It was possible for the export.log to get written and then git-annex was
interrupted, before it could graft in the exported tree. Which could
result in export.log referencing a tree that got garbage collected.
This is most of the way there, but not quite working.
The layout of migrate.tree/ needs to be changed to follow this approach.
git log will list all the files in tree order, so the new layout needs
to alternate old and new keys. Can that be done? git may not document
tree order, or may not preserve it here.
Alternatively, change to using git log --format=raw and extract
the tree header from that, then use
git diff --raw $tree:migrate.tree/old $tree:migrate.tree/new
That will be a little more expensive, but only when there are lots of
migrations.
Sponsored-by: Joshua Antonishen on Patreon
Noticed that Semigroup instance of Map is not suitable to use
for MapLog. For example, it behaved like this:
ghci> parseTrustLog "foo 1 timestamp=10\nfoo 2 timestamp=11" <> parseTrustLog "foo X timestamp=12"
fromList [(UUID "foo",LogEntry {changed = VectorClock 11s, value = SemiTrusted})]
Which was wrong, it lost the newer DeadTrusted value.
Luckily, nothing used that Semigroup when operating on a MapLog. And this
provides a safe instance.
Sponsored-by: Graham Spencer on Patreon
* Deal with clock skew, both forwards and backwards, when logging
information to the git-annex branch.
* GIT_ANNEX_VECTOR_CLOCK can now be set to a fixed value (eg 1)
rather than needing to be advanced each time a new change is made.
* Misuse of GIT_ANNEX_VECTOR_CLOCK will no longer confuse git-annex.
When changing a file in the git-annex branch, the vector clock to use is now
determined by first looking at the current time (or GIT_ANNEX_VECTOR_CLOCK
when set), and comparing it to the newest vector clock already in use in
that file. If a newer time stamp was already in use, advance it forward by
a second instead.
When the clock is set to a time in the past, this avoids logging with
an old timestamp, which would risk that log line later being ignored in favor
of "newer" line that is really not newer.
When a log entry has been made with a clock that was set far ahead in the
future, this avoids newer information being logged with an older timestamp
and so being ignored in favor of that future-timestamped information.
Once all clocks get fixed, this will result in the vector clocks being
incremented, until finally enough time has passed that time gets back ahead
of the vector clock value, and then it will return to usual operation.
(This latter situation is not ideal, but it seems the best that can be done.
The issue with it is, since all writers will be incrementing the last
vector clock they saw, there's no way to tell when one writer made a write
significantly later in time than another, so the earlier write might
arbitrarily be picked when merging. This problem is why git-annex uses
timestamps in the first place, rather than pure vector clocks.)
Advancing forward by 1 second is somewhat arbitrary. setDead
advances a timestamp by just 1 picosecond, and the vector clock could
too. But then it would interfere with setDead, which wants to be
overrulled by any change. So it could use 2 picoseconds or something,
but that seems weird. It could just as well advance it forward by a
minute or whatever, but then it would be harder for real time to catch
up with the vector clock when forward clock slew had happened.
A complication is that many log files contain several different peices of
information, and it may be best to only use vector clocks for the same peice
of information. For example, a key's location log file contains
InfoPresent/InfoMissing for each UUID, and it only looks at the vector
clocks for the UUID that is being changed, and not other UUIDs.
Although exactly where the dividing line is can be hard to determine.
Consider metadata logs, where a field "tag" can have multiple values set
at different times. Should it advance forward past the last tag?
Probably. What about when a different field is set, should it look at
the clocks of other fields? Perhaps not, but currently it does, and
this does not seems like it will cause any problems.
Another one I'm not entirely sure about is the export log, which is
keyed by (fromuuid, touuid). So if multiple repos are exporting to the
same remote, different vector clocks can be used for that remote.
It looks like that's probably ok, because it does not try to determine
what order things occurred when there was an export conflict.
Sponsored-by: Jochen Bartl on Patreon
This adds a separate journal, which does not currently get committed to
an index, but is planned to be committed to .git/annex/index-private.
Changes that are regarding a UUID that is private will get written to
this journal, and so will not be published into the git-annex branch.
All log writing should have been made to indicate the UUID it's
regarding, though I've not verified this yet.
Currently, no UUIDs are treated as private yet, a way to configure that
is needed.
The implementation is careful to not add any additional IO work when
privateUUIDsKnown is False. It will skip looking at the private journal
at all. So this should be free, or nearly so, unless the feature is
used. When it is used, all branch reads will be about twice as expensive.
It is very lucky -- or very prudent design -- that Annex.Branch.change
and maybeChange are the only ways to change a file on the branch,
and Annex.Branch.set is only internal use. That let Annex.Branch.get
always yield any private information that has been recorded, without
the risk that Annex.Branch.set might be called, with a non-private UUID,
and end up leaking the private information into the git-annex branch.
And, this relies on the way git-annex union merges the git-annex branch.
When reading a file, there can be a public and a private version, and
they are just concacenated together. That will be handled the same as if
there were two diverged git-annex branches that got union merged.
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.
Fixed that, and made parserLsTree accept the space as well as tab.
Fixes a reversion that made import of a tree from a special remote result in
a merge that deleted files that were not preferred content of that special
remote.
Especially from borg, where the content identifier logs
all end up being the same identical file!
But also, for other imports, the location tracking logs can,
in some cases, be identical files.
Bonus optimisation: Avoid looking up (and parsing when set)
GIT_ANNEX_VECTOR_CLOCK env var every time a log is written to.
Although the lookup does happen at startup even when no
log will be written now.
This makes sync a lot faster in the common case where there's no new
backup.
There's still room for it to be faster. Currently the old imported tree
has to be traversed, to generate the ImportableContents. Which then
gets turned around to generate the new imported tree, which is
identical. So, it would be possible to just return a "no new imports",
or an ImportableContents that has a way to graft in a tree. The latter
is probably too far to go to optimise this, unless other things need it.
The former might be worth it, but it's already pretty fast, since git
ls-tree is pretty fast.
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.
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.
Prevents merging the import from deleting the non-preferred files from
the branch it's merged into.
adjustTree previously appended the new list of items to the old, which
could result in it generating a tree with multiple files with the same
name. That is not good and confuses some parts of git. Gave it a
function to resolve such conflicts.
That allowed dealing with the problem of what happens when the import
contains some files (or subtrees) with the same name as files that were
filtered out of the export. The files from the import win.
This does not change the overall license of the git-annex program, which
was already AGPL due to a number of sources files being AGPL already.
Legally speaking, I'm adding a new license under which these files are
now available; I already released their current contents under the GPL
license. Now they're dual licensed GPL and AGPL. However, I intend
for all my future changes to these files to only be released under the
AGPL license, and I won't be tracking the dual licensing status, so I'm
simply changing the license statement to say it's AGPL.
(In some cases, others wrote parts of the code of a file and released it
under the GPL; but in all cases I have contributed a significant portion
of the code in each file and it's that code that is getting the AGPL
license; the GPL license of other contributors allows combining with
AGPL code.)
Like the earlier fixed one in Command.Export, it occurred when the same
tree was exported by multiple clones. Previous fix was incomplete since
several other places looked at the list of exported trees to detect when
there was an export conflict. Added a single unified function to avoid
missing any places it needed to be fixed.
This commit was sponsored by mo on Patreon.
There should be some speed gains here, especially for chunk and remote
state logs, which are queried once per key.
Now only old-format uuid-based logs still need to be converted to attoparsec.
Mostly didn't push the ByteStrings down very deep, but all of these log
files are not written to frequently at all, so slight remaining
innefficiency doesn't matter.
In Logs.UUID, removed the fixBadUUID code that cleaned up after a bug in
git-annex versions 3.20111105-3.20111110. In the unlikely event that a repo was
last touched by that ancient git-annex version, the descriptions of remotes
would appear missing when used with this version of git-annex. That is such minor
breakage, and so unlikely to still be a problem for any repos, that it was not
worth forward-porting that code to ByteString.
Probably not any particular speedup in this, since most of these logs
are not written to often. Possibly chunk log writing is sped up, but
writes to chunk logs are interleaved with expensive data transfers to
remotes, so unlikely to be a noticiable speedup.
Most of the individual logs are not converted yet, only presense logs
have an efficient ByteString Builder implemented so far. The rest
convert to and from String.
It was not getting old lines removed, because the tree graft confused
the updater, so it union merged from the previous git-annex branch,
which still contained the old lines. Fixed by carefully using setIndexSha.
This commit was supported by the NSF-funded DataLad project.
This breaks backwards compatibility, but only with unreleased versions of
git-annex, which I think is acceptable.
This commit was supported by the NSF-funded DataLad project.
This way, the temp files that might be left due to failure will be
cleaned up next time.
Also, nub the list of incomplete exports to avoid repeatedly adding the
same tree to it when running export repeatedly when it's failing.
This commit was supported by the NSF-funded DataLad project.
Not yet used, but essential for resuming cleanly.
Note that, in normmal operation, only one commit is made to export.log
during an export; the incomplete version only gets to the journal and
is then overwritten.
This commit was supported by the NSF-funded DataLad project.
So it will be available later and elsewhere, even after GC.
I first though to use git update-index to do this, but feeding it a line
with a tree object seems to always cause it to generate a git subtree
merge. So, fell back to using the Git.Tree interface to maniupulate the
trees, and not involving the git-annex branch index file at all.
This commit was sponsored by Andreas Karlsson.
