However, I don't yet have a reliable way to deal with files being modified
while they're being transferred. I have code that detects it on the sending
side, but the receiver is still free to move the wrong content into its
annex, and record that it has the content. So that's not acceptable, and
I'll need to work on it some more.
However, at this point I can use a direct mode repository as a remote and
transfer files from and to it.
* get/copy --auto: Transfer data even if it would exceed numcopies,
when preferred content settings want it.
* drop --auto: Fix dropping content when there are no preferred content
settings.
Files are now written to a tmp directory in the remote, and once all
chunks are written, etc, it's moved into the final place atomically.
For now, checkpresent still checks every single chunk of a file, because
the old method could leave partially transferred files with some chunks
present and others not.
Inject the required git-remote-xmpp into PATH when running xmpp git push.
Rest of the time it will not be in PATH, and git won't be able to talk to
xmpp remotes.
I now have this topology working:
assistant ---> {bare repo, special remote} <--- assistant
And, I think, also this one:
+----------- bare repo --------+
v v
assistant ---> special remote <--- assistant
While before with assistant <---> assistant connections, both sides got
location info updated after a transfer, in this topology, the bare repo
*might* get its location info updated, but the other assistant has no way to
know that it did. And a special remote doesn't record location info,
so transfers to it won't propigate out location log changes at all.
So, for these to work, after a transfer succeeds, the git-annex branch
needs to be pushed. This is done by recording a synthetic commit has
occurred, which lets the pusher handle pushing out the change (which will
include actually committing any still journalled changes to the git-annex
branch).
Of course, this means rather a lot more syncing action than happened
before. At least the pusher bundles together very close together pushes,
somewhat. Currently it just waits 2 seconds between each push.
I had been using -ignore-package monads-tf to deal with this, but
the XMPP library uses monads-tf, so that also ignores it. Instead,
use PackageImports to force use of mtl in my own code.
This was complicated quite a bit by needing to check numcopies. I optimised
that, so it only looks up numcopies once per file, no matter how many
remotes it checks to drop from. Although it did just occur to me that
it might be better to first check if it wants to drop content, and only
then check numcopies..
When rsyncProgress pipes rsync's stdout, this turns out to cause a ssh
process started by rsync to be left behind as a zombie. I don't know why,
but my recent zombie reaping cleanup was correct, it's just that this other
zombie, that's not directly started by git-annex, was no longer reaped
due to changes in the cleanup. Make rsyncProgress reap the zombie started
by rsync, as a workaround.
FWIW, the process tree looks like this. It seems like the rsync child
is for some reason starting but not waiting on this extra ssh process.
Ssh connection caching may be involved -- disabling it seemed to change
the shape of the tree, but did not eliminate the zombie.
9378 pts/14 S+ 0:00 | \_ rsync -p --progress --inplace -4 -e 'ssh' '-S' ...
9379 pts/14 S+ 0:00 | | \_ ssh ...
9380 pts/14 S+ 0:00 | | \_ rsync -p --progress --inplace -4 -e 'ssh' '-S' ...
9381 pts/14 Z+ 0:00 | \_ [ssh] <defunct>
If the autostart file lists a repository, for which a directory exists,
but there's not actually a valid git repo in there, the web app used to
try to use it, and see it wasn't valid, and then try to autostart again.
The ensuing runaway loop also ate memory, although not as fast as I was led
to belive was happening to someone on IRC yesterday. So that guy may have
had a different problem. But this seems otherwise a reasonable fit for the
circumstances described, if git-annex was started before something that
occurred during desktop login that made the repository available.
Both when queueing downloads, and uploads, consults the preferred content
settings.
I didn't make it check yet when requeing failed transfers or queuing
deferred downloads; dealing with the preferred content settings (or indeed,
other settings) changing while the assistant is running still needs work.
I'm down to 9 places in the code that can produce unwaited for zombies.
Most of these are pretty innocuous, at least for now, are only
used in short-running commands, or commands that run a set of
actions and explicitly reap zombies after each one.
The one from Annex.Branch.files could be trouble later,
since both Command.Fsck and Command.Unused can trigger it,
and the assistant will be doing those eventally. Ditto the one in
Git.LsTree.lsTree, which Command.Unused uses.
The only ones currently affecting the assistant though, are
in Git.LsFiles. Several threads use several of those.
(And yeah, using pipes or ResourceT would be a less ad-hoc approach,
but I don't really feel like ripping my entire code base apart right
now to change a foundation monad. Maybe one of these days..)
Nearly everything that's reading from git is operating on a small
amount of output and has been switched to use that. Only pipeNullSplit
stuff continues using the lazy version that yields zombies.
This includes a full parser for the boolean expressions in the log,
that compiles them into Matchers. Those matchers are not used yet.
A complication is that matching against an expression should never
crash git-annex with an error. Instead, vicfg checks that the expressions
parse. If a bad expression (or an expression understood by some future
git-annex version) gets into the log, it'll be ignored.
Most of the code in Limit couldn't fail anyway, but I did have to make
limitCopies check its parameter first, and return an error if it's bad,
rather than erroring at runtime.
One note: Deleted lines are not currently parsed as config changes.
That makes sense for trust settings. It may make sense to support deleted
lines as a way to clear group settings.
Incomplete; I need to finish parsing and saving. This will also be used
for editing transfer control expresssions.
Removed the group display from the status output, I didn't really
like that format, and vicfg can be used to see as well as edit rempository
group membership.
Simplified it using existing functions.
I doubt setSticky needs to return the FileMode; if it does for some
reason, it can be changed to use modifyFileMode'
Converted isSticky to a pure function for consistency with isSymlink.
Note that the sticky bit of a file can be tested thus:
isSticky . fileMode <$> getFileStatus file
Fix resuming of downloads, which do not have a transfer info file to read.
When checking upload progress, use the MVar, rather than re-reading
the info file.
Catch exceptions in the transfer action. Required a tryAnnex.
When a transfer fails, the progress info can be used to intelligently
retry it. If the transfer managed to make some progress, but did not
fully complete, then there's a good chance that a retry will finish it
(or at least make more progress).
Transfer info files are updated when the callback is called, updating
the number of bytes transferred.
Left unused p variables at every place the callback should be used.
Which is rather a lot..
Don't expose these as branches in refs/heads/. Instead hide them away in
refs/synced/ where only show-ref will find them.
Make unused only look at branches and tags, not these other things,
so it won't care if some stale sync ref used to use a file.
This means they don't need to be deleted, which could have
led to an incoming sync being missed.
This fixes a problem I was seeing in the assistant where two remotes would
attempt to sync with one another at the same time, and both failed pushing
the diverged git-annex branch. Then when both tried to resolve the failed
push, they each modified their git-annex branch, which again each blocked
the other from pushing into it. The result was that the git-annex
branches were perpetually diverged (despite having the same content!) and
once the assistant fell into this trap, it couldn't get out and always
had to do the slow push/fail/pull/merge/push/fail cycle.
They work fine. But I had to go to a lot of trouble to get Yesod to render
routes in a pure function. It may instead make more sense to have each
alert have an assocated IO action, and a single route that runs the IO
action of a given alert id. I just wish I'd realized that before the past
several hours of struggling with something Yesod really doesn't want to
allow.
Used by the assistant, rather than copy, this is faster because it avoids
using git ls-files, avoids checking the location log redundantly, and
runs in oneshot mode, avoiding making a commit to the git-annex branch
for every file transferred.
Found a very cheap way to determine when a disconnected remote has
diverged, and has new content that needs to be transferred: Piggyback on
the git-annex branch update, which already checks for divergence.
However, this does not check if new content has appeared locally while
disconnected, that should be transferred to the remote.
Also, this does not handle cases where the two git repos are in sync,
but their content syncing has not caught up yet.
This code could have its efficiency improved:
* When multiple remotes are synced, if any one has diverged, they're
all queued for transfer scans.
* The transfer scanner could be told whether the remote has new content,
the local repo has new content, or both, and could optimise its scan
accordingly.
This commit includes a paydown on technical debt incurred two years ago,
when I didn't know that it was bad to make custom Read and Show instances
for types. As the routes need Read and Show for Transfer, which includes a
Key, and deriving my own Read instance of key was not practical,
I had to finally clean that up.
So the compact Key read and show functions are now file2key and key2file,
and Read and Show are now derived instances.
Changed all code that used the old instances, compiler checked.
(There were a few places, particularly in Command.Unused, and the test
suite where the Show instance continue to be used for legitimate
comparisons; ie show key_x == show key_y (though really in a bloom filter))
Avoid crashing when "git annex get" fails to download from one location,
and falls back to downloading from a second location.
The problem is that git annex get calls download recursively from within
itself if the first download attempt fails. So the first time through, it
writes a transfer info file, which is then overwritten on the second,
recursive call. Then on cleanup, it tries to delete the file twice, which
of course doesn't work.
Fixed both by not crashing if the transfer file is removed, and by
changing Get to not run download recursively like that. It's the only
thing that did so, and it just seems like a bad idea.
git annex assistant --autostart will start separate daemons in each
listed autostart repo
running the webapp outside any git-annex repo will open it on the
first listed autostart repo
This prevents multiple runs of the assistant in the foreground, and lets
--stop stop foregrounded runs too.
The webapp firstrun case also now writes a pid file, once it's made the git
repo to put it in.
This avoids forking another process, avoids polling, fixes a race,
and avoids a rare forkProcess thread hang that I saw once time
when starting the webapp.
Make Utility.Process wrap the parts of System.Process that I use,
and add debug logging to them.
Also wrote some higher-level code that allows running an action
with handles to a processes stdin or stdout (or both), and checking
its exit status, all in a single function call.
As a bonus, the debug logging now indicates whether the process
is being run to read from it, feed it data, chat with it (writing and
reading), or just call it for its side effect.
Test suite now passes with -threaded!
I traced back all the hangs with -threaded to System.Cmd.Utils. It seems
it's just crappy/unsafe/outdated, and should not be used. System.Process
seems to be the cool new thing, so converted all the code to use it
instead.
In the process, --debug stopped printing commands it runs. I may try to
bring that back later.
Note that even SafeSystem was switched to use System.Process. Since that
was a modified version of code from System.Cmd.Utils, it needed to be
converted too. I also got rid of nearly all calls to forkProcess,
and all calls to executeFile, which I'm also doubtful about working
well with -threaded.
This *almost* works.
Along the way, I noticed that the --uuid parameter was being accidentially
passed after the --, so that has never been actually used by
git-annex-shell to verify it's running in the expected repository. Oops. Fixed.
Not yet tested and places git-annex-shell is run need to be modified to
pass the new field settings.
Note that rsyncServerSend was changed to fork, rather than directly exec
rsync, because it needs to keep the transfer lock held, and clean up the
transfer log when done.
In order to record a semi-useful filename associated with the key,
this required plumbing the filename all the way through to the remotes'
storeKey and retrieveKeyFile.
Note that there is potential for deadlock here, narrowly avoided.
Suppose the repos are A and B. A sends file foo to B, and at the same
time, B gets file foo from A. So, A locks its upload transfer info file,
and then locks B's download transfer info file. At the same time,
B is taking the two locks in the opposite order. This is only not a
deadlock because the lock code does not wait, and aborts. So one of A or
B's transfers will be aborted and the other transfer will continue.
Whew!
Note this is per-remote, so trying to get the same file from multiple
remotes can still let duplicate downloads run. (And uploading the same file
to multiple remotes is not duplicate at all of course.)
get, move, and copy are the only git-annex subcommands that transfer
files, but there's still git-annex-shell recvkey and sendkey to deal with too.
I considered modifying retrieveKeyFile or getViaTmp, but they are called
by other code that does not involve expensive file transfers (migrate)
or that does file transfers that should not be checked by this (fsck --from).
While this word may be less familiar to some users, it avoids the
connotation that version 2 is better than version 1, which is wrong
when the two variants were conflicting.
Note that, since this always pushes branch synced/master to the remote, it
assumes that master has already gotten all the commits that are on the
remote merged in. Otherwise, fast-forward prevention may prevent the push.
That's probably ok, because the next stage is to automatically detect
incoming pushes and merge.
Kqueue needs to remember which files failed to be added due to being open,
and retry them. This commit gets the data in place for such a retry thread.
Broke KeySource out into its own file, and added Eq and Ord instances
so it can be stored in a Set.
Was decoding the git-cat-file of the symlink target as utf8, but that can't
do, unix filenames are from the 70's and need this shiny disco
fileSystemEncoding.
Now it starts really, really fast! Down from 15 minutes or so on my big
tree to around 1 minute.
The trick is to remember the last time the daemon was running. Links with a
ctime from before that point don't need to be restaged on startup (as long
as they are correct), since the old daemon would have handled them already.
We also assume that if the daemon has never run before, any links that
already exist are good. The pre-commit hook fixes links, so this should be
a safe assumption.
Adds another MVar holding a DaemonStatus data structure. Also
allowed getting rid of the Annex.Fast hack. This data structure will
probably grow a lot of details about the daemon's status, that will
later be used by the webapp's UI.
The code to actually track when the daemon was last running is not written
yet. It's 3 am.
The idea, not yet done, is to use this to detect when a file
has an old change time, and avoid expensive restaging of the file.
If git-annex watch keeps track of the last time it finished a full scan,
then any symlink that is older than that time must have been scanned
before, so need not be added. (Relying on moving, copying, etc of a file
all updating its change time.)
Anyway, this info is available for free since inotify already checks it,
so it might as well make it available.
Now really only done in the startup scan.
It turns out to be quite hard for event handlers to know when the startup
scan is complete. I tried to make addWatch pass that info, but found
threading the state very difficult. For now, a quick hack, using the fast
flag.
Note that it's actually possible for inotify events to come in while the
startup scan is still ongoing. Due to my hack, the expensive check will
be done for files added in such inotify events.
This requires a relatively expensive test at file add time to see if it's
in git already. But it can be optimised to only happen during the startup
scan.
I thought this might be a lock conflict that explains the deadlock when
built with -threaded, but it seems not.. it still locks! It even locks
without the committer thread.
Indeed, it locks when running "git annex add"! -threaded is exposing some
other problem.
Still, this seems conceptually cleaner and did not add any inneficiencies.
Also added some high-level documentation about the threads used.
The commit thread now has access to a channel containing the times of
all uncommitted changes. This lets it be smart about detecting busy times
when a batch job is running (such as rm -rf, or untarring something, etc),
and avoid committing until it's done. While at the same time, instantly
committing one-off changes that the user is going to expect to see
immediately.
I had to use STM to implement the channel, because of
http://hackage.haskell.org/trac/ghc/ticket/4154
While this adds a dependency, I always wanted to use STM, so this actually
makes me happy. ;)
Also happy that shouldCommit is a pure function, so other commit smartness
strategies can easily be played with. Although the current one seems pretty
good.
There is one bug, for some reason it does double commits, every time.
While I was in there, I noticed and fixed a bug in the queue size
calculations. It was never encountered only because Queue.add was
only ever run with 1 file in the list.
This ensures that all special remotes show up in git annex status.
Before, a special remote that was not manually described, and was not
a current git remote, did not show up there, although initremote did list
it.
There's a race adding a new file to the annex: The file is moved to the
annex and replaced with a symlink, and then we git add the symlink. If
someone comes along in the meantime and replaces the symlink with
something else, such as a new large file, we add that instead. Which could
be bad..
This race is fixed by avoiding using git add, instead the symlink is
directly staged into the index.
It would be nice to make `git annex add` use this same technique.
I have not done so yet because it currently runs git update-index once per
file, which would slow does `git annex add`. A future enhancement would be
to extend the Git.Queue to include the ability to run update-index with
a list of Streamers.
Anything that tries to open the file for write, or delete the file,
or replace it with something else, will not affect the add.
Only if a process has the file open for write before add starts
can it still change it while (or after) it's added to the annex.
(fsck will catch this later of course)
Uses a MVar again, as there seems no other way to thread the state through
inotify events.
This is a rather unsatisfactory result. I had wanted to run them in
the same monad so that the git queue could be used to coleasce git commands
and speed things up. But, that led to fragility: If several files are
added, and one is removed before queue flush, git add will fail to add
any of them. So, the queue is still explicitly flushed after each add for
now.
TODO: Investigate using git add --ignore-errors. This would need to be done
in Command.Add. And, git add still exits nonzero with it, so would need
to avoid crashing on queue flush.
When a new file is annexed, a deletion event occurs when it's moved away
to be replaced by a symlink. Most of the time, there is no problimatic
race, because the same thread runs the add event as the deletion event.
So, once the symlink is in place, the deletion code won't run at all,
due to existing checks that a deleted file is really gone.
But there is a race at startup, as then the inotify thread is running
at the same time as the main thread, which does the initial tree walking
and annexing. It would be possible for the deletion inotify to run
in a perfect race with the addition, and remove the newly added symlink
from the git cache.
To solve this race, added event serialization via a MVar. We putMVar
before running each event, which blocks if an event is already running.
And when an event finishes (or crashes!), we takeMVar to free the lock.
Also, make rm -rf not spew warnings by passing --ignore-unmatch when
deleting directories.
And just like that, annexed files can be moved and copies around within
the tree, and are automatically fixed to point to the content, and staged
in git. Huzzah!
Delete still remains TODO, with its troublesome race during add..
Improved the inotify code, so it will also notice directory removal
and symlink creation.
In the watch code, optimised away a stat of a file that's being added,
that's done by Command.Add.start. This is the reason symlink creation is
handled separately from file creation, since during initial tree walk
at startup, a stat was already done, and can be reused.
Resetting an unlocked file to the branch head failed if it had just been
added, not committed, and unlocked, since the branch didbn't have it.
The code was concerned about dropping any changes that might be staged in the
index, but I cannot see why.
Baked into the code was an assumption that a repository's git directory
could be determined by adding ".git" to its work tree (or nothing for bare
repos). That fails when core.worktree, or GIT_DIR and GIT_WORK_TREE are
used to separate the two.
This was attacked at the type level, by storing the gitdir and worktree
separately, so Nothing for the worktree means a bare repo.
A complication arose because we don't learn where a repository is bare
until its configuration is read. So another Location type handles
repositories that have not had their config read yet. I am not entirely
happy with this being a Location type, rather than representing them
entirely separate from the Git type. The new code is not worse than the
old, but better types could enforce more safety.
Added support for core.worktree. Overriding it with -c isn't supported
because it's not really clear what to do if a git repo's config is read, is
not bare, and is then overridden to bare. What is the right git directory
in this case? I will worry about this if/when someone has a use case for
overriding core.worktree with -c. (See Git.Config.updateLocation)
Also removed and renamed some functions like gitDir and workTree that
misused git's terminology.
One minor regression is known: git annex add in a bare repository does not
print a nice error message, but runs git ls-files in a way that fails
earlier with a less nice error message. This is because before --work-tree
was always passed to git commands, even in a bare repo, while now it's not.
annex.ssh-options, annex.rsync-options, annex.bup-split-options.
And adjust types to avoid the bugs that broke several config settings
recently. Now "annex." prefixing is enforced at the type level.
This is incomplete, it does not honor it yet for hash directories
and other annex bookkeeping files. Some of that is not needed for a bare
repo; some of it may be.
So far this only handles auto-annexing new files that are created inside
the repository while it's running. To make this really useful,
it needs to at least:
- notice deleted files and stage the deletion
(tricky; there's a race with add..)
- notice renamed files, auto-fix the symlink, and stage the new file location
- periodically auto-commit staged changes
- honor .gitignore, not adding files it excludes
Also nice to have would be:
- Somehow sync remotes, possibly using a push sync like dvcs-autosync
does, so they are immediately updated.
- Somehow get content that is unavilable. This is problimatic with inotify,
since we only get an event once the user has tried (and failed) to read
from the file. Perhaps instead, automatically copy content that is added
out to remotes, with the goal of all repos eventually getting a copy,
if df allows.
- Drop files that have not been used lately, or meet some other criteria
(as long as there's a copy elsewhere).
- Perhaps automatically dropunused files that have been deleted,
although I cannot see a way to do that, since by the time the inotify
deletion event arrives, the file is deleted, and we cannot see what
its symlink pointed to! Alternatievely, perhaps automatically
do an expensive unused/dropunused cleanup process.
Some of this probably needs the currently stateless threads to maintain
a common state.
getConfig got a remote-specific config, and this confusing name caused it
to be used a couple of places that only were interested in global configs.
Rename to getRemoteConfig and make getConfig only get global configs.
There are no behavior changes here, but remote.<name>.annex-web-options
never actually worked (and per-remote web options is a very unlikely to be
useful case so I didn't make it work), so fix the documentation for it.
