snapshot: search for unrelated but reusable full-snapshot
# New Strategy Step: Reusing Snapshot Outside Of Parents' Chain.
If no suitable bases were found in the parent's chains, see if we could reuse
a full snapshot not directly related to the current revision. Such search can
be expensive, so we only search for snapshots appended to the revlog *after*
the bases used by the parents of the current revision (the one we just
tested). We assume the parent's bases were created because the previous
snapshots were unsuitable, so there are low odds they would be useful now.
This search gives a chance to reuse a delta chain unrelated to the current
revision. Without this re-use, topological branches would keep reopening new
full chains. Creating more and more snapshots as the repository grow.
In repositories with many topological branches, the lack of delta reuse can
create too many snapshots reducing overall compression to nothing. This
results in a very large repository and other usability issues.
For now, we still focus on creating level-1 snapshots. However, this principle
will play a large part in how we avoid snapshot explosion once we have more
snapshot levels.
# Effects On The Test Repository
In the test repository we created, we can see the beneficial effect of such
reuse. We need very few level-0 snapshots and the overall revlog size has
decreased.
The `hg debugrevlog` call, show a "lvl-2" snapshot. It comes from the existing
delta logic using the `prev` revision (revlog's tip) as the base. In this
specific case, it turns out the tip was a level-1 snapshot. This is a
coincidence that can be ignored.
Finding and testing against all these unrelated snapshots can have a
performance impact at write time. We currently focus on building good deltas
chain we build. Performance concern will be dealt with later in another
series.
--- a/mercurial/revlogutils/deltas.py Fri Sep 07 11:17:29 2018 -0400
+++ b/mercurial/revlogutils/deltas.py Fri Sep 07 11:17:30 2018 -0400
@@ -9,6 +9,7 @@
from __future__ import absolute_import
+import collections
import heapq
import struct
@@ -607,8 +608,18 @@
continue
group.append(rev)
if group:
+ # XXX: in the sparse revlog case, group can become large,
+ # impacting performances. Some bounding or slicing mecanism
+ # would help to reduce this impact.
yield tuple(group)
+def _findsnapshots(revlog, cache, start_rev):
+ """find snapshot from start_rev to tip"""
+ deltaparent = revlog.deltaparent
+ for rev in revlog.revs(start_rev):
+ if deltaparent(rev) == nullrev:
+ cache[nullrev].append(rev)
+
def _rawgroups(revlog, p1, p2, cachedelta):
"""Provides group of revision to be tested as delta base
@@ -656,6 +667,18 @@
for p in parents:
bases.append(deltachain(p)[0])
yield tuple(sorted(bases))
+ # No suitable base found in the parent chain, search if any full
+ # snapshots emitted since parent's base would be a suitable base for an
+ # intermediate snapshot.
+ #
+ # It give a chance to reuse a delta chain unrelated to the current
+ # revisions instead of starting our own. Without such re-use,
+ # topological branches would keep reopening new full chains. Creating
+ # more and more snapshot as the repository grow.
+ snapfloor = min(bases) + 1
+ snapshots = collections.defaultdict(list)
+ _findsnapshots(revlog, snapshots, snapfloor)
+ yield tuple(snapshots[nullrev])
# other approach failed try against prev to hopefully save us a
# fulltext.
--- a/tests/test-sparse-revlog.t Fri Sep 07 11:17:29 2018 -0400
+++ b/tests/test-sparse-revlog.t Fri Sep 07 11:17:30 2018 -0400
@@ -77,7 +77,7 @@
$ f -s .hg/store/data/*.d
- .hg/store/data/_s_p_a_r_s_e-_r_e_v_l_o_g-_t_e_s_t-_f_i_l_e.d: size=72315280
+ .hg/store/data/_s_p_a_r_s_e-_r_e_v_l_o_g-_t_e_s_t-_f_i_l_e.d: size=67810463
$ hg debugrevlog *
format : 1
flags : generaldelta
@@ -89,36 +89,39 @@
empty : 0 ( 0.00%)
text : 0 (100.00%)
delta : 0 (100.00%)
- snapshot : 145 ( 2.90%)
- lvl-0 : 15 ( 0.30%)
- lvl-1 : 130 ( 2.60%)
- deltas : 4856 (97.10%)
- revision size : 72315280
- snapshot : 18481085 (25.56%)
- lvl-0 : 3016019 ( 4.17%)
- lvl-1 : 15465066 (21.39%)
- deltas : 53834195 (74.44%)
+ snapshot : 126 ( 2.52%)
+ lvl-0 : 4 ( 0.08%)
+ lvl-1 : 120 ( 2.40%)
+ lvl-2 : 2 ( 0.04%)
+ deltas : 4875 (97.48%)
+ revision size : 67810463
+ snapshot : 14373347 (21.20%)
+ lvl-0 : 804235 ( 1.19%)
+ lvl-1 : 13535903 (19.96%)
+ lvl-2 : 33209 ( 0.05%)
+ deltas : 53437116 (78.80%)
chunks : 5001
0x78 (x) : 5001 (100.00%)
- chunks size : 72315280
- 0x78 (x) : 72315280 (100.00%)
+ chunks size : 67810463
+ 0x78 (x) : 67810463 (100.00%)
avg chain length : 18
max chain length : 45
- max chain reach : 32095083
- compression ratio : 23
+ max chain reach : 25808240
+ compression ratio : 25
uncompressed data size (min/max/avg) : 346468 / 346472 / 346471
- full revision size (min/max/avg) : 200990 / 201151 / 201067
- inter-snapshot size (min/max/avg) : 37202 / 173034 / 118962
- level-1 (min/max/avg) : 37202 / 173034 / 118962
- delta size (min/max/avg) : 10649 / 104791 / 11086
+ full revision size (min/max/avg) : 201014 / 201116 / 201058
+ inter-snapshot size (min/max/avg) : 11623 / 173150 / 111222
+ level-1 (min/max/avg) : 11623 / 173150 / 112799
+ level-2 (min/max/avg) : 14151 / 19058 / 16604
+ delta size (min/max/avg) : 10649 / 101790 / 10961
- deltas against prev : 4185 (86.18%)
- where prev = p1 : 4139 (98.90%)
+ deltas against prev : 4207 (86.30%)
+ where prev = p1 : 4164 (98.98%)
where prev = p2 : 0 ( 0.00%)
- other : 46 ( 1.10%)
- deltas against p1 : 647 (13.32%)
- deltas against p2 : 24 ( 0.49%)
+ other : 43 ( 1.02%)
+ deltas against p1 : 653 (13.39%)
+ deltas against p2 : 15 ( 0.31%)
deltas against other : 0 ( 0.00%)