+++ b/rdiff-backup/CHANGELOG
@@ -14,6 +14,9 @@ Added supplementary rdiff-backup-statistics utility for parsing
 rdiff-backup's statistics files (originally based off perl script by
 Dean Gaudet).
 
+rdiff-backup should now use much less memory than v1.1.1-1.1.4 if you
+have lots of hard links.
+
 
 New in v1.1.4 (2005/12/13)
 --------------------------
diff --git a/rdiff-backup/rdiff-backup.1 b/rdiff-backup/rdiff-backup.1
index e785e98..e13a9f0 100644
--- a/rdiff-backup/rdiff-backup.1
+++ b/rdiff-backup/rdiff-backup.1
@@ -314,11 +314,8 @@ rdiff-backup-data directory.  rdiff-backup will run slightly quicker
 and take up a bit less space.
 .TP
 .BI --no-hard-links
-Don't replicate hard links on destination side.  Note that because
-metadata is written to a separate file, hard link information will not
-be lost even if the --no-hard-links option is given (however, mirror
-files will not be linked).  If many hard-linked files are present,
-this option can drastically decrease memory usage.
+Don't replicate hard links on destination side.  If many hard-linked
+files are present, this option can drastically decrease memory usage.
 .TP
 .B --null-separator
 Use nulls (\\0) instead of newlines (\\n) as line separators, which
diff --git a/rdiff-backup/rdiff_backup/Hardlink.py b/rdiff-backup/rdiff_backup/Hardlink.py
index 855c512..1dddbbb 100644
--- a/rdiff-backup/rdiff_backup/Hardlink.py
+++ b/rdiff-backup/rdiff_backup/Hardlink.py
@@ -31,15 +31,15 @@ source side should only transmit inode information.
 """
 
 from __future__ import generators
-import cPickle
-import Globals, Time, rpath, log, robust, errno
+import Globals, Time, log, robust, errno
 
 # The keys in this dictionary are (inode, devloc) pairs.  The values
-# are a pair (index, remaining_links, dest_key) where index is the
-# rorp index of the first such linked file, remaining_links is the
-# number of files hard linked to this one we may see, and key is
+# are a pair (index, remaining_links, dest_key, sha1sum) where index
+# is the rorp index of the first such linked file, remaining_links is
+# the number of files hard linked to this one we may see, and key is
 # either (dest_inode, dest_devloc) or None, and represents the
-# hardlink info of the existing file on the destination.
+# hardlink info of the existing file on the destination.  Finally
+# sha1sum is the hash of the file if it exists, or None.
 _inode_index = None
 
 def initialize_dictionaries():
@@ -64,7 +64,9 @@ def add_rorp(rorp, dest_rorp = None):
 		if not dest_rorp: dest_key = None
 		elif dest_rorp.getnumlinks() == 1: dest_key = "NA"
 		else: dest_key = get_inode_key(dest_rorp)
-		_inode_index[rp_inode_key] = (rorp.index, rorp.getnumlinks(), dest_key)
+		digest = rorp.has_sha1() and rorp.get_sha1() or None
+		_inode_index[rp_inode_key] = (rorp.index, rorp.getnumlinks(),
+									  dest_key, digest)
 	return rp_inode_key
 
 def del_rorp(rorp):
@@ -73,12 +75,12 @@ def del_rorp(rorp):
 	rp_inode_key = get_inode_key(rorp)
 	val = _inode_index.get(rp_inode_key)
 	if not val: return
-	index, remaining, dest_key = val
+	index, remaining, dest_key, digest = val
 	if remaining == 1:
 		del _inode_index[rp_inode_key]
 		return 1
 	else:
-		_inode_index[rp_inode_key] = (index, remaining-1, dest_key)
+		_inode_index[rp_inode_key] = (index, remaining-1, dest_key, digest)
 		return 0
 
 def rorp_eq(src_rorp, dest_rorp):
@@ -95,11 +97,11 @@ def rorp_eq(src_rorp, dest_rorp):
 
 	if src_rorp.getnumlinks() < dest_rorp.getnumlinks(): return 0
 	src_key = get_inode_key(src_rorp)
-	index, remaining, dest_key = _inode_index[src_key]
+	index, remaining, dest_key, digest = _inode_index[src_key]
 	if dest_key == "NA":
 		# Allow this to be ok for first comparison, but not any
 		# subsequent ones
-		_inode_index[src_key] = (index, remaining, None)
+		_inode_index[src_key] = (index, remaining, None, None)
 		return 1
 	return dest_key == get_inode_key(dest_rorp)
 
@@ -114,6 +116,10 @@ def get_link_index(rorp):
 	"""Return first index on target side rorp is already linked to"""
 	return _inode_index[get_inode_key(rorp)][0]
 
+def get_sha1(rorp):
+	"""Return sha1 digest of what rorp is linked to"""
+	return _inode_index[get_inode_key(rorp)][3]
+
 def link_rp(diff_rorp, dest_rpath, dest_root = None):
 	"""Make dest_rpath into a link using link flag in diff_rorp"""
 	if not dest_root: dest_root = dest_rpath # use base of dest_rpath
diff --git a/rdiff-backup/rdiff_backup/backup.py b/rdiff-backup/rdiff_backup/backup.py
index 819ae91..181c918 100644
--- a/rdiff-backup/rdiff_backup/backup.py
+++ b/rdiff-backup/rdiff_backup/backup.py
@@ -295,11 +295,6 @@ class CacheCollatedPostProcess:
 		# after we're finished with them
 		self.dir_perms_list = []
 
-		# A dictionary of {index: source_rorp}.  We use this to
-		# hold the digest of a hard linked file so it only needs to be
-		# computed once.
-		self.inode_digest_dict = {}
-
 		# Contains list of (index, (source_rorp, diff_rorp)) pairs for
 		# the parent directories of the last item in the cache.
 		self.parent_list = []
@@ -326,8 +321,7 @@ class CacheCollatedPostProcess:
 
 		"""
 		if Globals.preserve_hardlinks and source_rorp:
-			if Hardlink.add_rorp(source_rorp, dest_rorp):
-				self.inode_digest_dict[source_rorp.index] = source_rorp
+			Hardlink.add_rorp(source_rorp, dest_rorp)
 		if (dest_rorp and dest_rorp.isdir() and Globals.process_uid != 0
 			and dest_rorp.getperms() % 01000 < 0700):
 			self.unreadable_dir_init(source_rorp, dest_rorp)
@@ -394,8 +388,7 @@ class CacheCollatedPostProcess:
 
 		"""
 		if Globals.preserve_hardlinks and source_rorp:
-			if Hardlink.del_rorp(source_rorp):
-				del self.inode_digest_dict[source_rorp.index]
+			Hardlink.del_rorp(source_rorp)
 
 		if not changed or success:
 			if source_rorp: self.statfileobj.add_source_file(source_rorp)
@@ -469,10 +462,10 @@ class CacheCollatedPostProcess:
 
 	def update_hardlink_hash(self, diff_rorp):
 		"""Tag associated source_rorp with same hash diff_rorp points to"""
-		orig_rorp = self.inode_digest_dict[diff_rorp.get_link_flag()]
-		if orig_rorp.has_sha1():
-			new_source_rorp = self.get_source_rorp(diff_rorp.index)
-			new_source_rorp.set_sha1(orig_rorp.get_sha1())
+		sha1sum = Hardlink.get_sha1(diff_rorp)
+		if not sha1sum: return
+		source_rorp = self.get_source_rorp(diff_rorp.index)
+		source_rorp.set_sha1(sha1sum)
 
 	def close(self):
 		"""Process the remaining elements in the cache"""
diff --git a/rdiff-backup/testing/commontest.py b/rdiff-backup/testing/commontest.py
index 04cdb93..e9b272e 100644
--- a/rdiff-backup/testing/commontest.py
+++ b/rdiff-backup/testing/commontest.py
@@ -386,4 +386,19 @@ def raise_interpreter(use_locals = None):
 	else: local_dict = globals()
 	code.InteractiveConsole(local_dict).interact()
 
-
+def getrefs(i, depth):
+	"""Get the i'th object in memory, return objects that reference it"""
+	import sys, gc, types
+	o = sys.getobjects(i)[-1]
+	for d in range(depth):
+		for ref in gc.get_referrers(o):
+			if type(ref) in (types.ListType, types.DictType,
+								types.InstanceType):
+				if type(ref) is types.DictType and ref.has_key('copyright'):
+					continue
+				o = ref
+				break
+		else:
+			print "Max depth ", d
+			return o
+	return o