Values in AnnexRead can be read more efficiently, without MVar overhead.
Only a few things have been moved into there, and the performance
increase so far is not likely to be noticable.
This is groundwork for putting more stuff in there, particularly a value
that indicates if debugging is enabled.
The obvious next step is to change option parsing to not run in the
Annex monad to set values in AnnexState, and instead return a pure value
that gets stored in AnnexRead.
Not yet used, but allows getting the size of items in the tree fairly
cheaply.
I noticed that CmdLine.Seek uses ls-tree and the feeds the files into
another long-running process to check their size. That would be an
example of a place that might be sped up by using this. Although in that
particular case, it only needs to know the size of unlocked files, not
locked. And since enabling --long probably doubles the ls-tree runtime
or more, the overhead of using it there may outwweigh the benefit.
Added LinkType to ProvidedInfo, and unified MatchingKey with
ProvidedInfo. They're both used in the same way, so there was no real
reason to keep separate.
Note that addLocked and addUnlocked still set matchNeedsFileName,
because to handle MatchingFile, they do need it. However, they
don't use it when MatchingInfo is provided. This should be ok,
the --branch case will be able skip checking matchNeedsFileName,
since it will provide a filename in any case.
Previously such nonsensical combinations always treated the matching option
as if it didn't match.
For now, made find --branch refuse matching options that need a
filename, because one is not provided to them in a way they'll use.
There's an open bug report to support it, but making it error out is
better than the old behavior of not finding what it was asked to.
Also, made --mimetype combined with eg --all work, by looking at the
object file when operating on keys.
Implemented by generalizing registerurl. Without the implicit batch mode
of registerurl since that is only a backwards compatability thing
(see commit 1d1054faa6).
When annex.stalldetection is not enabled, and a likely stall is detected,
display a suggestion to enable it.
Note that the progress meter display is not taken down when displaying
the message, so it will display like this:
0% 8 B 0 B/s
Transfer seems to have stalled. To handle stalling transfers, configure annex.stalldetection
0% 10 B 0 B/s
Although of course if it's really stalled, it will never update
again after the message. Taking down the progress meter and starting
a new one doesn't seem too necessary given how unusual this is,
also this does help show the state it was at when it stalled.
Use of uninterruptibleCancel here is ok, the thread it's canceling
only does STM transactions and sleeps. The annex thread that gets
forked off is separate to avoid it being canceled, so that it
can be joined back at the end.
A module cycle required moving from dupState the precaching of the
remote list. Doing it at startConcurrency should cover all the cases
where the remote list is used in concurrent actions.
This commit was sponsored by Kevin Mueller on Patreon.
Seems only fair, that, like git runs git-annex, git-annex runs
git-annex-foo.
Implementation relies on O.forwardOptions, so that any options are passed
through to the addon program. Note that this includes options before the
subcommand, eg: git-annex -cx=y foo
Unfortunately, git-annex eats the --help/-h options.
This is because it uses O.hsubparser, which injects that option into each
subcommand. Seems like this should be possible to avoid somehow, to let
commands display their own --help, instead of the dummy one git-annex
displays.
The two step searching mirrors how git works, it makes finding
git-annex-foo fast when "git annex foo" is run, but will also support fuzzy
matching, once findAllAddonCommands gets implemented.
This commit was sponsored by Dr. Land Raider on Patreon.
Since that can lead to data loss, which should never be enabled by an
option other than --force.
I suppose that using --trust was in some situation, safer than --force,
because it doesn't entirely disable checking for data loss, but only
disables checking involving data that is on the specified repository.
But it seems better to be able to say that data loss only happens with
--force.
This commit was sponsored by Graham Spencer on Patreon.
This is conceptually very simple, just making a 1 that was hard coded be
exposed as a config option. The hard part was plumbing all that, and
dealing with complexities like reading it from git attributes at the
same time that numcopies is read.
Behavior change: When numcopies is set to 0, git-annex used to drop
content without requiring any copies. Now to get that (highly unsafe)
behavior, mincopies also needs to be set to 0. It seemed better to
remove that edge case, than complicate mincopies by ignoring it when
numcopies is 0.
This commit was sponsored by Denis Dzyubenko on Patreon.
It got broken in several ways by the streaming seeking optimisations
around version 8.20201007.
Moved time limit checking out of the matcher, which was a hack in the
first place. So everywhere that uses Limit.getMatcher needs to check
time limit. Well, almost everywhere. Command.Info uses it, but it does
not make sense to time limit getting info. And Command.MultiCast uses it
just to build up a list of files that then get passed to a command, so
it would never have hit the timeout in a useful way.
This implementation is a little more expensive when at time limit than
necessary, since it continues seeking only to discard everything after the
time limit. I did try making it close the file handles to force a faster
shutdown, but that didn't work and hung. Could certianly be improved
somehow, but seeking is probably not the expensive bit when a time limit
is hit, so this seems acceptable for now.
MatchingKey is not the thing to use when matching on actual worktreee
files.
Fix reversion in 8.20201116 that made include= and exclude= in
preferred/required content expressions match a path relative to the current
directory, rather than the path from the top of the repository.
This is to avoid breakage when upgrading or downgrading git-annex with a
process running that uses the interface. It's better to keep the
compatability code for a few years than worry about such breakage.
This commit was sponsored by Brett Eisenberg on Patreon.
The problem was this line:
cleanup = and <$> sequence (map snd v)
That caused all of v to be held onto until the end, when the cleanup action
was run.
I could not seem to find a bang pattern that avoided the leak, so I
resorted to a IORef, rather clunky, but not a performance problem because
it will only be written once per git ls-files, so typically just 1 time.
This commit was sponsored by Mark Reidenbach on Patreon.
Anything that needs to examine the file content will fail to match,
or fall back to other available information. But the intent is that the
matcher be checked for matchNeedsFileContent and only be used if it does
not, so the exact behavior doesn't much matter as it should never
happen.
The real point of this is to not need to provide a dummy content file
when matching.
This commit was sponsored by Martin D on Patreon.
This was the last one marked as a zombie. There might be others I don't
know about, but except for in the hypothetical case of a thread dying
due to an async exception before it can wait on a process it started, I
don't know of any.
It would probably be safe to remove the reapZombies now, but let's wait
and so that in its own commit in case it turns out to cause problems.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
Sped up seeking for files to operate on, when using options like --copies
or --in, by around 20%.
Benchmark showed an increase for --copies from 155 seconds to 121
seconds, and --in remote will be similar to that.
For --in here, the speedup was less, 5-10% or so.
(both warm cache)
This commit was sponsored by Jack Hill on Patreon.
Before, -J1 was different than no -J: It makes concurrent-output be used
for display, and it actually can run concurrent jobs in some situations.
Eg, a perform stage and a cleanup stage can both run.
dupState is used in several places, and changes NonConcurrent to
Concurrent 1. My concern is that this might, in some case, enable
that concurrent behavior. And in particular, that it might get enabled in
--batch mode, when the user is not expecting concurrent output because
they did not pass -J.
While I don't have a test case where that happens and causes out of
order output, it looks like it could, and so prudent to make this change.
--batch combined with -J now runs batch requests concurrently for many
commands. Before, the combination was accepted, but did not enable
concurrency. Since the output of batch requests can be in any order, --json
with the new "input" field is recommended to be used, to determine which
batch request each response corresponds to.
If --json is not used, batch mode still runs concurrently, using the usual
concurrent-output. That will not be very useful for most batch mode users,
probably, but who knows.
If a program was using --batch -J before, and was parsing non-json output,
this could break it. But, it was relying on git-annex not supporting
concurrency despite it being enabled, so it should have expected concurrent
output. So, I think that's ok.
annex.jobs does not enable concurrency in --batch mode, because that would
confuse programs that use --batch but don't expect concurrency.
This is prep for making batchCommandAction use commandAction,
which will enable concurrency for batch mode. Since commandAction can't
return anything, have to handle the case of a CommandStart that chooses
to do nothing in a different way.
No behavior changes (hopefully), just adding SeekInput and plumbing it
through to the JSON display code for later use.
Over the course of 2 grueling days.
withFilesNotInGit reimplemented in terms of seekHelper
should be the only possible behavior change. It seems to test as
behaving the same.
Note that seekHelper dummies up the SeekInput in the case where
segmentPaths' gives up on sorting the expanded paths because there are
too many input paths. When SeekInput later gets exposed as a json field,
that will result in it being a little bit wrong in the case where
100 or more paths are passed to a git-annex command. I think this is a
subtle enough problem to not matter. If it does turn out to be a
problem, fixing it would require splitting up the input
parameters into groups of < 100, which would make git ls-files run
perhaps more than is necessary. May want to revisit this, because that
fix seems fairly low-impact.
Avoid complaining that a file with "is beyond a symbolic link" when the
filepath is absolute and the symlink in question is not actually inside the
git repository.
This assumes that inodes remain stable while the command is running.
I think they always will, the filesystems where they are unstable change
them across mounts. (If inodes were not stable, it would just complain about
symlinks in the path that are not inside the working tree.)
(On windows, I don't want to assume anything about inodes, they could be
random numbers for all I know. But if they were, this would still be ok, as
long as windows doesn't have symlinks that are detected by isSymbolicLink.
Which seems a fair bet.)
Part of workTreeItems is trying detect a case
where git porcelain refuses to process a file, and where
git ls-files silently outputs nothing. But, it's hard to perfectly
replicate git's behavior, and besides, git's behavior could change.
So it could be that we warn, but then git ls-files does not skip over
it, and so git-annex also processes it after warning about it.
So, if we think we have a problem with a parameter, display the warning,
and skip processing it at all.
Implementing this was complicated by needing to handle the case where
all command-line parameters get filtered out this way. Which is
different than the case where there are none, because we don't want to
operate on all files in this new case..
And add a test case for that.
This certianly loses some of the 2x performance improvement in file
seeking that seekFilteredKeys led to, because now it has to stat the
worktree files again. Without benchmarking, I expect there will still be
a sizable improvement, and also the git-annex branch precaching that
seekFilteredKeys can do will still be a win of its approach.
Also worth noting that lookupKey, when the file DNE, check if it's in an
adjusted branch with hidden files, and if so, finds the key for the
file anyway. That was intended to make git-annex sync --content be able
to process those files, but a side effect was that, when a file was
deleted but the deletion not yet staged, git-annex commands used to
still list it. That was actually a bug. This commit fixes that bug too.
(git-annex sync --content on such a branch does not use seekFilteredKeys
so was not affected by the reversion or by this behavior change)
This commit was sponsored by Jake Vosloo on Patreon.
This was a bit disappointing, I was hoping for a 2x speedup. But, I think
the metadata lookup is wasting a lot of time and also needs to be made to
stream.
The changes to catObjectStreamLsTree were benchmarked to not also speed
up --all around 3% more. Seems I managed to make it polymorphic after all.
This solves the same problem as commit b4d0f6dfc2
but in a better way, that should make processing pointer files maximally
fast. If there is a mixture of pointer files and symlinks, the first
symlinks until the pointer file are handled maximally fast, while the
ones after that go via the slightly slower path.
This removes all calls to inAnnex, except for some involving --batch.
It may be that the batch code could get a similar speedup, but I don't
know if people habitually pass a huge number of files through --batch
that git-annex does not need to do anything to process, so I skipped it
for now.
A few calls to ifAnnexed remain, and might be worth doing more to
convert. In particular, Command.Sync has one that would probably speed
it up by a good amount.
(also removed some dead code from Command.Lock)
This is all good, except for one small problem... When a pointer file
has to be fed into the metadata cat-file, it's possible for a
non-pointer file that comes after it to get fed into the main cat-file
first, so the two files will be processed in a different order than the
user specified.
So, while this is the fast way, I guess I'll have to change it to be
slower, but sequential..
This is only implemented for git-annex get so far. It makes git-annex
get nearly twice as fast in a repo with 10k files, all of them present!
But, see the TODO for some caveats.
This assumes that no location log files will have a newline or carriage
return in their name. catObjectStream skips any such files due to
cat-file not supporting them.
Keys have been prevented from containing newlines since 2011,
commit 480495beb4. If some old repo
had a key with a newline in it, --all will just skip processing that key.
Other things, like .git/annex/unused files certianly assume no newlines in
keys too, and AFAICR, such keys never actually worked.
Carriage return is escaped by preSanitizeKeyName since 2013. WORM keys
generated before that point could perhaps contain a CR. (URL probably not,
http probably doesn't support an URL with a raw CR in it.) So, added
a warning in fsck about such keys. Although, fsck --all will naturally
skip them, so won't be able to warn about them. Not entirely
satisfactory, but I'll bet there are not really any such keys in
existence.
Thanks to Lukey for finding this optimisation.
The cache was removed way back in 2012,
commit 3417c55189
Then I forgot I had removed it! I remember clearly multiple times when I
thought, "this reads the same data twice, but the cache will avoid that
being very expensive".
The reason it was removed was it messed up the assistant noticing when
other processes made changes. That same kind of problem has recently
been addressed when adding the optimisation to avoid reading the journal
unnecessarily.
Indeed, enableInteractiveJournalAccess is run in just the
right places, so can just piggyback on it to know when it's not safe
to use the cache.
Always run Git.Config.store, so when the git config gets reloaded,
the override gets re-added to it, and changeGitRepo then calls extractGitConfig
on it and sees the annex.* settings from the override.
Remove any prior occurance of -c v and add it to the end. This way,
-c foo=1 -c foo=2 -c foo=1 will pass -c foo=1 to git, rather than -c foo=2
Note that, if git had some multiline config that got built up by
multiple -c's, this would not work still. But it never worked because
before the bug got fixed in the first place, the -c value was repeated
many times, so the multivalue thing would have been wrong. I don't think
-c can be used with multiline configs anyway, though git-config does
talk about them?
Added annex.skipunknown git config, that can be set to false to change the
behavior of commands like `git annex get foo*`, to not skip over files/dirs
that are not checked into git and are explicitly listed in the command
line.
Significant complexity was needed to handle git-annex add, which uses some
git ls-files calls, but needs to not use --error-unmatch because of course
the files are not known to git.
annex.skipunknown is planned to change to default to false in a
git-annex release in early 2022. There's a todo for that.
Fix a crash or potentially not all files being exported when sync -J
--content is used with an export remote.
Crash as described in fixed bug report.
waitForAllRunningCommandActions inserted in several points where all the
commandActions started before need to have finished before moving on to
the next stage of the export. A race across those points could have
maybe resulted in not all files being exported, or a wrong tree being
export.
For example, changeExport starting up an action like
a rename of A to B. Then, with that action still running, fillExport
uploading a new A, *before* the rename occurred. That race seems
unlikely to have happened. There are some other ones that this also
fixes.
This relicates git's behavior. It adds a few stat calls for the command
line parameters, so there is some minor slowdown, but even with thousands
of parameters it will not be very noticable, and git does the same statting
in similar circumstances.
Note that this does not prevent eg "git annex add symlink"; the symlink
will be added to git as usual. And "git annex find symlink" will silently
list nothing as well. It's only "symlink/foo" or "subdir/symlink/foo" that
triggers the warning.
Avoid running a large number of git cat-file child processes when run with
a large -J value.
This implementation takes care to avoid adding any overhead to git-annex
when run without -J. When run with -J, there is a small bit of added
overhead, to manipulate the resource pool. That optimisation added a
fair bit of complexity.
This does mean that RemoteDaemon.Transport.Tor's call runs it, otherwise
no change, but this is groundwork for doing more such expensive actions
in dupState.
Git has an obnoxious special case in git config, a line "foo" is the same
as "foo = true". That means there is no way to examine the output of
git config and tell if it was run with --null or not, since a "foo"
in the first line could be such a boolean, or could be followed by its
value on the next line if --null were used.
So, rather than trying to do such a detection, track the style of config
at all the points where it's generated.
The only price paid is one additional MVar read per write to the journal.
Presumably writing a journal file dominiates over a MVar read time by
several orders of magnitude.
--batch does not get the speedup because then it needs to notice when
another process has made a change. Also made the assistant and other damon
modes bypass the optimisation, which would not help them anyway.
The 'fail' method has been moved to the 'MonadFail' class. I made the changes
so that the code still compiles with previous versions of 'base' that don't
have the new MonadFail class exported by Prelude yet.
It's important that it be clear that it overrides a config, such that
reloading the git config won't change it, and in particular, setConfig
won't change it.
Most of the calls to changeGitConfig were actually after setConfig,
which was redundant and unncessary. So removed those.
The only remaining one, besides --debug, is in the handling of
repository-global config values. That one's ok, because the
way mergeGitConfig is implemented, it does not override any value that
is set in git config. If a value with a repo-global setting was passed
to setConfig, it would set it in the git config, reload the git config,
re-apply mergeGitConfig, and use the newly set value, which is the right
thing.
The git add behavior changes could be avoided if it turns out to be
really annoying, but then it would need to behave the old way when
annex.dotfiles=false and the new way when annex.dotfiles=true. I'd
rather not have the config option result in such divergent behavior as
`git annex add .` skipping a dotfile (old) vs adding to annex (new).
Note that the assistant always adds dotfiles to the annex.
This is surprising, but not new behavior. Might be worth making it also
honor annex.dotfiles, but I wonder if perhaps some user somewhere uses
it and keeps large files in a directory that happens to begin with a
dot. Since dotfiles and dotdirs are a unix culture thing, and the
assistant users may not be part of that culture, it seems best to keep
its current behavior for now.
git-annex find is now RawFilePath end to end, no string conversions.
So is git-annex get when it does not need to get anything.
So this is a major milestone on optimisation.
Benchmarks indicate around 30% speedup in both commands.
Probably many other performance improvements. All or nearly all places
where a file is statted use RawFilePath now.
Adds a dependency on filepath-bytestring, an as yet unreleased fork of
filepath that operates on RawFilePath.
Git.Repo also changed to use RawFilePath for the path to the repo.
This does eliminate some RawFilePath -> FilePath -> RawFilePath
conversions. And filepath-bytestring's </> is probably faster.
But I don't expect a major performance improvement from this.
This is mostly groundwork for making Annex.Location use RawFilePath,
which will allow for a conversion-free pipleline.
The parser and looking up config keys in the map should both be faster
due to using ByteString.
I had hoped this would speed up startup time, but any improvement to
that was too small to measure. Seems worth keeping though.
Note that the parser breaks up the ByteString, but a config map ends up
pointing to the config as read, which is retained in memory until every
value from it is no longer used. This can change memory usage
patterns marginally, but won't affect git-annex.
Finally builds (oh the agoncy of making it build), but still very
unmergable, only Command.Find is included and lots of stuff is badly
hacked to make it compile.
Benchmarking vs master, this git-annex find is significantly faster!
Specifically:
num files old new speedup
48500 4.77 3.73 28%
12500 1.36 1.02 66%
20 0.075 0.074 0% (so startup time is unchanged)
That's without really finishing the optimization. Things still to do:
* Eliminate all the fromRawFilePath, toRawFilePath, encodeBS,
decodeBS conversions.
* Use versions of IO actions like getFileStatus that take a RawFilePath.
* Eliminate some Data.ByteString.Lazy.toStrict, which is a slow copy.
* Use ByteString for parsing git config to speed up startup.
It's likely several of those will speed up git-annex find further.
And other commands will certianly benefit even more.
Goal is to make git-annex faster by using ByteString for all the
worktree traversal. For now, this is focusing on Command.Find,
in order to benchmark how much it helps. (All other commands are
temporarily disabled)
Currently in a very bad unbuildable in-between state.
See the comment for a trace of the deadlock.
Added a new StartStage. New worker threads begin in the StartStage.
Once a thread is ready to do work, it moves away from the StartStage,
and no thread will ever transition back to it.
A thread that blocks waiting on another thread that is processing
the same key will block while in the StartStage. That other thread
will never switch back to the StartStage, and so the deadlock is avoided.
Drop support for building with ghc older than 8.4.4, and with older
versions of serveral haskell libraries than will be included in Debian 10.
The only remaining version ifdefs in the entire code base are now a couple
for aws!
This commit should only be merged after the Debian 10 release.
And perhaps it will need to wait longer than that; it would make
backporting new versions of git-annex to Debian 9 (stretch) which
has been actively happening as recently as this year.
This commit was sponsored by Ilya Shlyakhter.
When all worker threads are running and enteringStage is called,
it waits for an idle slot. If all off the other threads then call it in
turn, a deadlock occurrs.
This is the same problem I didn't actually fix in
5a9842d7ed.
Fixed by doing two separate STM transactions, the first replaces its
active thread with an idle thread, and the second waits for another idle
thread. That guarantees there will eventually be an idle thread to find.
The changes to WorkerPool were necessary because it can't add an idle
thread containing the Annex state and go on to run an action using that
same state, so I had to remove the Annex state from IdleWorker.
Rather than limiting it to PerformStage and CleanupStage, this opens it
up so any number of stages can be added as needed by commands.
Each concurrent command has a set of stages that it uses, and only
transitions between those can block waiting for a free slot in the
worker pool. Calling enteringStage for some other stage does not block,
and has very little overhead.
Note that while before the Annex state was duplicated on the first call
to commandAction, this now happens earlier, in startConcurrency.
That means that seek stage actions should that use startConcurrency
and then modify Annex state won't modify the state of worker threads
they then start. I audited all of them, and only Command.Seek
did so; prepMerge changes the working directory and so has to come
before startConcurrency.
Also, the remote list is built before duplicating the state, which means
that it gets built earlier now than it used to. This would only have an
effect of making commands that end up not needing to perform any actions
unncessary build the remote list (only when they're run with concurrency
enable), but that's a minor overhead compared to commands seeking
through the work tree and determining they don't need to do anything.
I couldn't find a way to avoid the deadlock w/o rewriting it to clearly
not have one. I'm not quite sure what was the actual cause of the
deadlock.
This makes me unsure how I now know it clearly doesn't have a
deadlock. But, it was easy to reproduce before (just call it twice in a
row) and doesn't happen now.
