lfs: fix the stall and corruption issue when concurrently uploading blobs
We've avoided the issue up to this point by gating worker usage with an
experimental config. See 10e62d5efa73, and the thread linked there for some of
the initial diagnosis, but essentially some data was being read from the blob
before an error occurred and `keepalive` retried, but didn't rewind the file
pointer. So the leading data was lost from the blob on the server, and the
connection stalled, trying to send more data than available.
In trying to recreate this, I was unable to do so uploading from Windows to
CentOS 7. But it reproduced every time going from CentOS 7 to another CentOS 7
over https.
I found recent fixes in the FaceBook repo to address this[1][2]. The commit
message for the first is:
The KeepAlive HTTP implementation is bugged in it's retry logic, it supports
reading from a file pointer, but doesn't support rewinding of the seek cursor
when it performs a retry. So it can happen that an upload fails for whatever
reason and will then 'hang' on the retry event.
The sequence of events that get triggered are:
- Upload file A, goes OK. Keep-Alive caches connection.
- Upload file B, fails due to (for example) failing Keep-Alive, but LFS file
pointer has been consumed for the upload and fd has been closed.
- Retry for file B starts, sets the Content-Length properly to the expected
file size, but since file pointer has been consumed no data will be uploaded,
causing the server to wait for the uploaded data until either client or
server reaches a timeout, making it seem as our mercurial process hangs.
This is just a stop-gap measure to prevent this behavior from blocking Mercurial
(LFS has retry logic). A proper solutions need to be build on top of this
stop-gap measure: for upload from file pointers, we should support fseek() on
the interface. Since we expect to consume the whole file always anyways, this
should be safe. This way we can seek back to the beginning on a retry.
I ported those two patches, and it works. But I see that `url._sendfile()` does
a rewind on `httpsendfile` objects[3], so maybe it's better to keep this all in
one place and avoid a second seek. We may still want the first FaceBook patch
as extra protection for this problem in general. The other two uses of
`httpsendfile` are in the wire protocol to upload bundles, and to upload
largefiles. Neither of these appear to use a worker, and I'm not sure why
workers seem to trigger this, or if this could have happened without a worker.
Since `httpsendfile` already has a `close()` method, that is dropped. That
class also explicitly says there's no `__len__` attribute, so that is removed
too. The override for `read()` is necessary to avoid the progressbar usage per
file.
[1] https://github.com/facebookexperimental/eden/commit/c350d6536d90c044c837abdd3675185644481469
[2] https://github.com/facebookexperimental/eden/commit/77f0d3fd0415e81b63e317e457af9c55c46103ee
[3] https://www.mercurial-scm.org/repo/hg/file/5.2.2/mercurial/url.py#l176
Differential Revision: https://phab.mercurial-scm.org/D7962
#require unix-permissions
test that new files created in .hg inherit the permissions from .hg/store
$ mkdir dir
just in case somebody has a strange $TMPDIR
$ chmod g-s dir
$ cd dir
$ cat >printmodes.py <<EOF
> from __future__ import absolute_import, print_function
> import os
> import sys
>
> allnames = []
> isdir = {}
> for root, dirs, files in os.walk(sys.argv[1]):
> for d in dirs:
> name = os.path.join(root, d)
> isdir[name] = 1
> allnames.append(name)
> for f in files:
> name = os.path.join(root, f)
> allnames.append(name)
> allnames.sort()
> for name in allnames:
> suffix = name in isdir and '/' or ''
> print('%05o %s%s' % (os.lstat(name).st_mode & 0o7777, name, suffix))
> EOF
$ cat >mode.py <<EOF
> from __future__ import absolute_import, print_function
> import os
> import sys
> print('%05o' % os.lstat(sys.argv[1]).st_mode)
> EOF
$ umask 077
$ hg init repo
$ cd repo
$ chmod 0770 .hg/store .hg/cache .hg/wcache
before commit
store can be written by the group, other files cannot
store is setgid
$ "$PYTHON" ../printmodes.py .
00700 ./.hg/
00600 ./.hg/00changelog.i
00770 ./.hg/cache/
00600 ./.hg/requires
00770 ./.hg/store/
00770 ./.hg/wcache/
$ mkdir dir
$ touch foo dir/bar
$ hg ci -qAm 'add files'
after commit
working dir files can only be written by the owner
files created in .hg can be written by the group
(in particular, store/**, dirstate, branch cache file, undo files)
new directories are setgid
$ "$PYTHON" ../printmodes.py .
00700 ./.hg/
00600 ./.hg/00changelog.i
00770 ./.hg/cache/
00660 ./.hg/cache/branch2-served
00660 ./.hg/cache/rbc-names-v1
00660 ./.hg/cache/rbc-revs-v1
00660 ./.hg/dirstate
00660 ./.hg/fsmonitor.state (fsmonitor !)
00660 ./.hg/last-message.txt
00600 ./.hg/requires
00770 ./.hg/store/
00660 ./.hg/store/00changelog.i
00660 ./.hg/store/00manifest.i
00770 ./.hg/store/data/
00770 ./.hg/store/data/dir/
00660 ./.hg/store/data/dir/bar.i (reporevlogstore !)
00660 ./.hg/store/data/foo.i (reporevlogstore !)
00770 ./.hg/store/data/dir/bar/ (reposimplestore !)
00660 ./.hg/store/data/dir/bar/b80de5d138758541c5f05265ad144ab9fa86d1db (reposimplestore !)
00660 ./.hg/store/data/dir/bar/index (reposimplestore !)
00770 ./.hg/store/data/foo/ (reposimplestore !)
00660 ./.hg/store/data/foo/b80de5d138758541c5f05265ad144ab9fa86d1db (reposimplestore !)
00660 ./.hg/store/data/foo/index (reposimplestore !)
00660 ./.hg/store/fncache (repofncache !)
00660 ./.hg/store/phaseroots
00660 ./.hg/store/undo
00660 ./.hg/store/undo.backupfiles
00660 ./.hg/store/undo.phaseroots
00660 ./.hg/undo.backup.dirstate
00660 ./.hg/undo.bookmarks
00660 ./.hg/undo.branch
00660 ./.hg/undo.desc
00660 ./.hg/undo.dirstate
00770 ./.hg/wcache/
00711 ./.hg/wcache/checkisexec
007.. ./.hg/wcache/checklink (re)
00600 ./.hg/wcache/checklink-target
00660 ./.hg/wcache/manifestfulltextcache (reporevlogstore !)
00700 ./dir/
00600 ./dir/bar
00600 ./foo
$ umask 007
$ hg init ../push
before push
group can write everything
$ "$PYTHON" ../printmodes.py ../push
00770 ../push/.hg/
00660 ../push/.hg/00changelog.i
00770 ../push/.hg/cache/
00660 ../push/.hg/requires
00770 ../push/.hg/store/
00770 ../push/.hg/wcache/
$ umask 077
$ hg -q push ../push
after push
group can still write everything
$ "$PYTHON" ../printmodes.py ../push
00770 ../push/.hg/
00660 ../push/.hg/00changelog.i
00770 ../push/.hg/cache/
00660 ../push/.hg/cache/branch2-base
00660 ../push/.hg/dirstate
00660 ../push/.hg/requires
00770 ../push/.hg/store/
00660 ../push/.hg/store/00changelog.i
00660 ../push/.hg/store/00manifest.i
00770 ../push/.hg/store/data/
00770 ../push/.hg/store/data/dir/
00660 ../push/.hg/store/data/dir/bar.i (reporevlogstore !)
00660 ../push/.hg/store/data/foo.i (reporevlogstore !)
00770 ../push/.hg/store/data/dir/bar/ (reposimplestore !)
00660 ../push/.hg/store/data/dir/bar/b80de5d138758541c5f05265ad144ab9fa86d1db (reposimplestore !)
00660 ../push/.hg/store/data/dir/bar/index (reposimplestore !)
00770 ../push/.hg/store/data/foo/ (reposimplestore !)
00660 ../push/.hg/store/data/foo/b80de5d138758541c5f05265ad144ab9fa86d1db (reposimplestore !)
00660 ../push/.hg/store/data/foo/index (reposimplestore !)
00660 ../push/.hg/store/fncache (repofncache !)
00660 ../push/.hg/store/undo
00660 ../push/.hg/store/undo.backupfiles
00660 ../push/.hg/store/undo.phaseroots
00660 ../push/.hg/undo.bookmarks
00660 ../push/.hg/undo.branch
00660 ../push/.hg/undo.desc
00660 ../push/.hg/undo.dirstate
00770 ../push/.hg/wcache/
Test that we don't lose the setgid bit when we call chmod.
Not all systems support setgid directories (e.g. HFS+), so
just check that directories have the same mode.
$ cd ..
$ hg init setgid
$ cd setgid
$ chmod g+rwx .hg/store
$ chmod g+s .hg/store 2> /dev/null || true
$ mkdir dir
$ touch dir/file
$ hg ci -qAm 'add dir/file'
$ storemode=`"$PYTHON" ../mode.py .hg/store`
$ dirmode=`"$PYTHON" ../mode.py .hg/store/data/dir`
$ if [ "$storemode" != "$dirmode" ]; then
> echo "$storemode != $dirmode"
> fi
$ cd ..
$ cd .. # g-s dir