# HG changeset patch # User Benoit Boissinot # Date 1135710589 21600 # Node ID 14d1f1868bf636296e3323aef660355c2800af73 # Parent 96b47ef8f7405a6844546605991382996b64fee2 cleanup of revlog.group when repository is local revlog.group cached every chunk from the revlog, the behaviour was needed to minimize the roundtrip with old-http. We now cache the revlog data ~4MB at a time. The memory used server side when pulling goes down to 35Mo maximum whereas without the patch more than 160Mo was used when cloning the linux kernel repository. The time used by cloning is higher mainly because of the check in revlog.revision. before 110.25user 20.90system 2:52.00elapsed 76%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+708707minor)pagefaults 0swaps after 117.56user 18.86system 2:50.43elapsed 80%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (0major+748366minor)pagefaults 0swaps diff -r 96b47ef8f740 -r 14d1f1868bf6 mercurial/revlog.py --- a/mercurial/revlog.py Tue Dec 27 12:52:25 2005 -0600 +++ b/mercurial/revlog.py Tue Dec 27 13:09:49 2005 -0600 @@ -188,6 +188,7 @@ self.datafile = datafile self.opener = opener self.cache = None + self.chunkcache = None try: i = self.opener(self.indexfile).read() @@ -473,6 +474,35 @@ """apply a list of patches to a string""" return mdiff.patches(t, pl) + def chunk(self, rev): + start, length = self.start(rev), self.length(rev) + end = start + length + + def loadcache(): + cache_length = max(4096 * 1024, length) # 4Mo + df = self.opener(self.datafile) + df.seek(start) + self.chunkcache = (start, df.read(cache_length)) + + if not self.chunkcache: + loadcache() + + cache_start = self.chunkcache[0] + cache_end = cache_start + len(self.chunkcache[1]) + if start >= cache_start and end <= cache_end: + # it is cached + offset = start - cache_start + else: + loadcache() + offset = 0 + + #def checkchunk(): + # df = self.opener(self.datafile) + # df.seek(start) + # return df.read(length) + #assert s == checkchunk() + return decompress(self.chunkcache[1][offset:offset + length]) + def delta(self, node): """return or calculate a delta between a node and its predecessor""" r = self.rev(node) @@ -481,10 +511,7 @@ return self.diff(self.revision(self.node(r - 1)), self.revision(node)) else: - f = self.opener(self.datafile) - f.seek(self.start(r)) - data = f.read(self.length(r)) - return decompress(data) + return self.chunk(r) def revision(self, node): """return an uncompressed revision of a given""" @@ -494,33 +521,22 @@ # look up what we need to read text = None rev = self.rev(node) - start, length, base, link, p1, p2, node = self.index[rev] - end = start + length - if base != rev: start = self.start(base) + base = self.base(rev) # do we have useful data cached? if self.cache and self.cache[1] >= base and self.cache[1] < rev: base = self.cache[1] - start = self.start(base + 1) text = self.cache[2] - last = 0 - - f = self.opener(self.datafile) - f.seek(start) - data = f.read(end - start) - - if text is None: - last = self.length(base) - text = decompress(data[:last]) + else: + text = self.chunk(base) bins = [] for r in xrange(base + 1, rev + 1): - s = self.length(r) - bins.append(decompress(data[last:last + s])) - last = last + s + bins.append(self.chunk(r)) text = mdiff.patches(text, bins) + p1, p2 = self.parents(node) if node != hash(text, p1, p2): raise RevlogError(_("integrity check failed on %s:%d") % (self.datafile, rev)) @@ -650,7 +666,7 @@ #print "next x" gx = x.next() - def group(self, nodelist, lookup, infocollect = None): + def group(self, nodelist, lookup, infocollect=None): """calculate a delta group Given a list of changeset revs, return a set of deltas and @@ -660,7 +676,6 @@ changesets. parent is parent[0] """ revs = [self.rev(n) for n in nodelist] - needed = dict.fromkeys(revs, 1) # if we don't have any revisions touched by these changesets, bail if not revs: @@ -671,88 +686,30 @@ p = self.parents(self.node(revs[0]))[0] revs.insert(0, self.rev(p)) - # for each delta that isn't contiguous in the log, we need to - # reconstruct the base, reconstruct the result, and then - # calculate the delta. We also need to do this where we've - # stored a full version and not a delta - for i in xrange(0, len(revs) - 1): - a, b = revs[i], revs[i + 1] - if a + 1 != b or self.base(b) == b: - for j in xrange(self.base(a), a + 1): - needed[j] = 1 - for j in xrange(self.base(b), b + 1): - needed[j] = 1 - - # calculate spans to retrieve from datafile - needed = needed.keys() - needed.sort() - spans = [] - oo = -1 - ol = 0 - for n in needed: - if n < 0: continue - o = self.start(n) - l = self.length(n) - if oo + ol == o: # can we merge with the previous? - nl = spans[-1][2] - nl.append((n, l)) - ol += l - spans[-1] = (oo, ol, nl) - else: - oo = o - ol = l - spans.append((oo, ol, [(n, l)])) - - # read spans in, divide up chunks - chunks = {} - for span in spans: - # we reopen the file for each span to make http happy for now - f = self.opener(self.datafile) - f.seek(span[0]) - data = f.read(span[1]) - - # divide up the span - pos = 0 - for r, l in span[2]: - chunks[r] = decompress(data[pos: pos + l]) - pos += l - # helper to reconstruct intermediate versions def construct(text, base, rev): - bins = [chunks[r] for r in xrange(base + 1, rev + 1)] + bins = [self.chunk(r) for r in xrange(base + 1, rev + 1)] return mdiff.patches(text, bins) # build deltas - deltas = [] for d in xrange(0, len(revs) - 1): a, b = revs[d], revs[d + 1] - n = self.node(b) + na = self.node(a) + nb = self.node(b) if infocollect is not None: - infocollect(n) + infocollect(nb) # do we need to construct a new delta? if a + 1 != b or self.base(b) == b: - if a >= 0: - base = self.base(a) - ta = chunks[self.base(a)] - ta = construct(ta, base, a) - else: - ta = "" - - base = self.base(b) - if a > base: - base = a - tb = ta - else: - tb = chunks[self.base(b)] - tb = construct(tb, base, b) + ta = self.revision(na) + tb = self.revision(nb) d = self.diff(ta, tb) else: - d = chunks[b] + d = self.chunk(b) - p = self.parents(n) - meta = n + p[0] + p[1] + lookup(n) + p = self.parents(nb) + meta = nb + p[0] + p[1] + lookup(nb) l = struct.pack(">l", len(meta) + len(d) + 4) yield l yield meta diff -r 96b47ef8f740 -r 14d1f1868bf6 mercurial/statichttprepo.py --- a/mercurial/statichttprepo.py Tue Dec 27 12:52:25 2005 -0600 +++ b/mercurial/statichttprepo.py Tue Dec 27 13:09:49 2005 -0600 @@ -35,6 +35,8 @@ self.changelog = changelog.changelog(self.opener) self.tagscache = None self.nodetagscache = None + self.encodepats = None + self.decodepats = None def dev(self): return -1