The general delta heuristic to select a delta do not scale with the number of
branch. The delta base is frequently too far away to be able to reuse a chain
according to the "distance" criteria. This leads to insertion of larger delta (or
even full text) that themselves push the bases for the next delta further away
leading to more large deltas and full texts. This full text and frequent
recomputation throw Mercurial performance in disarray.
For example of a slightly large repository
280 000 files (2 150 000 versions)
430 000 changesets (10 000 topological heads)
Number below compares repository with and without the distance criteria:
manifest size:
with: 21.4 GB
without: 0.3 GB
store size:
with: 28.7 GB
without 7.4 GB
bundle last 15 00 revisions:
with: 800 seconds
971 MB
without: 50 seconds
73 MB
unbundle time (of the last 15K revisions):
with: 1150 seconds (~19 minutes)
without: 35 seconds
Similar issues has been observed in other repositories.
Adding a new option or "feature" on stable is uncommon. However, given that this
issues is making Mercurial practically unusable, I'm exceptionally targeting
this patch for stable.
What is actually needed is a full rework of the delta building and reading
logic. However, that will be a longer process and churn not suitable for stable.
In the meantime, we introduces a quick and dirty mitigation of this in the
'experimental' config space. The new option introduces a way to set the maximum
amount of memory usable to store a diff in memory. This extend the ability for
Mercurial to create chains without removing all safe guard regarding memory
access. The option should be phased out when core has a more proper solution
available.
Setting the limit to '0' remove all limits, setting it to '-1' use the default
limit (textsize x 4).
I've seen revlog._deltachain() appear in a number of performance
profiles. I suspect there are 2 reasons for this:
1. Delta chain resolution performs many index lookups, thus triggering
population of index tuples. Creating possibly tens of thousands of
PyObject will have overhead.
2. Delta chain resolution is a tight loop.
By moving delta chain resolution to C, we can defer instantiation
of full index entry tuples and make the loop faster courtesy of
not running in Python.
We can measure the impact to delta chain resolution via
`hg perflogrevision` using the mozilla-central repo with a recent
manifest having delta chain length of 33726:
$ hg perfrevlogrevision -m 364895
! full
! wall 0.367585 comb 0.370000 user 0.340000 sys 0.030000 (best of 27)
! wall 0.357581 comb 0.360000 user 0.350000 sys 0.010000 (best of 28)
! deltachain
! wall 0.010644 comb 0.010000 user 0.010000 sys 0.000000 (best of 270)
! wall 0.000292 comb 0.000000 user 0.000000 sys 0.000000 (best of 8729)
$ hg perfrevlogrevision --cache -m 364895
! deltachain
! wall 0.003904 comb 0.000000 user 0.000000 sys 0.000000 (best of 712)
! wall 0.000284 comb 0.000000 user 0.000000 sys 0.000000 (best of 9926)
The first test measures savings from both not instantiating index
entries and moving to C. The second test (which doesn't clear the
index caches) essentially isolates the benefits of moving from Python
to C. It still shows a 13.7x speedup (versus 36.4x). And there are
multiple milliseconds of savings within the critical path for resolving
revision data. I think that justifies the existence of C code.
A more striking example of the benefits of this change can be
demonstrated by timing `hg debugdeltachain -m` for the mozilla-central
repo:
$ time hg debugdeltachain -m > /dev/null
before: 1057.4s
after: 503.3s
PyPy2.7 5.8.0: 220.0s
It's worth noting that the C code isn't as optimal as it could be.
We're still instantiating a new PyObject for every revision. A future
optimization would be to reuse the PyObject on the cached index tuple.
We could potentially also get wins by using a memory array of raw
integers. There is also room for a delta chain cache on revlog
instances. Of course, the best optimization is to implement revlog
reading outside of Python so Python doesn't need to be concerned
about the relatively expensive index entries and operations on them.
It seems like the reason for "content" is that the variable contains
the nodes that the changegroup "contains", see c2901cddb53f (revlog:
make addgroup returns a list of node contained in the added source,
2012-01-13), but "nodes" seems much clearer.
The comment seems to refer to code that was deleted in a46638c640d8
(revlog.addgroup(): always use _addrevision() to add new revlog
entries, 2010-10-08).
There are a number of improvements we want to make to revlogs
that will require a new version - version 2. It is unclear what the
full set of improvements will be or when we'll be done with them.
What I do know is that the process will likely take longer than a
single release, will require input from various stakeholders to
evaluate changes, and will have many contentious debates and
bikeshedding.
It is unrealistic to develop revlog version 2 up front: there
are just too many uncertainties that we won't know until things
are implemented and experiments are run. Some changes will also
be invasive and prone to bit rot, so sitting on dozens of patches
is not practical.
This commit introduces skeleton support for version 2 revlogs in
a way that is flexible and not bound by backwards compatibility
concerns.
An experimental repo requirement for denoting revlog v2 has been
added. The requirement string has a sub-version component to it.
This will allow us to declare multiple requirements in the course
of developing revlog v2. Whenever we change the in-development
revlog v2 format, we can tweak the string, creating a new
requirement and locking out old clients. This will allow us to
make as many backwards incompatible changes and experiments to
revlog v2 as we want. In other words, we can land code and make
meaningful progress towards revlog v2 while still maintaining
extreme format flexibility up until the point we freeze the
format and remove the experimental labels.
To enable the new repo requirement, you must supply an experimental
and undocumented config option. But not just any boolean flag
will do: you need to explicitly use a value that no sane person
should ever type. This is an additional guard against enabling
revlog v2 on an installation it shouldn't be enabled on. The
specific scenario I'm trying to prevent is say a user with a
4.4 client with a frozen format enabling the option but then
downgrading to 4.3 and accidentally creating repos with an
outdated and unsupported repo format. Requiring a "challenge"
string should prevent this.
Because the format is not yet finalized and I don't want to take
any chances, revlog v2's version is currently 0xDEAD. I figure
squatting on a value we're likely never to use as an actual revlog
version to mean "internal testing only" is acceptable. And
"dead" is easily recognized as something meaningful.
There is a bunch of cleanup that is needed before work on revlog
v2 begins in earnest. I plan on doing that work once this patch
is accepted and we're comfortable with the idea of starting down
this path.
The idea is simple. If the given node id prefix is 'ff...f', add +1 to the
number of matches (e.g. ambiguous if partial + maybewdir > 1).
This patch also fixes id() revset and shortest() template since _partialmatch()
can raise WdirUnsupported exception.
This will allow us to map repo["ff..."] to workingctx. _partialmatch() will
be updated later. I tried "return wdirrev" in place of raising the exception,
but earlier exception seemed better.
When we try to run, 'hg debugrevspec 'branch(wdir())'', it throws an index error
and blows up. Lets raise the WdirUnsupported if wdir() is passed so that we can
catch that later.
revlog.parentrevs() is called while evaluating ^ operator in revsets. When wdir
is passed, it raises IndexError. This patch raises WdirUnsupported if wdir is
passed in the function. The error will be caugth in future patches.
RevlogNG is not such a good name when it is no longer the
newest revlog version. Since we'll soon have revlog version 2,
let's remove some references to it.
First, the logic around the if..elif..elif was subtly wrong
and sub-optimal because all branches would be tested as long as
the revlog was valid. This patch changes things so it behaves like
a switch statement over the revlog version.
While I was here, I also tweaked error strings to make them
consistent and to read better.
Feature flag constants don't need "NG" in the name because they will
presumably apply to non-"NG" version revlogs.
All feature flag constants should also share a similar naming
convention to identify them as such.
And, "RevlogNG" isn't a great internal name since it isn't obvious it
maps to version 1 revlogs. Plus, "NG" (next generation) is only a good
name as long as it is the latest version. Since we're talking about
version 2, now is as good a time as any to move on from that naming.
This completes our rename of internal revlog methods to
distinguish between low-level raw revlog data "segments" and
higher-level, per-revision "chunks."
perf.py has been updated to consult both names so it will work
against older Mercurial versions.
Currently, "chunk" is overloaded in revlog terminology to mean
multiple things. One of them refers to a segment of raw data from
the revlog. This commit renames various methods only used within
revlog.py to have "segment" in their name instead of "chunk."
While I was here, I also made the names more descriptive. e.g.
"_loadchunk()" becomes "_readsegment()" because it actually does
I/O.
Previously, revlog.size equals to revlog.rawsize. However, the flag
processor framework could make a difference - "size" could mean the length
of len(revision(raw=False)), while "rawsize" means len(revision(raw=True)).
This patch makes it so.
This corrects "hg status" output when flag processor is involved. The call
stack looks like:
basectx.status -> workingctx._buildstatus -> workingctx._dirstatestatus
-> workingctx._checklookup -> filectx.cmp -> filelog.cmp -> filelog.size
-> revlog.size
Previously, revlog.revision(raw=False) may try to apply the delta chain
on _cache hit. That happens if flags are non-empty. This patch makes rawtext
reused so delta chain application is avoided.
"_cache" and "rev" are moved a bit to avoid unnecessary assignments.
When writing the revlog-ng index, the third field is len(rawtext). See
revlog._addrevision:
textlen = len(rawtext)
....
e = (offset_type(offset, flags), l, textlen,
base, link, p1r, p2r, node)
self.index.insert(-1, e)
Therefore, revlog.index[rev][2] returned by revlog.rawsize should be
len(rawtext), where "rawtext" is revlog.revision(raw=True).
Unfortunately it's hard to add a test for this code path because "if l >= 0"
catches most cases.
If cache hit and flags are empty, no flag processor runs and "text" equals
to "rawtext". So we check flags, and return rawtext.
This resolves performance issue introduced by a previous patch.
All 3 users of _addrevision use raw:
- addrevision: passing rawtext to _addrevision
- addgroup: passing rawtext and raw=True to _addrevision
- clone: passing rawtext to _addrevision
There is no real user using _addrevision(raw=False). On the other hand,
_addrevision is low-level code dealing with raw revlog deltas and rawtexts.
It should not transform rawtext to non-raw text.
This patch removes the "raw" parameter from "_addrevision", and does some
rename and doc change to make it clear that "_addrevision" expects rawtext.
Archeology shows 886a08012bbe added "raw" flag to "_addrevision", follow-ups
fe1e206cb389 and 1cfa6239c923 seem to make the flag unnecessary.
test-revlog-raw.py no longer complains.
See the added comment. revdiff is meant to output the raw delta that will be
written to revlog. It should use raw.
test-revlog-raw.py now shows "addgroupcopy test passed", but there is more
to fix.
Using external content provided by flagprocessor when building revlog delta
is wrong, because deltas are applied to raw contents in revlog.
This patch fixes the above issue by adding "raw=True".
test-revlog-raw.py now shows "local test passed", but there is more to fix.
As documented at revlog.__init__, revlog._cache stores raw text.
The current read and write usage of "_cache" in revlog.revision lacks of
raw=True check.
This patch fixes that by adding check about raw, and storing rawtext
explicitly in _cache.
Note: it may slow down cache hit code path when raw=False and flags=0. That
performance issue will be fixed in a later patch.
test-revlog-raw now points us to a new problem.
Narrowhg has been using "1 << 14" as its revlog flag value for a long
time. We (Google) have many repos with that value in production
already. When the same value was reserved for EXTSTORED, it made those
repos invalid. Upgrading them will be a little painful. We should
clearly have reserved the value for narrowhg a long time ago. Since
the EXTSTORED flag is not yet in any release and Facebook also says
they have not started using it in production, so it should be okay to
change it. This patch gives the current value (1 << 14) back to
narrowhg and gives a new value (1 << 13) to EXTSTORED.
The final part of integrating the compression manager APIs into
revlog storage is the plumbing for repositories to advertise they
are using non-zlib storage and for revlogs to instantiate a non-zlib
compression engine.
The main intent of the compression manager work was to zstd all
of the things. Adding zstd to revlogs has proved to be more involved
than other places because revlogs are... special. Very small inputs
and the use of delta chains (which are themselves a form of
compression) are a completely different use case from streaming
compression, which bundles and the wire protocol employ. I've
conducted numerous experiments with zstd in revlogs and have yet
to formalize compression settings and a storage architecture that
I'm confident I won't regret later. In other words, I'm not yet
ready to commit to a new mechanism for using zstd - or any other
compression format - in revlogs.
That being said, having some support for zstd (and other compression
formats) in revlogs in core is beneficial. It can allow others to
conduct experiments.
This patch introduces *highly experimental* support for non-zlib
compression formats in revlogs. Introduced is a config option to
control which compression engine to use. Also introduced is a namespace
of "exp-compression-*" requirements to denote support for non-zlib
compression in revlogs. I've prefixed the namespace with "exp-"
(short for "experimental") because I'm not confident of the
requirements "schema" and in no way want to give the illusion of
supporting these requirements in the future. I fully intend to drop
support for these requirements once we figure out what we're doing
with zstd in revlogs.
A good portion of the patch is teaching the requirements system
about registered compression engines and passing the requested
compression engine as an opener option so revlogs can instantiate
the proper compression engine for new operations.
That's a verbose way of saying "we can now use zstd in revlogs!"
On an `hg pull` conversion of the mozilla-unified repo with no extra
redelta settings (like aggressivemergedeltas), we can see the impact
of zstd vs zlib in revlogs:
$ hg perfrevlogchunks -c
! chunk
! wall 2.032052 comb 2.040000 user 1.990000 sys 0.050000 (best of 5)
! wall 1.866360 comb 1.860000 user 1.820000 sys 0.040000 (best of 6)
! chunk batch
! wall 1.877261 comb 1.870000 user 1.860000 sys 0.010000 (best of 6)
! wall 1.705410 comb 1.710000 user 1.690000 sys 0.020000 (best of 6)
$ hg perfrevlogchunks -m
! chunk
! wall 2.721427 comb 2.720000 user 2.640000 sys 0.080000 (best of 4)
! wall 2.035076 comb 2.030000 user 1.950000 sys 0.080000 (best of 5)
! chunk batch
! wall 2.614561 comb 2.620000 user 2.580000 sys 0.040000 (best of 4)
! wall 1.910252 comb 1.910000 user 1.880000 sys 0.030000 (best of 6)
$ hg perfrevlog -c -d 1
! wall 4.812885 comb 4.820000 user 4.800000 sys 0.020000 (best of 3)
! wall 4.699621 comb 4.710000 user 4.700000 sys 0.010000 (best of 3)
$ hg perfrevlog -m -d 1000
! wall 34.252800 comb 34.250000 user 33.730000 sys 0.520000 (best of 3)
! wall 24.094999 comb 24.090000 user 23.320000 sys 0.770000 (best of 3)
Only modest wins for the changelog. But manifest reading is
significantly faster. What's going on?
One reason might be data volume. zstd decompresses faster. So given
more bytes, it will put more distance between it and zlib.
Another reason is size. In the current design, zstd revlogs are
*larger*:
debugcreatestreamclonebundle (size in bytes)
zlib: 1,638,852,492
zstd: 1,680,601,332
I haven't investigated this fully, but I reckon a significant cause of
larger revlogs is that the zstd frame/header has more bytes than
zlib's. For very small inputs or data that doesn't compress well, we'll
tend to store more uncompressed chunks than with zlib (because the
compressed size isn't smaller than original). This will make revlog
reading faster because it is doing less decompression.
Moving on to bundle performance:
$ hg bundle -a -t none-v2 (total CPU time)
zlib: 102.79s
zstd: 97.75s
So, marginal CPU decrease for reading all chunks in all revlogs
(this is somewhat disappointing).
$ hg bundle -a -t <engine>-v2 (total CPU time)
zlib: 191.59s
zstd: 115.36s
This last test effectively measures the difference between zlib->zlib
and zstd->zstd for revlogs to bundle. This is a rough approximation of
what a server does during `hg clone`.
There are some promising results for zstd. But not enough for me to
feel comfortable advertising it to users. We'll get there...
Now that compression engines declare their header in revlog chunks
and can decompress revlog chunks, we refactor revlog.decompress()
to use them.
Making full use of the property that revlog compressor objects are
reusable, revlog instances now maintain a dict mapping an engine's
revlog header to a compressor object. This is not only a performance
optimization for engines where compressor object reuse can result in
better performance, but it also serves as a cache of header values
so we don't need to perform redundant lookups against the compression
engine manager. (Yes, I measured and the overhead of a function call
versus a dict lookup was observed.)
Replacing the previous inline lookup table with a dict lookup was
measured to make chunk reading ~2.5% slower on changelogs and ~4.5%
slower on manifests. So, the inline lookup table has been mostly
preserved so we don't lose performance. This is unfortunate. But
many decompression operations complete in microseconds, so Python
attribute lookup, dict lookup, and function calls do matter.
The impact of this change on mozilla-unified is as follows:
$ hg perfrevlogchunks -c
! chunk
! wall 1.953663 comb 1.950000 user 1.920000 sys 0.030000 (best of 6)
! wall 1.946000 comb 1.940000 user 1.910000 sys 0.030000 (best of 6)
! chunk batch
! wall 1.791075 comb 1.800000 user 1.760000 sys 0.040000 (best of 6)
! wall 1.785690 comb 1.770000 user 1.750000 sys 0.020000 (best of 6)
$ hg perfrevlogchunks -m
! chunk
! wall 2.587262 comb 2.580000 user 2.550000 sys 0.030000 (best of 4)
! wall 2.616330 comb 2.610000 user 2.560000 sys 0.050000 (best of 4)
! chunk batch
! wall 2.427092 comb 2.420000 user 2.400000 sys 0.020000 (best of 5)
! wall 2.462061 comb 2.460000 user 2.400000 sys 0.060000 (best of 4)
Changelog chunk reading is slightly faster but manifest reading is
slower. What gives?
On this repo, 99.85% of changelog entries are zlib compressed (the 'x'
header). On the manifest, 67.5% are zlib and 32.4% are '\0'. This patch
swapped the test order of 'x' and '\0' so now 'x' is tested first. This
makes changelogs faster since they almost always hit the first branch.
This makes a significant percentage of manifest '\0' chunks slower
because that code path now performs an extra test. Yes, I too can't
believe we're able to measure the impact of an if..elif with simple
string compares. I reckon this code would benefit from being written
in C...
This commit swaps in the just-added revlog compressor API into
the revlog class.
Instead of implementing zlib compression inline in compress(), we
now store a cached-on-first-use revlog compressor on each revlog
instance and invoke its "compress()" method.
As part of this, revlog.compress() has been refactored a bit to use
a cleaner code flow and modern formatting (e.g. avoiding
parenthesis around returned tuples).
On a mozilla-unified repo, here are the "compress" times for a few
commands:
$ hg perfrevlogchunks -c
! wall 5.772450 comb 5.780000 user 5.780000 sys 0.000000 (best of 3)
! wall 5.795158 comb 5.790000 user 5.790000 sys 0.000000 (best of 3)
$ hg perfrevlogchunks -m
! wall 9.975789 comb 9.970000 user 9.970000 sys 0.000000 (best of 3)
! wall 10.019505 comb 10.010000 user 10.010000 sys 0.000000 (best of 3)
Compression times did seem to slow down just a little. There are
360,210 changelog revisions and 359,342 manifest revisions. For the
changelog, mean time to compress a revision increased from ~16.025us to
~16.088us. That's basically a function call or an attribute lookup. I
suppose this is the price you pay for abstraction. It's so low that
I'm not concerned.
Upcoming patches will convert revlogs to use the compression engine
APIs to perform all things compression. The yet-to-be-introduced
APIs support a persistent "compressor" object so the same object
can be reused for multiple compression operations, leading to
better performance. In addition, compression engines like zstd
may wish to tweak compression engine state based on the revlog
(e.g. per-revlog compression dictionaries).
A global and shared decompress() function will shortly no longer
make much sense. So, we move decompress() to be a method of the
revlog class. It joins compress() there.
On the mozilla-unified repo, we can measure the impact of this change
on reading performance:
$ hg perfrevlogchunks -c
! chunk
! wall 1.932573 comb 1.930000 user 1.900000 sys 0.030000 (best of 6)
! wall 1.955183 comb 1.960000 user 1.930000 sys 0.030000 (best of 6)
! chunk batch
! wall 1.787879 comb 1.780000 user 1.770000 sys 0.010000 (best of 6
! wall 1.774444 comb 1.770000 user 1.750000 sys 0.020000 (best of 6)
"chunk" appeared to become slower but "chunk batch" got faster. Upon
further examination by running both sets multiple times, the numbers
appear to converge across all runs. This tells me that there is no
perceived performance impact to this refactor.
revlog.compress() compares the compressed size to the input size
and throws away the compressed data if it is larger than the input.
This is the correct thing to do, as storing compressed data that
is larger than the input takes up more storage space and makes reading
slower.
However, the comparison was implemented inconsistently. For the
streaming compression mode, we threw away the result if it was
greater than or equal to the input size. But for the one-shot
compression, we threw away the compression only if it was greater
than the input size!
This patch changes the comparison for the simple case so it is
consistent with the streaming case.
As a few tests demonstrate, this adds 1 byte to some revlog entries.
This is because of an added 'u' header on the chunk. It seems
somewhat wrong to increase the revlog size here. However, IMO the cost
of 1 byte in storage is insignificant compared to the performance gains
of avoiding decompression. This patch should invite questions around
the heuristic for throwing away compressed data. For example, I'd argue
we should be more liberal about rejecting compressed data, additionally
doing so where the number of bytes saved fails to reach a threshold.
But we can have this discussion another time.
Upcoming patches will introduce functionality for in-place
repository/store "upgrades." Copying the contents of a revlog
feels sufficiently low-level to warrant being in the revlog
class. So this commit implements that functionality.
Because full delta recomputation can be *very* expensive (we're
talking several hours on the Firefox repository), we support
multiple modes of execution with regards to delta (re)use. This
will allow repository upgrades to choose the "level" of
processing/optimization they wish to perform when converting
revlogs.
It's not obvious from this commit, but "addrevisioncb" will be
used for progress reporting.
