[PATCH 1 of 3 v4] branchcache: introduce revbranchcache for caching of revision branch names

Mads Kiilerich mads at kiilerich.com
Fri Jan 9 16:03:58 UTC 2015


# HG changeset patch
# User Mads Kiilerich <madski at unity3d.com>
# Date 1420671663 -3600
#      Thu Jan 08 00:01:03 2015 +0100
# Node ID 388a25e2964fba5be3e278628c99b325c5cbbafd
# Parent  bb3ee61cfaa1d3f085c7577f48f566a0f31a6985
branchcache: introduce revbranchcache for caching of revision branch names

It is expensive to retrieve the branch name of a revision. Very expensive when
creating a changectx and calling .branch() every time - slightly less when
using changelog.branchinfo().

Now, to speed things up, provide a way to cache the results on disk in an
efficient format. Each branchname is assigned a number, and for each revision
we store the number of the corresponding branch name. The branch names are
stored in a dedicated file which is strictly append only.

Branch names are usually reused across several revisions, and the total list of
branch names will thus be so small that it is feasible to read the whole set of
names before using the cache. It will however do that it might be more
efficient to use the changelog for retrieving the branch info for a single
revision.

The revision entries are stored in another file. This file is usually append
only, but if the repository has been modified, the file will be truncated and
the relevant parts rewritten on demand.

The entries for each revision are 8 bytes each, and the whole revision file
will thus be 1/8 of 00changelog.i.

Each revision entry contains the first 4 bytes of the corresponding node hash.
This is used as a check sum that always is verified before the entry is used.
That check is relatively expensive but it makes sure history modification is
detected and handled correctly. It will also detect and handle most revision
file corruptions.

This is just a cache. A new format can always be introduced if other
requirements or ideas make that seem like a good idea. Rebuilding the cache is
not really more expensive than it was to run for example 'hg log -b branchname'
before this cache was introduced.

This new method is still unused but promise to make some operations several
times faster once it actually is used.

Abandoning Python 2.4 would make it possible to implement this more efficiently
by using struct classes and pack_into. The Python code could probably also be
micro optimized or it could be implemented very efficiently in C where it would
be easy to control the data access.

diff --git a/mercurial/branchmap.py b/mercurial/branchmap.py
--- a/mercurial/branchmap.py
+++ b/mercurial/branchmap.py
@@ -9,6 +9,8 @@ from node import bin, hex, nullid, nullr
 import encoding
 import util
 import time
+from array import array
+from struct import calcsize, pack, unpack
 
 def _filename(repo):
     """name of a branchcache file for a given repo or repoview"""
@@ -285,3 +287,150 @@ class branchcache(dict):
         duration = time.time() - starttime
         repo.ui.log('branchcache', 'updated %s branch cache in %.4f seconds\n',
                     repo.filtername, duration)
+
+# Revision branch info cache
+
+_rbcversion = '-v1'
+_rbcnames = 'cache/rbc-names' + _rbcversion
+_rbcrevs = 'cache/rbc-revs' + _rbcversion
+# [4 byte hash prefix][4 byte branch name number with sign bit indicating open]
+_rbcrecfmt = '>4sI'
+_rbcrecsize = calcsize(_rbcrecfmt)
+_rbcnodelen = 4
+_rbcbranchidxmask = 0x7fffffff
+_rbccloseflag = 0x80000000
+
+class revbranchcache(object):
+    """Persistent cache, mapping from revision number to branch name and close.
+    This is a low level cache, independent of filtering.
+
+    Branch names are stored in rbc-names in internal encoding separated by 0.
+    rbc-names is append-only, and each branch name is only stored once and will
+    thus have a unique index.
+
+    The branch info for each revision is stored in rbc-revs as constant size
+    records. The whole file is read into memory, but it is only 'parsed' on
+    demand. The file is usually append-only but will be truncated if repo
+    modification is detected.
+    The record for each revision contains the first 4 bytes of the
+    corresponding node hash, and the record is only used if it still matches.
+    Even a completely trashed rbc-revs fill thus still give the right result
+    while converging towards full recovery ... assuming no incorrectly matching
+    node hashes.
+    The record also contains 4 bytes where 31 bits contains the index of the
+    branch and the last bit indicate that it is a branch close commit.
+    The usage pattern for rbc-revs is thus somewhat similar to 00changelog.i
+    and will grow with it but be 1/8th of its size.
+    """
+
+    def __init__(self, repo):
+        assert repo.filtername is None
+        self._names = [] # branch names in local encoding with static index
+        self._rbcrevs = array('c') # structs of type _rbcrecfmt
+        self._rbcsnameslen = 0
+        try:
+            bndata = repo.vfs.read(_rbcnames)
+            self._rbcsnameslen = len(bndata) # for verification before writing
+            self._names = [encoding.tolocal(bn) for bn in bndata.split('\0')]
+        except (IOError, OSError), inst:
+            repo.ui.debug("couldn't read revision branch cache names: %s\n" %
+                          inst)
+        if self._names:
+            try:
+                data = repo.vfs.read(_rbcrevs)
+                self._rbcrevs.fromstring(data)
+            except (IOError, OSError), inst:
+                repo.ui.debug("couldn't read revision branch cache: %s\n" %
+                              inst)
+        # remember number of good records on disk
+        self._rbcrevslen = min(len(self._rbcrevs) // _rbcrecsize,
+                               len(repo.changelog))
+        if self._rbcrevslen == 0:
+            self._names = []
+        self._rbcnamescount = len(self._names) # number of good names on disk
+        self._namesreverse = dict((b, r) for r, b in enumerate(self._names))
+
+    def branchinfo(self, changelog, rev):
+        """Return branch name and close flag for rev, using and updating
+        persistent cache."""
+        rbcrevidx = rev * _rbcrecsize
+
+        # if requested rev is missing, add and populate all missing revs
+        if len(self._rbcrevs) < rbcrevidx + _rbcrecsize:
+            first = len(self._rbcrevs) // _rbcrecsize
+            self._rbcrevs.extend('\0' * (len(changelog) * _rbcrecsize -
+                                         len(self._rbcrevs)))
+            for r in xrange(first, len(changelog)):
+                self._branchinfo(r)
+
+        # fast path: extract data from cache, use it if node is matching
+        reponode = changelog.node(rev)[:_rbcnodelen]
+        cachenode, branchidx = unpack(
+            _rbcrecfmt, buffer(self._rbcrevs, rbcrevidx, _rbcrecsize))
+        close = bool(branchidx & _rbccloseflag)
+        if close:
+            branchidx &= _rbcbranchidxmask
+        if cachenode == reponode:
+            return self._names[branchidx], close
+        # fall back to slow path and make sure it will be written to disk
+        self._rbcrevslen = min(self._rbcrevslen, rev)
+        return self._branchinfo(rev)
+
+    def _branchinfo(self, changelog, rev):
+        """Retrieve branch info from changelog and update _rbcrevs"""
+        b, close = changelog.branchinfo(rev)
+        if b in self._namesreverse:
+            branchidx = self._namesreverse[b]
+        else:
+            branchidx = len(self._names)
+            self._names.append(b)
+            self._namesreverse[b] = branchidx
+        reponode = changelog.node(rev)
+        if close:
+            branchidx |= _rbccloseflag
+        rbcrevidx = rev * _rbcrecsize
+        rec = array('c')
+        rec.fromstring(pack(_rbcrecfmt, reponode, branchidx))
+        self._rbcrevs[rbcrevidx:rbcrevidx + _rbcrecsize] = rec
+        return b, close
+
+    def write(self, repo):
+        """Save branch cache if it is dirty."""
+        if self._rbcnamescount < len(self._names):
+            try:
+                if self._rbcnamescount != 0:
+                    f = repo.vfs.open(_rbcnames, 'ab')
+                    if f.tell() == self._rbcsnameslen:
+                        f.write('\0')
+                    else:
+                        f.close()
+                        self._rbcnamescount = 0
+                        self._rbcrevslen = 0
+                if self._rbcnamescount == 0:
+                    f = repo.vfs.open(_rbcnames, 'wb')
+                f.write('\0'.join(encoding.fromlocal(b)
+                                  for b in self._names[self._rbcnamescount:]))
+                self._rbcsnameslen = f.tell()
+                f.close()
+            except (IOError, OSError, util.Abort), inst:
+                repo.ui.debug("couldn't write revision branch cache names: "
+                              "%s\n" % inst)
+                return
+            self._rbcnamescount = len(self._names)
+
+        start = self._rbcrevslen * _rbcrecsize
+        if start != len(self._rbcrevs):
+            self._rbcrevslen = min(len(repo.changelog),
+                                   len(self._rbcrevs) // _rbcrecsize)
+            try:
+                f = repo.vfs.open(_rbcrevs, 'ab')
+                if f.tell() != start:
+                    f.seek(start)
+                    f.truncate()
+                end = self._rbcrevslen * _rbcrecsize
+                f.write(self._rbcrevs[start:end])
+                f.close()
+            except (IOError, OSError, util.Abort), inst:
+                repo.ui.debug("couldn't write revision branch cache: %s\n" %
+                              inst)
+                return


More information about the Mercurial-devel mailing list