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.
Add the ability for revlog objects to process revision flags and apply
registered transforms on read/write operations.
This patch introduces:
- the 'revlog._processflags()' method that looks at revision flags and applies
flag processors registered on them. Due to the need to handle non-commutative
operations, flag transforms are applied in stable order but the order in which
the transforms are applied is reversed between read and write operations.
- the 'addflagprocessor()' method allowing to register processors on flags.
Flag processors are defined as a 3-tuple of (read, write, raw) functions to be
applied depending on the operation being performed.
- an update on 'revlog.addrevision()' behavior. The current flagprocessor design
relies on extensions to wrap around 'addrevision()' to set flags on revision
data, and on the flagprocessor to perform the actual transformation of its
contents. In the lfs case, this means we need to process flags before we meet
the 2GB size check, leading to performing some operations before it happens:
- if flags are set on the revision data, we assume some extensions might be
modifying the contents using the flag processor next, and we compute the
node for the original revision data (still allowing extension to override
the node by wrapping around 'addrevision()').
- we then invoke the flag processor to apply registered transforms (in lfs's
case, drastically reducing the size of large blobs).
- finally, we proceed with the 2GB size check.
Note: In the case a cachedelta is passed to 'addrevision()' and we detect the
flag processor modified the revision data, we chose to trust the flag processor
and drop the cachedelta.
Adding the ability to passing flags to addrevision instead of simply passing
default flags to _addrevision will allow extensions relying on flag transforms
to wrap around addrevision() in order to update revlog flags.
The first use case of this patch will be the lfs extension marking nodes as
stored externally when the contents are larger than the defined threshold.
One of the reasons leading to setting flags in addrevision() wrappers in the
flag processor design is that it allows to detect files larger than the 2GB
limit before the check is performed, which allows lfs to transform the contents
into metadata.
This patch introduces a new 'raw' argument (defaults to False) to revlog's
revision() and _addrevision() methods.
When the 'raw' argument is set to True, it indicates the revision data should be
handled as raw data by the flagprocessor.
Note: Given revlog.addgroup() calls are restricted to changegroup generation, we
can always set raw to True when calling revlog._addrevision() from
revlog.addgroup().
This patch adds a line that ensures we are not setting by mistake a set of flags
overlfowing the 2 bytes they are allocated. Given the way the data is packed in
the revlog header, overflowing 2 bytes will result in setting a wrong offset.
In many cases, _chunkraw() is called with startrev==endrev. When
this is true, we can avoid an extra index lookup and some other
minor operations.
On the mozilla-unified repo, `hg perfrevlogchunks -c` says this
has the following impact:
! read w/ reused fd
! wall 0.371846 comb 0.370000 user 0.350000 sys 0.020000 (best of 27)
! wall 0.337930 comb 0.330000 user 0.300000 sys 0.030000 (best of 30)
! read batch w/ reused fd
! wall 0.014952 comb 0.020000 user 0.000000 sys 0.020000 (best of 197)
! wall 0.014866 comb 0.010000 user 0.000000 sys 0.010000 (best of 196)
So, we've gone from ~25x slower than batch to ~22.5x slower.
At this point, there's probably not much else we can do except
implement an optimized function in the index itself, including in C.
When I implemented `hg perfrevlogchunks`, one of the things that
stood out was N * _chunk() calls was ~38x slower than 1
_chunks() call. Specifically, on the mozilla-unified repo:
N*_chunk: 0.528997s
1*_chunks: 0.013735s
This repo has 352,097 changesets. So the average time per changeset
comes out to:
N*_chunk: 1.502us
1*_chunks: 0.039us
If you extrapolate these numbers to a repository with 1M changesets,
that comes out to 1.502s versus 0.039s, which is significant.
At these latencies, Python attribute lookups and function calls
matter. So, this patch inlines some code to cut down on that overhead.
The impact of this patch on N*_chunk() calls is clear:
! wall 0.528997 comb 0.520000 user 0.500000 sys 0.020000 (best of 19)
! wall 0.367723 comb 0.370000 user 0.350000 sys 0.020000 (best of 27)
So, we go from ~38x slower to ~27x. A nice improvement. But there's
still a long way to go.
It's worth noting that functionality like revsets perform changelog
lookups one revision at a time. So this code path is worth optimizing.
Index entries are ordered tuples. We have accessors in the revlog
class to map tuple offsets to names. To help reinforce the order,
reorder the methods so they match the order of elements in the
tuple. While I'm here, also sneak in some minimal documentation.
This allows revlog-style files to be written out with checkambig=True
easily.
Because avoiding file stat ambiguity is needed only for filecache-ed
manifest and changelog, this patch does:
- use False for default value of checkambig
- focus only on writing changes of index file out
This patch also adds optional argument checkambig to _divert/_delay
for changelog, to safely accept checkambig specified in revlog
layer. But this argument can be fully ignored, because:
- changes are written into other than index file, if name != target
- changes are never written into index file, otherwise
(into pending file by _divert, or into in-memory buffer by _delay)
This is a part of ExactCacheValidationPlan.
https://www.mercurial-scm.org/wiki/ExactCacheValidationPlan
Profiling using statprof revealed a hotspot during changegroup
application calculating delta chain bases on generaldelta repos.
Essentially, revlog._addrevision() was performing a lot of redundant
work tracing the delta chain as part of determining when the chain
distance was acceptable. This was most pronounced when adding
revisions to manifests, which can have delta chains thousands of
revisions long.
There was a delta chain base cache on revlogs before, but it only
captured a single revision. This was acceptable before generaldelta,
when _addrevision would build deltas from the previous revision and
thus we'd pretty much guarantee a cache hit when resolving the delta
chain base on a subsequent _addrevision call. However, it isn't
suitable for generaldelta because parent revisions aren't necessarily
the last processed revision.
This patch converts the delta chain base cache to an LRU dict cache.
The cache can hold multiple entries, so generaldelta repos have a
higher chance of getting a cache hit.
The impact of this change when processing changegroup additions is
significant. On a generaldelta conversion of the "mozilla-unified"
repo (which contains heads of the main Firefox repositories in
chronological order - this means there are lots of transitions between
heads in revlog order), this change has the following impact when
performing an `hg unbundle` of an uncompressed bundle of the repo:
before: 5:42 CPU time
after: 4:34 CPU time
Most of this time is saved when applying the changelog and manifest
revlogs:
before: 2:30 CPU time
after: 1:17 CPU time
That nearly a 50% reduction in CPU time applying changesets and
manifests!
Applying a gzipped bundle of the same repo (effectively simulating a
`hg clone` over HTTP) showed a similar speedup:
before: 5:53 CPU time
after: 4:46 CPU time
Wall time improvements were basically the same as CPU time.
I didn't measure explicitly, but it feels like most of the time
is saved when processing manifests. This makes sense, as large
manifests tend to have very long delta chains and thus benefit the
most from this cache.
So, this change effectively makes changegroup application (which is
used by `hg unbundle`, `hg clone`, `hg pull`, `hg unshelve`, and
various other commands) significantly faster when delta chains are
long (which can happen on repos with large numbers of files and thus
large manifests).
In theory, this change can result in more memory utilization. However,
we're caching a dict of ints. At most we have 200 ints + Python object
overhead per revlog. And, the cache is really only populated when
performing read-heavy operations, such as adding changegroups or
scanning an individual revlog. For memory bloat to be an issue, we'd
need to scan/read several revisions from several revlogs all while
having active references to several revlogs. I don't think there are
many operations that do this, so I don't think memory bloat from the
cache will be an issue.
Before fd1bb7c, if the C index.partialmatch raises RevlogError, the Python
code raises "ambiguous identifier" error immediately, which is efficient.
fd1bb7c took hidden revisions into consideration and forced the slow path
enumerating the changelog to double-check hidden revisions. But it's not
necessary if we know the revlog has no hidden revisions.
This patch adds back the fast path for unfiltered revlogs.
Previously, we likely called _chunkraw() multiple times in order to
ensure it didn't change out from under us. I'm pretty certain this code
had its origins in the days where we attempted to have thread safety of
localrepository and thus revlog instances.
revlog instances are already not thread safe for writing. And, as of
Mercurial 3.6, hgweb uses a separate localrepository instance per
request, so there should only be a single thread reading a revlog at
a time. We more or less decided that attempting to make classes like
revlog thread safe is a lost cause.
So, this patch removes thread safety from _chunks. As a result, we make
one less call into _chunkraw() when the initial read isn't serviced
by the cache. This translates to savings of 4 function calls overall
and possibly prevents the creation of an additional buffer view into the
cache. I doubt this translates into any real world performance wins
because decompression will almost certainly dwarf time spent in
_chunks(). But it does make the code simpler, so it is an improvement.
A subsequent patch will refactor _chunks() and the calculation of the
offset will no longer occur in that function. Prepare by returning the
offset from _chunkraw().
Revlog chunks can be stored uncompressed. If the first byte of the
raw data is \0, we store the data as is. Else we prefix it with 'u'.
Before, we performed a string slice to strip out the 'u' prefix.
With this patch, we use a buffer to avoid an extra memory copy and
associated garbage collection overhead.
I was unable to verify any performance impact of this patch. For both
mozilla-central and the hg repos, the number of manifest revisions
with 'u' prefixes is very small - under 1%. So this change likely
isn't called enough to have an impact on manifest reading. However,
the reasoning behind this change is solid, so it should be safe.
This code is already written in multiple locations.
While this code needs to be fast and extracting it to its own function
adds overhead, code paths reading delta chains typically read,
decompress, and do binary patching on revlog data from the delta chain.
This other work (especially zlib decompression) almost certainly
accounts for a lot more time than the overhead of introducing a Python
function call. So I'm not worried about the performance impact of this
change.
clearcaches() was added several years ago in 1e47437a1ca7 as part
of implementing a perf command. Since revlog instances have many caches
and since the spirit of this mostly unused method is to facilitate
performance testing, I think it's appropriate for all the revlog's
caches to get cleared when it is called.
Revlogs were recently refactored to open file handles in "a+" and use a
persistent file handle for reading and writing. This drastically
reduced the number of file handles being opened.
Unfortunately, it appears that some versions of Solaris lose the file
offset when performing a write after the handle has been seeked.
The simplest workaround is to seek to EOF on files opened in a+ mode
before writing to them, which is what this patch does.
Ideally, this code would exist in the vfs layer. However, this would
require creating a proxy class for file objects in order to provide a
custom implementation of write(). This would add overhead. Since
revlogs are the only files we open in a+ mode, the one-off workaround
in revlog.py should be sufficient.
This patch appears to have little to no impact on performance on my
Linux machine.
In the most complex case, we try using the incoming delta base, then
we try both parents, and then we try the previous revlog entry. If
none of these result in a good delta, we natually use the null
revision as base. However, we sometimes consider the nullrev before we
have exhausted our other options. Specifically, when both parents are
null, we use the nullrev as delta base if it produces a good delta
(according to _isgooddelta()), and we fail to try the previous revlog
entry as delta base. After e60126c6093d (addrevision: use general
delta when the incoming base delta is bad, 2015-12-01), it can also
happen for non-merge commits when the incoming delta is not good.
The Firefox repo (from many months back) shrinks a tiny bit with this
patch: from 1.855GB to 1.830GB (1.4%). The hg repo itself shrinks even
less: by less than 0.1%. There may be repos that get larger instead.
This undoes the unexplained test change in e60126c6093d.
We always want to call _isgooddelta() before accepting a delta. We
mostly call the function right after building the delta, but not
always. Instead, we have an extra call at the end of the big code
block. Let's make it consistent, so we call _isgooddelta() right after
builddelta() and exactly once per delta. That also lets us rely on
"delta is None" to mean we didn't find a good delta.
The tested delta revisions are added to the 'tested' set. These are
the same revisions we pass to builddelta(). However, in one case, we
add builddelta(rev)[3] to the set intead of adding 'rev' itself. In
that particular case, that element is the same as the function's input
revision (because self._generaldelta is true), so the effect is the
same. Still, let's just add the function's input revision to avoid
confusing future readers.
We unify the delta selection process to be a simple three options process:
- try to use the incoming delta (if lazydeltabase is on)
- try to find a suitable parents to delta against (if gd is on)
- try to delta against the tipmost revision
The first of this option that yield a valid delta will be used.
The test change in 'test-generaldelta.t' show this behavior as we use a delta
against the parent instead of a full delta when the incoming delta is not
suitable.
This as some impact on 'test-bundle.t' because a delta somewhere changes. It
does not seems to change the test semantic and have been ignored.
The old code have multiple explicit tests and code duplications. This makes it
hard to improve the code. We rewrite the logic in a more generic way, not
changing anything of the computed result.
The final goal here is to eventually be able to:
- factor out the default fallback case "try against 'prev'" in a single place
- allow 'lazydeltabase' case to use the smarter general delta code path when
the incoming base does not provide us with a good delta.
Before this change, the 'lazydeltabase' would blindly build a delta using the
base provided by the incoming bundle and try to use it. If that base was far
down the revlog, the delta would be seen as "no good" and we would fall back to
a full text revision.
We now check if the delta is good and fallback to a computing a delta again the
tipmost revision otherwise (as we would do without general delta).
Later changesets will improve the logic to compute the fallback delta using the
general delta logic.
We would like to be able to exit the delta generation block without a valid
delta (for a more flexible control flow). So we make sure we do not expand the
"delta" content unless we actually have a delta.
We can do it one level lower because 'delta' is initialised at None anyway. Not
adding a level to the assignment prevent a line length issue.
The new convention is that bundles contain changegroups. bundle1
happens to *only* be a changegroup, but bundle2 is a more featureful
container that isn't something you can pass to addgroup().
The current behavior of revlogs is to flush the data file when writing
data to it. Tracing system calls revealed that changegroup processing
incurred numerous write(2) calls for values much smaller than the
default buffer size (Python defaults to 4096, but it can be adjusted
based on detected block size at run time by CPython).
The reason we flush revlogs is so readers have all data available.
For example, the current code in revlog.py will re-open the revlog
file (instead of seeking an existing file handle) to read the text
of a revision. This happens when starting a new delta chain when
adding several revisions from changegroups, for example. Yes, this
is likely sub-optimal (we should probably be sharing file descriptors
between readers and writers to avoid the flushing and associated
overhead of re-opening files).
While flushing revlogs is necessary, it appears all callers are
diligent about flushing files before a read is performed (see
buildtext() in _addrevision()), making the flush in
_writeentry() redundant and unncessary. So, we remove it. In practice,
this means we incur a write(2) a) when the buffer is full (typically
4096 bytes) b) when a new delta chain is created rather than after
every added revision. This applies to every revlog, but by volume
it mostly impacts filelogs.
Removing the redundant flush from _writeentry() significantly
reduces the number of write(2) calls during changegroup processing on
my Linux machine. When applying a changegroup of the hg repo based on
my local repo, the total number of write(2) calls during application
of the mercurial/localrepo.py revlogs dropped from 1,320 to 217 with
this patch applied. Total I/O related system calls dropped from 1,577
to 474.
When unbundling a mozilla-central gzipped bundle (264,403 changesets
with 1,492,215 changes to 222,507 files), total write(2) calls
dropped from 1,252,881 to 827,106 and total system calls dropped from
3,601,259 to 3,178,636 - a reduction of 425,775!
While the system call reduction is significant, it appears
to have no impact on wall time on my Linux and Windows machines. Still,
fewer syscalls is fewer syscalls. Surely this can't hurt. If nothing
else, it makes examining remaining system call usage simpler and opens
the door to experimenting with the performance impact of different
buffer sizes.
_addrevision() may need to read from revlogs as part of computing
deltas. Previously, we would flush existing file handles and open
a new, short-lived file handle to perform the reading.
If we have an existing file handle, it seems logical to reuse it
for reading instead of opening a new file handle. This patch
makes that the new behavior.
After this patch, revlog files are only reopened when adding
revisions if the revlog is switched from inline to non-inline.
On Linux when unbundling a bundle of the mozilla-central repo, this
patch has the following impact on system call counts:
Call Before After Delta
write 827,639 673,390 -154,249
open 700,103 684,089 -16,014
read 74,489 74,489 0
fstat 493,924 461,896 -32,028
close 249,131 233,117 -16,014
stat 242,001 242,001 0
lstat 18,676 18,676 0
lseek 20,268 20,268 0
ioctl 14,652 13,173 -1,479
TOTAL 3,180,758 2,930,679 -250,079
It's worth noting that many of the open() calls fail due to missing
files. That's why there are many more open() calls than close().
Despite the significant system call reduction, this change does not
seem to have a significant performance impact on Linux.
On Windows 10 (not a VM, on a SSD), this patch appears to reduce
unbundle time for mozilla-central from ~960s to ~920s. This isn't
as significant as I was hoping. But a decrease it is nonetheless.
Still, Windows unbundle performance is still >2x slower than Linux.
Despite the lack of significant gains, fewer system calls is fewer
system calls. If nothing else, this will narrow the focus of potential
areas to optimize in the future.
An upcoming patch will teach revlogs to use the existing file
handle to read revision data instead of opening a new file handle
just for quick reads. For this to work, files must be opened for
reading as well.
This patch is merely cosmetic: there are no behavior changes.
Currently, the low-level revlog reading code always opens a new file
handle. In some key scenarios, the revlog is already opened and an
existing file handle could be used to read. This patch paves the
road to that by teaching various revlog reading functions to accept
an optional existing file handle to read from.
revlog instances can cache the full text of a single revision. Typically
the most recently read revision is cached.
When adding a delta group via addgroup() and _addrevision(), the
full text isn't always computed: sometimes only the passed in delta is
sufficient for adding a new revision to the revlog.
When writing the changelog from a delta group, the just-added full
text revision is always read immediately after it is written because
the changegroup code needs to extract the set of files from the entry.
In other words, revision() is *always* being called and caching the full
text of the just-added revision is guaranteed to result in a cache hit,
making the cache worthwhile.
This patch adds support to _addrevision() for always building and
caching the full text. This option is currently only active when
processing changelog entries from a changegroup.
While the total number of revision() calls is the same, the location
matters: buildtext() calls into revision() on the base revision when
building the full text of the just-added revision. Since the previous
revision's _addrevision() built the full text and the the previous
revision is likely the base revision, this means that the base
revision's full text is likely cached and can be used to compute the
current full text from just a delta. No extra I/O required.
The end result is the changelog isn't opened and read after adding every
revision from a changegroup.
On my 2013 MacBook Pro running OS X 10.10.5 from an SSD and Python 2.7,
this patch impacted the time taken to apply ~262,000 changesets from a
mozilla-central gzip bundle:
before: ~43s
after: ~32s
~25% reduction in changelog processing times. Not bad.
The purpose of this code was to provide thread safety. With the
conversion of hgweb to use separate localrepository instances per
request/thread, we should no longer have any consumers that need to
access revlog instances from multiple threads. Remove the code.
This adds an option for delta'ing against both p1 and p2 when applying merge
revisions and picking whichever is smallest.
Some before and after stats on manifest.d size:
internal large repo:
before: 1.2 GB
after: 930 MB
mozilla-central:
before: 261 MB
after: 92 MB
The old generaldelta heuristic was "if p1 (or p2) was closer than the last full text,
use it, otherwise use prev". This was problematic when a repo contained multiple
branches that were very different. If commits to branch A were pushed, and the
last full text was branch B, it would generate a fulltext. Then if branch B was
pushed, it would generate another fulltext. The problem is that the last
fulltext (and delta'ing against `prev` in general) has no correlation with the
contents of the incoming revision, and therefore will always have degenerate
cases.
According to the blame, that algorithm was chosen to minimize the chain length.
Since there is already code that protects against that (the delta-vs-fulltext
code), and since it has been improved since the original generaldelta algorithm
went in (2011), I believe the chain length criteria will still be preserved.
The new algorithm always diffs against p1 (or p2 if it's closer), unless the
resulting delta will fail the delta-vs-fulltext check, in which case we delta
against prev.
Some before and after stats on manifest.d size.
internal large repo
old heuristic - 2.0 GB
new heuristic - 1.2 GB
mozilla-central
old heuristic - 242 MB
new heuristic - 261 MB
The regression in mozilla central is due to the new heuristic choosing p2r as
the delta when it's closer to the tip. Switching the algorithm to always prefer
p1r brings the size back down (242 MB). This is result of the way in which
mozilla does merges and pushes, and the result could easily swing the other
direction in other repos (depending on if they merge X into Y or Y into X), but
will never be as degenerate as before.
I future patch will address the regression by introducing an optional, even more
aggressive delta heuristic which will knock the mozilla manifest size down
dramatically.
This moves the textlen calculation to be above the delta chooser. Since textlen
is needed for calling isgooddelta, we need it above the delta chooser so future
patches can call isgooddelta.
This moves the delta vs fulltext comparison to its own function. This will allow
us to reuse the function in future patches for more efficient delta choices. As
a side effect, this will also allow extensions to modify our delta criteria.
This option will make repositories created as general delta by default but will
not make Mercurial aggressively recompute deltas for all incoming bundle.
Instead, the delta contained in the bundle will be used. This will allow us to
start having general delta repositories created everywhere without triggering
massive recomputation costs for all new clients cloning from old servers.
A subsequent patch will add a feature that performs iterative
computation as changesets are added from a changegroup. To facilitate
this type of processing in a generic manner, we add a mechanism for
calling a function whenever a revision is added via revlog.addgroup().
There are potential performance concerns with this callback, as using it
will flush the revlog after every revision is added.
Python 2.6 introduced the "except type as instance" syntax, replacing
the "except type, instance" syntax that came before. Python 3 dropped
support for the latter syntax. Since we no longer support Python 2.4 or
2.5, we have no need to continue supporting the "except type, instance".
This patch mass rewrites the exception syntax to be Python 2.6+ and
Python 3 compatible.
This patch was produced by running `2to3 -f except -w -n .`.
Before we were relying on _pack to error out when trying to pass an
integer that was too large for the "i" format specifier. Now we check
this earlier so we can form a better error message.
The error message unfortunately must exclude the filename at this
level of the call stack. The problem is that this name is not
available here, and the error can be triggered by a large manifest or
by a large file itself. Although perhaps we could provide the name of
a revlog index file (from the revlog object, instead of the revlogio
object), this seems like too much leakage of internal data structures.
It's not ideal already that an error message even mentions revlogs,
but this does seem unavoidable here.
I forgot to include this change as a previous diff and the native code to
compute the phases was never called. The AttributeError was silently caught and
the pure implementation was used instead.
The checkinlinesize function, which converts inline revlogs to non-inline,
uses the current transaction's "data" field to determine how to update the
transaction after the conversion.
This change works around the missing data field, which is not in the
transaction after a strip.
A censored revision stored in a revlog should have the censored revlog index
flag bit set. This implies we must know if a revision is censored before we
add it to the revlog. When adding revisions from exchanged deltas, we would
prefer to determine this flag without decoding every single full text.
This change introduces a heuristic based on assumptions around the Mercurial
delta format and filelog metadata. Since deltas which produce a censored
revision must be full-replacement deltas, we can read the delta's first bytes
to check the filelog metadata. Since "censored" is the alphabetically first
filelog metadata key, censored filelog revisions have a well-known prefix we
can look for.
For more on the design and background of the censorship feature, see:
http://mercurial.selenic.com/wiki/CensorPlan
A delta against a censored revision is either received through exchange and
written blindly to a revlog, or it is created by the revlog itself. This
change ensures the latter process creates deltas which fully replace all
data in a censored base using a single patch operation.
Recipients of a delta against a censored base will verify that the delta is in
this full-replace format. Other recipients will use the delta as normal.
For background and broader design of the censorship feature, see:
http://mercurial.selenic.com/wiki/CensorPlan
When a delta received through exchange is added to a revlog, it will very
often be expanded to a full text by applying the delta to its base. If
that delta is of a particular form, we can avoid decoding the base revision.
This avoids an exception if the base revision is censored.
For background and broader design of the censorship feature, see:
http://mercurial.selenic.com/wiki/CensorPlan
To ensure interoperability when clones disagree about which file nodes are
censored, a restriction is made on deltas based on censored nodes. Any such
delta must replace the full text of the base in a single patch.
If the recipient of a delta considers the base to be censored and the delta
is not in the expected form, the recipient must reject it, as it can't know
if the source has also censored the base.
For background and broader design of the censorship feature, see:
http://mercurial.selenic.com/wiki/CensorPlan
The iscensored method will be used by the exchange layer to reject
nonconforming deltas involving censored revisions (and to produce
conforming deltas).
For background and broader design of the censorship feature, see:
http://mercurial.selenic.com/wiki/CensorPlan
Because revlog implements __iter__, "rev in revlog" works but does silly O(n)
lookup unexpectedly. So it seems good to add fast version of __contains__.
This allows "rev in repo.changelog" in the next patch.
When receiving a delta via exchange, three possible storage outcomes emerge:
1. The delta is added directly to the revlog. ("fast-path")
2. A freshly-computed delta with a different base is stored.
3. The new revision's fulltext is computed and stored outright.
Both (2) and (3) require materializing the full text of the new revision by
applying the delta to its base. This is typically followed by a hash check.
The new flags argument allows callers to _addrevision to signal that they
expect that hash check to fail. We can use this opportunity to verify that
expectation. If the hash fails, require the flag be set; if the hash passes,
require the flag be unset.
Rather than simply eliding the hash check, this approach provides some
assurance that the censored flag is not applied to valid revisions.
Read more at: http://mercurial.selenic.com/wiki/CensorPlan
For revlog index flags to be useful to other parts of Mercurial, they need to
be settable when writing revisions. The current use case for revlog index
flags is the censorship feature: http://mercurial.selenic.com/wiki/CensorPlan
While the censor flag could be inferred in _addrevision by interrogating the
text/delta being added, that would bury the censorship logic and
inappropriately couple it to all revision creation.
This flag bit will be used to cheaply signal censorship presence to upper
layers (exchange, verify). It indicates that censorship metadata is present
but does not attest to the verifiability of that metadata.
For the censorship design, see: http://mercurial.selenic.com/wiki/CensorPlan
This dumb cache works surprisingly well: on a repository with typical delta
chains ~50k in length, unbundling a linear series of 5000 revisions (changelogs
and manifests only) went from 60 seconds to 3.
This doesn't affect normal clones since they'd be bound by the CPU bound below
anyway -- it does, however, improve generaldelta clones significantly.
This also results in better deltaing for generaldelta clones -- in generaldelta
clones, we calculate deltas with respect to the closest base if it has a higher
revision number than either parent. If the base is on a significantly different
branch, this can result in pointlessly massive deltas. This reduces the number
of bases and hence the number of bad deltas.
Empirically, for a highly branchy repository, this resulted in an improvement
of around 15% to manifest size.
This is a very silly case and not particularly likely to happen in the wild,
but it turns out we can hit it in a couple of places. As we tune the storage
parameters we're likely to hit more such cases.
The affected test cases all have smaller revlogs now.
The current heuristic for deciding between storing delta and full texts
is based on ratio of (sizeofdeltas)/(sizeoffulltext).
In some cases (for example a manifest for ahuge repo) this approach
can result in extremely long delta chains (~30,000) which are very slow to
read. (In the case of a manifest ~500ms are added to every hg command because of that).
This commit introduces "revlog.maxchainlength" configuration variable that will
limit delta chain length.
The chain length was computed correctly only when generaldelta
feature was enabled. Now it's fixed.
When generaldelta is disabled the base revision in revlog index is not
the revision we have delta against - it's always previous revision.
Instead of incorrect chainbaseandlen in command.py we are now using two
single-responsibility functions in revlog.py:
- chainbase(rev)
- chainlen(rev)
Only chainlen(rev) was missing so it was written to mimic the way the
chain of deltas is actually found during file reconstruction.