Joey Hess [Mon, 17 Oct 2022 19:56:19 +0000 (15:56 -0400)]
More robust handling of ErrorBusy when writing to sqlite databases
While ErrorBusy and other exceptions were caught and the write retried for
up to 10 seconds, it was still possible for git-annex to eventually
give up and error out without writing to the database. Now it will retry
as long as necessary.
This does mean that, if one git-annex process is suspended just as sqlite
has locked the database for writing, another git-annex that tries to write
it it might get stuck retrying forever. But, that could already happen when
opening the sqlite database, which retries forever on ErrorBusy. This is an
area where git-annex is known to not behave well, there's a todo about the
general case of it.
Probably not cooincidentally, I emailed sqlite-users about such a
situation in Feb 2015.
https://www.mail-archive.com/sqlite-users@mailinglists.sqlite.org/msg90580.html
Noone ever replied to me, but at least now I understand why it does that.
Since it's documented now, it's no longer a bug.
Joey Hess [Wed, 12 Oct 2022 19:21:19 +0000 (15:21 -0400)]
avoid uncessary keys db writes; doubled speed!
When running eg git-annex get, for each file it has to read from and
write to the keys database. But it's reading exclusively from one table,
and writing to a different table. So, it is not necessary to flush the
write to the database before reading. This avoids writing the database
once per file, instead it will buffer 1000 changes before writing.
Benchmarking getting 1000 small files from a local origin,
git-annex get now takes 13.62s, down from 22.41s!
git-annex drop now takes 9.07s, down from 18.63s!
Wowowowowowowow!
(It would perhaps have been better if there were separate databases for
the two tables. At least it would have avoided this complexity. Ah well,
this is better than splitting the table in a annex.version upgrade.)
Joey Hess [Wed, 12 Oct 2022 17:50:46 +0000 (13:50 -0400)]
avoid flushing keys db queue after each Annex action
The flush was only done Annex.run' to make sure that the queue was flushed
before git-annex exits. But, doing it there means that as soon as one
change gets queued, it gets flushed soon after, which contributes to
excessive writes to the database, slowing git-annex down.
(This does not yet speed git-annex up, but it is a stepping stone to
doing so.)
Database queues do not autoflush when garbage collected, so have to
be flushed explicitly. I don't think it's possible to make them
autoflush (except perhaps if git-annex sqitched to using ResourceT..).
The comment in Database.Keys.closeDb used to be accurate, since the
automatic flushing did mean that all writes reached the database even
when closeDb was not called. But now, closeDb or flushDb needs to be
called before stopping using an Annex state. So, removed that comment.
In Remote.Git, change to using quiesce everywhere that it used to use
stopCoProcesses. This means that uses on onLocal in there are just as
slow as before. I considered only calling closeDb on the local git remotes
when git-annex exits. But, the reason that Remote.Git calls stopCoProcesses
in each onLocal is so as not to leave git processes running that have files
open on the remote repo, when it's on removable media. So, it seemed to make
sense to also closeDb after each one, since sqlite may also keep files
open. Although that has not seemed to cause problems with removable
media so far. It was also just easier to quiesce in each onLocal than
once at the end. This does likely leave performance on the floor, so
could be revisited.
In Annex.Content.saveState, there was no reason to close the db,
flushing it is enough.
The rest of the changes are from auditing for Annex.new, and making
sure that quiesce is called, after any action that might possibly need
it.
After that audit, I'm pretty sure that the change to Annex.run' is
safe. The only concern might be that this does let more changes get
queued for write to the db, and if git-annex is interrupted, those will be
lost. But interrupting git-annex can obviously already prevent it from
writing the most recent change to the db, so it must recover from such
lost data... right?
Joey Hess [Fri, 7 Oct 2022 17:19:17 +0000 (13:19 -0400)]
take lock in checkLogFile and calcLogFile
move: Fix openFile crash with -J
This does make them a bit slower, although usually the log file is not
very big, so even when it's being rewritten, they will not block for
long taking the lock. Still, little slowdowns may add up when moving a lot
file files.
A less expensive fix would be to use something lower level than openFile
that does not check if the file is already open for write by another
thread. But GHC does not seem to provide anything convenient; even mkFD
checks for a writing thread.
fullLines is no longer necessary since these functions no longer will
read the file while it's being written.
Joey Hess [Mon, 3 Oct 2022 17:48:40 +0000 (13:48 -0400)]
avoid combining multiple words provided to trust/untrust/dead
* trust, untrust, semitrust, dead: Fix behavior when provided with
multiple repositories to operate on.
* trust, untrust, semitrust, dead: When provided with no parameters,
do not operate on a repository that has an empty name.
The man page and usage already indicated that multiple repos could be
provided to these commands, but they actually used unwords to combine
everything into string, and found a repo matching that string. This was
especially bad when no parameters resulted in the empty string and some
repo happened to have an empty description.
This does change the behavior, and it's possible someone relied on the
current behavior to eg, trust a repo by name with the name not quoted into
a single parameter. But fixing the empty string bug and matching the
documentation are worth breaking that usage.
Note that git-annex init/reinit do still unwords multiple parameters when
provided to them. That is inconsistent behavior, but it certianly seems
possible that something does run git-annex init with an unquoted
description, and I don't think it's worth breaking that just to make it more
consistent with these other commands.
Support "inbackend" in preferred content expressions
Well, actually, fix a typo that has always been in the implementation of
that. "inbacked" used to work, but let's not tell users about that; they
might try to use it and expect git-annex to keep supporting the typo..
I noticed that, using just the man pages, there is no real description
of what backends are, or what ones are available. Except for some
examples.
Added a git-annex-backends man page, that is just a stub, but at least
describes what they basically are, and tells how to find the supported
ons, and links to the backends web page.
Commit ff0927bde98e7b7fd4a7ef49b61b8770c6df294d broke this, it made it
try to read all of the input before looking for the code. But, wormhole
keeps running until it sends the file, so that caused a deadlock. Oops.
p2p: Pass wormhole the --appid option before the receive/send command, as
it does not accept that option after the command
I'm left wondering, did I get this wrong from the beginning, or did
wormhole change its option parser? I'm reminded of the change in 0.8.2
where it silently changed what FD the pairing code was output to.
But, looking at the wormhole source, it was at least putting --appid before
send in its test suite from the introduction of the option.
So I think probably this has always been broken. On 2021-12-31 the --appid
option was enabled, and it took until now for someone to try
git-annex p2p --pair and notice that flag day broke it..
The user sets these hooks deliberately so they should always be run. For
example this allows hooks to be used to manage file permissions on NTFS
volumes in WSL1.
Avoided the bug that had by making the upgrade log be updated after each
upgrade step. So, after upgrade from v8 to v9, the log is updated, and
so Upgrade.V9's timeOfUpgrade check will find that it was upgraded
recently and so won't let it skip ahead to v10.
projects / git-annex.git / log