This regresses performance of 'hg branches', presumably because it's
visiting the revlog in the wrong order. This suggests we either need
to fix the branch code or add some read-behind to mitigate the effect.
This showed up in a statprof profile of "hg svn rebuildmeta", which
is read-intensive on the changelog. This two-line patch improved
the performance of that command by 10%.
This greatly speeds up node->rev lookups, with results that are
often user-perceptible: for instance, "hg --time log" of the node
associated with rev 1000 on a linux-2.6 repo improves from 0.3
seconds to 0.03. I have not found any instances of slowdowns.
The new perfnodelookup command in contrib/perf.py demonstrates the
speedup more dramatically, since it performs no I/O. For a single
lookup, the new code is about 40x faster.
These changes also prepare the ground for the possibility of further
improving the performance of prefix-based node lookups.
This list will contains any node see in the source, not only the added one.
This is intended to allow phase to be move according what was pushed by client
not only what was added.
The usual contract is that close() makes your writes permanent, so
atomictempfile's use of close() to *discard* writes (and rename() to
keep them) is rather unexpected. Thus, change it so close() makes
things permanent and add a new discard() method to throw them away.
discard() is only used internally, in __del__(), to ensure that writes
are discarded when an atomictempfile object goes out of scope.
I audited mercurial.*, hgext.*, and ~80 third-party extensions, and
found no one using the existing semantics of close() to discard
writes, so this should be safe.
This greatly improves the speed of the bundling process, and often reduces the
bundle size considerably. (Although if the repository is already ordered, this
has little effect on both time and bundle size.)
For non-generaldelta clients, the reduced bundle size translates to a reduced
repository size, similar to shrinking the revlogs (which uses the exact same
algorithm). For generaldelta clients the difference is minor.
When the new bundle format comes, reordering will not be necessary since we
can then store the deltaparent relationsships directly. The eventual default
behavior for clients and servers is presented in the table below, where "new"
implies support for GD as well as the new bundle format:
old client new client
old server old bundle, no reorder old bundle, no reorder
new server, non-GD old bundle, no reorder[1] old bundle, no reorder[2]
new server, GD old bundle, reorder[3] new bundle, no reorder[4]
[1] reordering is expensive on the server in this case, skip it
[2] client can choose to do its own redelta here
[3] reordering is needed because otherwise the pull does a lot of extra
work on the server
[4] reordering isn't needed because client can get deltabase in bundle
format
Currently, the default is to reorder on GD-servers, and not otherwise. A new
setting, bundle.reorder, has been added to override the default reordering
behavior. It can be set to either 'auto' (the default), or any true or false
value as a standard boolean setting, to either force the reordering on or off
regardless of generaldelta.
Some timing data from a relatively branch test repository follows. All
bundling is done with --all --type none options.
Non-generaldelta, non-shrunk repo:
-----------------------------------
Size: 276M
Without reorder (default):
Bundle time: 14.4 seconds
Bundle size: 939M
With reorder:
Bundle time: 1 minute, 29.3 seconds
Bundle size: 381M
Generaldelta, non-shrunk repo:
-----------------------------------
Size: 87M
Without reorder:
Bundle time: 2 minutes, 1.4 seconds
Bundle size: 939M
With reorder (default):
Bundle time: 25.5 seconds
Bundle size: 381M
defversion was a property (later option) on the store opener, used to propagate
the changelog revlog format to the other revlogs, so they would be created with
the same format.
This required that the changelog instance was created before any other revlog;
an invariant that wasn't directly enforced (or documented) anywhere.
We now use the revlogv1 requirement instead, which is transfered to the store
opener options. If this option is missing, v0 revlogs are created.
Without this change, pulls (and clones) into a generaldelta repository could
generate very inefficient revlogs, the size of which could be at least twice
the original size.
This was caused by the generated delta chains covering too large distances,
causing new chains to be built far too often. This change addresses the
problem by forcing a delta against second parent or against the previous
revision, when the first parent delta is in danger of creating a long chain.
The bug didn't cause corruption, and thus wasn't caught in hg verify or in
tests. It could lead to delta chains longer than normally allowed, by
affecting the code that decides when to add a full revision. This could,
in turn, lead to performance regression.
With generaldelta switched on, deltas are always computed against the first
parent when adding revisions. This is done regardless of what revision the
incoming bundle, if any, is deltaed against.
The exact delta building strategy is subject to change, but this will not
affect compatibility.
Generaldelta is switched off by default.
Generaldelta is a new revlog global flag. When it's turned on, the base field
of each revision entry holds the deltaparent instead of the base revision of
the current delta chain.
This allows for great potential flexibility when generating deltas, as any
revision can serve as deltaparent. Previously, the deltaparent for revision r
was hardcoded to be r - 1.
The base revision of the delta chain can still be accessed as before, since it
is now computed in an iterative fashion, following the deltaparents backwards.
This is in preparation for generaldelta, where the revlog entry base field is
reinterpreted as the deltaparent. For that reason we also rename the base
function to chainbase.
Without generaldelta, performance is unaffected, but generaldelta will suffer
from this in _addrevision, since delta chains will be walked repeatedly.
A cache has been added to eliminate this problem completely.
Adds a new discovery method based on repeatedly sampling the still
undecided subset of the local node graph to determine the set of nodes
common to both the client and the server.
For small differences between client and server, it uses about the same
or slightly fewer roundtrips than the old tree-based discovery. For
larger differences, it typically reduces the number of roundtrips
drastically (from 150 to 4, for instance).
The old discovery code now lives in treediscovery.py, the new code is
in setdiscovery.py.
Still missing is a hook for extensions to contribute nodes to the
initial sample. For instance, Augie's remotebranches could contribute
the last known state of the server's heads.
Credits for the actual sampler and computing common heads instead of
bases go to Benoit Boissinot.
getbundle(common, heads) -> bundle
Returns the changegroup for all ancestors of heads which are not ancestors of common. For both
sets, the heads are included in the set.
Intended to eventually supercede changegroupsubset and changegroup. Uses heads of common region
to exclude unwanted changesets instead of bases of desired region, which is more useful and
easier to implement.
Designed to be extensible with new optional arguments (which will have to be guarded by
corresponding capabilities).
Add missing calls to close() to many places where files are
opened. Relying on reference counting to catch them soon-ish is not
portable and fails in environments with a proper GC, such as PyPy.
Now that the nodemap is lazily created, we use linear scanning back
from tip for typical node to rev mapping. Given that nodemap creation
is O(n log n) and revisions searched for are usually very close to
tip, this is often a significant performance win for a small number of
searches.
When we do end up building a nodemap for bulk lookups, the scanning
function is replaced with a hash lookup.
We were not returning the correct result if nullrev was in revs, as we
are checking parent(currentrev) != nullrev before yielding currentrev
test-convert-hg-startrev was wrong: if we start converting from rev -1 and
onwards, all the descendants of -1 (full repo) should be converted.
When parentdelta is enabled, we choose the delta that has the minimum
distance to its base. Otherwise, base may be sufficiently far away to
require a full version, resulting in greatly reduced compression.
While reading changegroup if a node with missing parents is encountered,
we add a punched entry in the index with null parents for the missing
parent node.
Rename some variables, making the name more obvious (in particular "cache" was
actually two different variable.
Move code around, moving the index preloading before the deltachain computation,
without that index preloading was useless (everything was read in deltachain).
Refactor revlog._addrevision() and put the correct base rev in the
parent-delta case: base(rev) should always be equal to the first full snapshot
that is needed by the delta chain, in both parent-delta and tip-delta case.
Before this fix, the base rev was in most case wrong (and in the case where
p1 == nullid, this triggered the bug from issue2337). This means that
repositories converted to parent-delta earlier are corrupted and needs to be
reconverted.
index flag to identify a revision as punched, i.e. it contains no data.
REVIDX_PUNCHED_FLAG = 2, is used to mark a revision as punched.
REVIDX_KNOWN_FLAGS is the accumulation of all index flags.
This is similar to the __builtin__.cmp behaviour, but still not
straightforward, as the dailylife meaning of a comparison usually is
"find out if they are different".
Previously, it only returned revisions that were in the revlog when it
was originally opened; revisions added since then were invisible.
This broke revlog._partialmatch() and therefore repo.lookup().
(Credit to Benoit Boissinot for simplifying my original test script
and for the actual fix.)
This lets us change the threshold at which a *.d file will be split
out, which should make it much easier to construct test cases that
probe revlogs with a separate data file.
(issue2137)
This lets us change the threshold at which a *.d file will be split
out, which should make it much easier to construct test cases that
probe revlogs with a separate data file.
(issue2137)
If a buffer of an mutable object is passed to revlog.addrevision(), the revlog
will happily store it in its cache. Later when the revlog reuses the cached
entry, if the manifest modified the object in-between, all kind of bugs
appears.
We fix it by:
- passing immutable objects to addrevision() if they are already available
- only storing the text in the cache if it's of str type
Then we can remove the conversion of the cache entry to str() during
retrieval. That was probably just there hiding the bug for the common cases
but not really fixing it.
- chunk to _chunk
- _prime to _chunkraw
- _chunkclear for cache clearing
- _chunk calls _chunkraw
- clean up _prime a bit
- simplify users in revision and checkinlinesize
- drop file descriptor passing (we're better off opening fds lazily
The built-in None object is a singleton and it is therefore safe to
compare memory addresses with is. It is also faster, how much depends
on the object being compared. For a simple type like str I get:
| s = "foo" | s = None
----------+-----------+----------
s == None | 0.25 usec | 0.21 usec
s is None | 0.17 usec | 0.17 usec
Uses a transaction instance from the local repository to journal the
truncation of revlog files, such that if a strip only partially completes,
hg recover will be able to finish the truncate of all the files.
The potential unbundling of changes that have been backed up to be restored
later will, in case of an error, have to be unbundled manually. The
difference is that it will be possible to recover the repository state so
the unbundle can actually succeed.
Because we often compute sha1(nullid), it's interesting to copy a precomputed
hash of nullid instead of computing everytime the same hash. Similarly, when
one of the parents is null, we can avoid a < comparison (sort).
Overall, this change adds a string equality comparison on each hash() call,
but when p2 is null, we drop one string < comparison, and copy a hash instead
of computing it. Since it is common to have revisions with only one parent,
this change makes hash() 25% faster when cloning a big repository.
They are unnecessary. I did leave them in localrepo.py where there is
something like:
_junk = foo()
_junk = None
to free memory early. I don't know if just `foo()` will free the return
value as early.
- create error.py for exception classes to reduce demandloading
- move revlog exceptions to it
- change users to import error and drop revlog import if possible
changegroup() has a problem when nodes which does not descend from a node
in <bases> are added to remote after the discovery phase.
If that happens, changegroup() won't send the correct set of nodes, ie.
some nodes will be missing.
To correct it we have to find the set of nodes that both remote and self
have (called <common>), and send all the nodes not in <common>.
This fix has some overhead, in the worst case it will re-send a whole branch.
A proper fix to avoid this overhead might be to change the protocol so that
the <common> nodes are sent (instead of the <bases> of the missing nodes).
Previously, an unknown node id would lead to the following error:
abort: 00changelog.i@343445453433: no node!
All other unknown revision would instead display as:
abort: unknown revision '343445453'!
The former error message has been suppressed in favor of the latter.
This patch adds two methods to revlog:
- ancestors: given a list of revisions returns their ancestors
- descendants: given a list of revisions return their descendants
If there's no inline data, revlog.revision opens the data file every
time it's called. This is useful if we're going to call chunk many
times, but, if we're going to call it only once, it's better to let
chunk open the file - if we're lucky, all the data we're going to need
is already cached and we won't need to even look at the file.
When we remove revision N from the repository, all revisions >= N are
affected: either it's a descendant from N and will also be removed, or
it's not a descendant of N and will be renumbered.
As a consequence, we have to (at least temporarily) remove all filelog
and manifest revisions that have a linkrev >= N, readding some of them
later.
Unfortunately, it's possible to have a revlog with two revisions
r1 and r2 such that r1 < r2, but linkrev(r1) > linkrev(r2). If we try
to strip revision linkrev(r1) from the repository, we'll also lose
revision r2 when we truncate this revlog.
We already use changegroupsubset to create a temporary changegroup
containing the revisions that have to be restored, but that function is
unable to detect that we also wanted to save the r2 in the case above.
So we manually calculate these extra nodes and pass it to changegroupsubset.
This should fix issue764.
Python's zlib apparently makes an internal copy of strings passed to
compress(). To avoid this, compress strings 1M at a time, then join
them at the end if the result would be smaller than the original.
For initial commits of large but compressible files, this cuts peak
memory usage nearly in half.
- use a buffer to extract the delta from a chunk
- avoid concatenating to a compressed delta
- use a buffer to directly extra full text from a trivial delta
- delete chunk and delta objects after use
- handle chunk headers separately rather than prepending them to
(potentially large) chunks
- break large chunks into 1M pieces for compression
- don't prepend file metadata onto (potentially large) file data
To avoid extra memory usage and performance issues with large files,
generate a trivial delta header for deltas against the null revision
rather than calling the usual delta generator.
We append the delta header to meta rather than prepending it to data
to avoid a large allocate and copy.
We want to store version information about the revlog in the first
entry of its index. The code in packentry was using some heuristics
to detect whether this was the first entry, but these heuristics could
fail in some cases (e.g. rev 0 was empty; rev 1 descends directly from
the nullid and is stored as a delta).
We now give the revision number to packentry to avoid heuristics.
This function is fairly performance sensitive, so we make a couple
ugly tweaks:
- keep all entries packed so we needn't test entry types
- fold index lookup/load into unpack call to eliminate
local variable setting
- remove unused defaults for p1, p2, and text
- reduce some if/else
- use better variable names
- remove some extra variables
- remove some obsolete corner tests
- simply first entry handling for revlogng
- simply inline vs outofline writeout
We expand our index by one entry so that index[nullrev] points to a
unique entry, the null revision. This naturally eliminates numerous
extra tests in the performance-sensitive index access functions, most
of which are now trivial again.
Adding new entries is now done with insert(-1, e) rather than
append(e).
This way can use one additional bit, and when encountering invalid revlogs
with the first bit set don't produce python warnings or strange error messages.