The part we are using for pull is now used for push too. As we no longer use
pushkey, pushkey hooks are no longer triggered. This is an obvious backward
incompatible change. We could artificially trigger the pushkey hook within the
bundle2 part, but this seemed too hacky to me.
An option would be to disable by default this new mechanism for a couple of
versions to help people migrate to `txnclose-phase`. I took the liberal and
optimistic path to just turn it on by default directly.
.. bc::
Push no longer triggers a pushkey hook when updating phases. Use the new
`txnclose-phase` and `txnclose-phase` hooks instead.
(Applies when both server and client use version 4.4 or above).
We are about to switch phase pushing from using pushkey to using a the new
dedicated binary part. We introduce the push race detection on its own to help
detect potential impact.
Upon pull or unbundle, we display a message with the range of new revisions
fetched. This revision range could readily be used after a pull to look out
what's new with 'hg log'. The algorithm takes care of filtering "obsolete"
revisions that might be present in transaction's "changes" but should not be
displayed to the end user.
A new bundle2 capability 'phases' has been added. If 'heads' is part of the
supported value for 'phases', the server supports reading and sending 'phase-
heads' bundle2 part.
Server is now able to process a 'phases' boolean parameter to 'getbundle'. If
'True', a 'phase-heads' bundle2 part will be included in the bundle with phase
information relevant to the whole pulled set. If this method is available the
phases listkey namespace will no longer be listed.
Beside the more efficient encoding of the data, this new method will greatly
improve the phase exchange efficiency for repositories with non-served
changesets (obsolete, secret) since we'll no longer send data about the
filtered heads.
Add a new 'devel.legacy.exchange' config item to allow fallback to the old
'listkey in bundle2' method.
Reminder: the pulled set is not just the changesets bundled by the pull. It
also contains changeset selected by the "pull specification" on the client
side (eg: everything for bare pull). One of the reason why the 'pulled set' is
important is to make sure we can move -common- nodes to public.
A few tests hardcode errno numbers and/or descriptions in the output, causing
test failures on platforms where these values are different.
Differential Revision: https://phab.mercurial-scm.org/D362
Now that we garantee that branchmap cache is updated at the end of the
transaction we can drop this update. This removes a problematic case with
nested transaction where the new cache could be written on disk before the
transaction is finished (and even roll-backed)
Such premature cache write was visible in the following test:
* tests/test-acl.t
* tests/test-rebase-conflicts.t
In addition, running the cache update later means having more date about the
state of the repository (in particular: phases). So we can generate caches with
more information. This creates harmless changes to the following tests:
* tests/test-hardlinks-whitelisted.t
* tests/test-hardlinks.t
* tests/test-phases.t
* tests/test-tags.t
* tests/test-inherit-mode.t
We are about to remove the branchmap cache update in changegroup application.
There is a debug message alongside this update that we do not want to loose. We
move the message beforehand to simplify the test update in the next changeset.
The message move is quite noisy and isolating that noise is useful.
Most tests update are just line reordering since the message is issued at a
later point during the transaction.
After this changes, the message is displayed in more case since local commit
creation also issue it.
Revlog can now be configured to store full snapshot only. This is used on the
changelog. However, the changegroup packing was still recomputing deltas to be
sent over the wire.
We now just reuse the full snapshot directly in this case, skipping delta
computation. This provides use with a large speed up(-30%):
# perfchangegroupchangelog on mercurial
! wall 2.010326 comb 2.020000 user 2.000000 sys 0.020000 (best of 5)
! wall 1.382039 comb 1.380000 user 1.370000 sys 0.010000 (best of 8)
# perfchangegroupchangelog on pypy
! wall 5.792589 comb 5.780000 user 5.780000 sys 0.000000 (best of 3)
! wall 3.911158 comb 3.920000 user 3.900000 sys 0.020000 (best of 3)
# perfchangegroupchangelog on mozilla central
! wall 20.683727 comb 20.680000 user 20.630000 sys 0.050000 (best of 3)
! wall 14.190204 comb 14.190000 user 14.150000 sys 0.040000 (best of 3)
Many tests have to be updated because of the change in bundle content. All
theses update have been verified. Because diffing changelog was not very
valuable, the resulting bundle have similar size (often a bit smaller):
# full bundle of mozilla central
with delta: 1142740533B
without delta: 1142173300B
So this is a win all over the board.
This patch disables delta chains on changelogs. After this patch, new
entries on changelogs - including existing changelogs - will be stored
as the fulltext of that data (likely compressed). No delta computation
will be performed.
An overview of delta chains and data justifying this change follows.
Revlogs try to store entries as a delta against a previous entry (either
a parent revision in the case of generaldelta or the previous physical
revision when not using generaldelta). Most of the time this is the
correct thing to do: it frequently results in less CPU usage and smaller
storage.
Delta chains are most effective when the base revision being deltad
against is similar to the current data. This tends to occur naturally
for manifests and file data, since only small parts of each tend to
change with each revision. Changelogs, however, are a different story.
Changelog entries represent changesets/commits. And unless commits in a
repository are homogonous (same author, changing same files, similar
commit messages, etc), a delta from one entry to the next tends to be
relatively large compared to the size of the entry. This means that
delta chains tend to be short. How short? Here is the full vs delta
revision breakdown on some real world repos:
Repo % Full % Delta Max Length
hg 45.8 54.2 6
mozilla-central 42.4 57.6 8
mozilla-unified 42.5 57.5 17
pypy 46.1 53.9 6
python-zstandard 46.1 53.9 3
(I threw in python-zstandard as an example of a repo that is homogonous.
It contains a small Python project with changes all from the same
author.)
Contrast this with the manifest revlog for these repos, where 99+% of
revisions are deltas and delta chains run into the thousands.
So delta chains aren't as useful on changelogs. But even a short delta
chain may provide benefits. Let's measure that.
Delta chains may require less CPU to read revisions if the CPU time
spent reading smaller deltas is less than the CPU time used to
decompress larger individual entries. We can measure this via
`hg perfrevlog -c -d 1` to iterate a revlog to resolve each revision's
fulltext. Here are the results of that command on a repo using delta
chains in its changelog and on a repo without delta chains:
hg (forward)
! wall 0.407008 comb 0.410000 user 0.410000 sys 0.000000 (best of 25)
! wall 0.390061 comb 0.390000 user 0.390000 sys 0.000000 (best of 26)
hg (reverse)
! wall 0.515221 comb 0.520000 user 0.520000 sys 0.000000 (best of 19)
! wall 0.400018 comb 0.400000 user 0.390000 sys 0.010000 (best of 25)
mozilla-central (forward)
! wall 4.508296 comb 4.490000 user 4.490000 sys 0.000000 (best of 3)
! wall 4.370222 comb 4.370000 user 4.350000 sys 0.020000 (best of 3)
mozilla-central (reverse)
! wall 5.758995 comb 5.760000 user 5.720000 sys 0.040000 (best of 3)
! wall 4.346503 comb 4.340000 user 4.320000 sys 0.020000 (best of 3)
mozilla-unified (forward)
! wall 4.957088 comb 4.950000 user 4.940000 sys 0.010000 (best of 3)
! wall 4.660528 comb 4.650000 user 4.630000 sys 0.020000 (best of 3)
mozilla-unified (reverse)
! wall 6.119827 comb 6.110000 user 6.090000 sys 0.020000 (best of 3)
! wall 4.675136 comb 4.670000 user 4.670000 sys 0.000000 (best of 3)
pypy (forward)
! wall 1.231122 comb 1.240000 user 1.230000 sys 0.010000 (best of 8)
! wall 1.164896 comb 1.160000 user 1.160000 sys 0.000000 (best of 9)
pypy (reverse)
! wall 1.467049 comb 1.460000 user 1.460000 sys 0.000000 (best of 7)
! wall 1.160200 comb 1.170000 user 1.160000 sys 0.010000 (best of 9)
The data clearly shows that it takes less wall and CPU time to resolve
revisions when there are no delta chains in the changelogs, regardless
of the direction of traversal. Furthermore, not using a delta chain
means that fulltext resolution in reverse is as fast as iterating
forward. So not using delta chains on the changelog is a clear CPU win
for reading operations.
An example of a user-visible operation showing this speed-up is revset
evaluation. Here are results for
`hg perfrevset 'author(gps) or author(mpm)'`:
hg
! wall 1.655506 comb 1.660000 user 1.650000 sys 0.010000 (best of 6)
! wall 1.612723 comb 1.610000 user 1.600000 sys 0.010000 (best of 7)
mozilla-central
! wall 17.629826 comb 17.640000 user 17.600000 sys 0.040000 (best of 3)
! wall 17.311033 comb 17.300000 user 17.260000 sys 0.040000 (best of 3)
What about 00changelog.i size?
Repo Delta Chains No Delta Chains
hg 7,033,250 6,976,771
mozilla-central 82,978,748 81,574,623
mozilla-unified 88,112,349 86,702,162
pypy 20,740,699 20,659,741
The data shows that removing delta chains from the changelog makes the
changelog smaller.
Delta chains are also used during changegroup generation. This
operation essentially converts a series of revisions to one large
delta chain. And changegroup generation is smart: if the delta in
the revlog matches what the changegroup is emitting, it will reuse
the delta instead of recalculating it. We can measure the impact
removing changelog delta chains has on changegroup generation via
`hg perfchangegroupchangelog`:
hg
! wall 1.589245 comb 1.590000 user 1.590000 sys 0.000000 (best of 7)
! wall 1.788060 comb 1.790000 user 1.790000 sys 0.000000 (best of 6)
mozilla-central
! wall 17.382585 comb 17.380000 user 17.340000 sys 0.040000 (best of 3)
! wall 20.161357 comb 20.160000 user 20.120000 sys 0.040000 (best of 3)
mozilla-unified
! wall 18.722839 comb 18.720000 user 18.680000 sys 0.040000 (best of 3)
! wall 21.168075 comb 21.170000 user 21.130000 sys 0.040000 (best of 3)
pypy
! wall 4.828317 comb 4.830000 user 4.820000 sys 0.010000 (best of 3)
! wall 5.415455 comb 5.420000 user 5.410000 sys 0.010000 (best of 3)
The data shows eliminating delta chains makes the changelog part of
changegroup generation slower. This is expected since we now have to
compute deltas for revisions where we could recycle the delta before.
It is worth putting this regression into context of overall changegroup
times. Here is the rough total CPU time spent in changegroup generation
for various repos while using delta chains on the changelog:
Repo CPU Time (s) CPU Time w/ compression
hg 4.50 7.05
mozilla-central 111.1 222.0
pypy 28.68 75.5
Before compression, removing delta chains from the changegroup adds
~4.4% overhead to hg changegroup generation, 1.3% to mozilla-central,
and 2.0% to pypy. When you factor in zlib compression, these percentages
are roughly divided by 2.
While the increased CPU usage for changegroup generation is unfortunate,
I think it is acceptable because the percentage is small, server
operators (those likely impacted most by this) have other mechanisms
to mitigate CPU consumption (namely reducing zlib compression level and
pre-generated clone bundles), and because there is room to optimize this
in the future. For example, we could use the nullid as the base revision,
effectively encoding the full revision for each entry in the changegroup.
When doing this, `hg perfchangegroupchangelog` nearly halves:
mozilla-unified
! wall 21.168075 comb 21.170000 user 21.130000 sys 0.040000 (best of 3)
! wall 11.196461 comb 11.200000 user 11.190000 sys 0.010000 (best of 3)
This looks very promising as a future optimization opportunity.
It's worth that the changes in test-acl.t to the changegroup part size.
This is because revision 6 in the changegroup had a delta chain of
length 2 before and after this patch the base revision is nullrev.
When the base revision is nullrev, cg2packer.deltaparent() hardcodes
the *previous* revision from the changegroup as the delta parent.
This caused the delta in the changegroup to switch base revisions,
the delta to change, and the size to change accordingly. While the
size increased in this case, I think sizes will remain the same
on average, as the delta base for changelog revisions doesn't matter
too much (as this patch shows). So, I don't consider this a regression.
In order to give us the freedom to change the changegroup3 format,
let's hide it behind an experimental config. Since it is required by
treemanifests, that will override the cg3 config.
I'm not entirely happy with using a trailing / on a "file" entry for
transferring a treemanifest. We've discussed putting some flags on
each file header[0], but I'm unconvinced that's actually any better:
if we were going to add another feature to the cg format we'd still be
doing a version bump anyway to cg4, so I'm inclined to not spend time
coming up with a more sophisticated format until we actually know what
the next feature we want to stuff in a changegroup will be.
Test changes outside test-treemanifest.t are only due to the new CG3
bundlecap showing up in the wire protocol.
Many thanks to adgar@google.com and martinvonz@google.com for helping
me with various odd corners of the changegroup and treemanifest API.
0: It's not hard refactoring, nor is it a lot of work. I'm just
disinclined to do speculative work when it's not clear what the
customer would actually be.
When using bundle2, the phase pushkey parts are now made mandatory. As a
result, failure to update the bookmark server side will result in the transaction
being aborted.
The current behavior (with bundle1) is to let the rest of the push succeed if
the pushkey call (phases, bookmarks) failed (this comes from the fact that each
item is sent in its own command).
We kept this behavior with bundle2, which is highly debatable, but let us keep
thing as they are now as a start. We are about to enforce 'mandatory' pushkey
part as 'mandatory' successful, so we need to marks parts as advisory to
preserve the current (debatable) behavior.
.hgtags fnodes cache entries can be expensive to compute, especially
if there are hundreds of even thousands of them. This patch implements
support for receiving a bundle2 part that contains a mapping of
changeset to .hgtags fnodes.
An upcoming patch will teach the server to send this part, allowing
clients to bypass having to redundantly compute these values.
A number of tests changed due to the client advertising the "hgtagsfnodes"
capability.
This test makes extensive use of --debug so moving to bundle2 based exchange
has a massive impact. We do it early to reduce the noise create by a future
usage of bundle2 as the default protocol.
When the progress extension is not enabled, each call to 'ui.progress' used to
issue a debug message. This results is a very verbose output and often redundant
in tests. Dropping it makes tests less volatile to factor they do not meant to
test.
We had to alter the sed trick in 'test-rename-merge2.t'. Sed is used to drop all
output from a certain point and hidding the progress output remove its anchor.
So we anchor on something else.
A future patch is going to add some extra commented-out boilerplate to
the top of .hg/hgrc during clone. In order to make this test not
require regular updates, switch to searching for [hooks] or [acl] and
print file from the first match to that pattern.
The discovery of necessary bookmark updates is now done within the "discovery
phase". This opens the door to the inclusion of bookmarks in a unified bundle2
push.
This input will have to travel over the wire anyway, so we feed the peer method
with a simple binary stream and rely on the server side to use `readbundle`
to create the python object.
The test output changes because the bundle is created marginally sooner and the
debug output interleaves in a different way.
Localrepo now supports the unbundle method of pushing changegroups. We
plan to use the unbundle call for bundle2 so it is important that all
peers supports it. The `peer.unbundle` and `peer.addchangegroup` code
path have small difference so cause some test output changes. None of those
changes seems problematic.
The strip code used a trick to lower the cost of branchcache update after a
strip. However is less necessary since we have branchcache collaboration.
Invalid branchcache are likely to be cheaply rebuilt again a near subset of the
repo.
Moreover, this trick would need update to be relevant in the now filtered
repository world. It currently update the unfiltered branchcache that few people
cares about. Make it smarter on that aspect would need complexes update of the
calling logic
So this mechanism is:
- Arguably needed,
- Currently irrelevant,
- Hard to update
and I'm dropping it.
We now update the branchcache in all case by courtesy of the read only reader.
This changeset have a few expected impact on the testsuite are different cache
are updated.
Now that changelog filtering is in place, it's become evident that
naming the filters according to the set of revs _not_ included in the
filtered changelog is confusing. This is especially evident in the
collaborative branch cache scheme.
This changes the names of the filters to reflect the revs that _are_
included:
hidden -> visible
unserved -> served
mutable -> immutable
impactable -> base
repoview.filteredrevs is renamed to filterrevs, so that callers read a
bit more sensibly, e.g.:
filterrevs('visible') # filter revs according to what's visible
It filters all mutable changesets, leaving only public changeset unfiltered.
This filtering set is expected to be much more stable that the previous one as
public changeset are unlikely to disapear.
The only official use of this filter is for branchcache.
Filtered repository are *subset* of unfiltered repository. This means that a
filtered branchmap could be use to compute the unfiltered version.
And filtered version happen to be subset of each other:
- "all() - unserved()" is a subset of "all() - hidden()"
- "all() - hidden()" is a subset of "all()"
This means that branchmap with "unfiltered" filter can be used as a base for
"hidden" branchmap that itself could be used as a base for unfiltered
branchmap.
unserved < hidden < None
This changeset implements this mechanism. If the on disk branchcache is not valid
we use the branchcache of the nearest subset as base instead of computing it from
scratch. Such fallback can be cascaded multiple time is necessary.
Note that both "hidden" and "unserved" set are a bit volatile. We will add more
stable filtering in next changesets.
This changeset enables collaboration between no filtering and "unserved"
filtering. Fixing performance regression introduced by 7bff5f37cb97
Disables this simple optimisation to allow coming more powerfull approach: cache
collaboration.
Our goal is to have branchcache collaborate. This means that unfiltered
branchcache will fallback to some filtered branchcache if invalid. We can't have
the filtered branchcache to use the unfiltered one. That would loop.
The "!" prefix allows you to prevent anyone except a given user or group to push
changesets in a given branch or path.
This patch enables a use case suggested by a user (Julien Bonnet):
There's a branch that only a given user (or group) should be able to push to,
and you don't want to restrict access to any other branch that may be created.
With the "!" prefix, you simply deny access to "!givenuser" (or "!@givengroup").
If the working dir parent was destroyed by rollback, then the old
behaviour is perfectly reasonable: restore dirstate, branch, and
bookmarks. That way the working dir moves back to an existing
changeset rather than becoming an orphan.
But if the working dir parent was unaffected -- say, you updated to an
older changeset and then did rollback -- then it's silly to restore
dirstate and branch. So don't do that. Leave the status of the working
dir alone. (But always restore bookmarks, because that file refers to
changeset IDs that may have been destroyed.)
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.
This speeds up the in-memory version of debugbuilddag that I'm
working on considerably for the case where we want to build just
a 00changelog.i (for discovery tests, for instance).
There are a couple of test changes because node ids in tests
have changed.
The changes to the patch names in test-mq-qdelete.t were required
because they could collide with nodeid abbreviations and newly
actually do (patch "c" collides with id "cafe..." for patch "b").
This is a long desired cleanup and paves the way for new discovery.
To specify subsets for bundling changes, all code should use the heads
of the desired subset ("heads") and the heads of the common subset
("common") to be excluded from the bundled set. These can be used
revlog.findmissing instead of revlog.nodesbetween.
This fixes an actual bug exposed by the change in test-bundle-r.t
where we try to bundle a changeset while specifying that said changeset
is to be assumed already present in the target. This used to still
bundle the changeset. It no longer does. This is similar to the bugs
fixed by the recent switch to heads/common for incoming/pull.
