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
$ cat > correct.py <<NO_CHECK_EOF
> def toto(arg1, arg2):
> del arg2
> return (5 + 6, 9)
> NO_CHECK_EOF
$ cat > wrong.py <<NO_CHECK_EOF
> def toto( arg1, arg2):
> del(arg2)
> return ( 5+6, 9)
> def badwrap():
> return 1 + \\
> 2
> NO_CHECK_EOF
$ cat > quote.py <<NO_CHECK_EOF
> # let's use quote in comments
> (''' ( 4x5 )
> but """\\''' and finally''',
> """let's fool checkpatch""", '1+2',
> '"""', 42+1, """and
> ( 4-1 ) """, "( 1+1 )\" and ")
> a, '\\\\\\\\', "\\\\\\" x-2", "c-1"
> NO_CHECK_EOF
$ cat > classstyle.py <<NO_CHECK_EOF
> class newstyle_class(object):
> pass
>
> class oldstyle_class:
> pass
>
> class empty():
> pass
>
> no_class = 1:
> pass
> NO_CHECK_EOF
$ check_code="$TESTDIR"/../contrib/check-code.py
$ "$check_code" ./wrong.py ./correct.py ./quote.py ./classstyle.py
./wrong.py:1:
> def toto( arg1, arg2):
gratuitous whitespace in () or []
./wrong.py:2:
> del(arg2)
Python keyword is not a function
./wrong.py:3:
> return ( 5+6, 9)
gratuitous whitespace in () or []
missing whitespace in expression
./wrong.py:5:
> return 1 + \
Use () to wrap long lines in Python, not \
./quote.py:5:
> '"""', 42+1, """and
missing whitespace in expression
./classstyle.py:4:
> class oldstyle_class:
old-style class, use class foo(object)
./classstyle.py:7:
> class empty():
class foo() creates old style object, use class foo(object)
[1]
$ cat > python3-compat.py << NO_CHECK_EOF
> foo <> bar
> reduce(lambda a, b: a + b, [1, 2, 3, 4])
> dict(key=value)
> NO_CHECK_EOF
$ "$check_code" python3-compat.py
python3-compat.py:1:
> foo <> bar
<> operator is not available in Python 3+, use !=
python3-compat.py:2:
> reduce(lambda a, b: a + b, [1, 2, 3, 4])
reduce is not available in Python 3+
python3-compat.py:3:
> dict(key=value)
dict() is different in Py2 and 3 and is slower than {}
[1]
$ cat > foo.c <<NO_CHECK_EOF
> void narf() {
> strcpy(foo, bar);
> // strcpy_s is okay, but this comment is not
> strcpy_s(foo, bar);
> }
> NO_CHECK_EOF
$ "$check_code" ./foo.c
./foo.c:2:
> strcpy(foo, bar);
don't use strcpy, use strlcpy or memcpy
./foo.c:3:
> // strcpy_s is okay, but this comment is not
don't use //-style comments
[1]
$ cat > is-op.py <<NO_CHECK_EOF
> # is-operator comparing number or string literal
> x = None
> y = x is 'foo'
> y = x is "foo"
> y = x is 5346
> y = x is -6
> y = x is not 'foo'
> y = x is not "foo"
> y = x is not 5346
> y = x is not -6
> NO_CHECK_EOF
$ "$check_code" ./is-op.py
./is-op.py:3:
> y = x is 'foo'
object comparison with literal
./is-op.py:4:
> y = x is "foo"
object comparison with literal
./is-op.py:5:
> y = x is 5346
object comparison with literal
./is-op.py:6:
> y = x is -6
object comparison with literal
./is-op.py:7:
> y = x is not 'foo'
object comparison with literal
./is-op.py:8:
> y = x is not "foo"
object comparison with literal
./is-op.py:9:
> y = x is not 5346
object comparison with literal
./is-op.py:10:
> y = x is not -6
object comparison with literal
[1]
$ cat > for-nolineno.py <<NO_CHECK_EOF
> except:
> NO_CHECK_EOF
$ "$check_code" for-nolineno.py --nolineno
for-nolineno.py:0:
> except:
naked except clause
[1]
$ cat > warning.t <<NO_CHECK_EOF
> $ function warnonly {
> > }
> $ diff -N aaa
> $ function onwarn {}
> NO_CHECK_EOF
$ "$check_code" warning.t
$ "$check_code" --warn warning.t
warning.t:1:
> $ function warnonly {
warning: don't use 'function', use old style
warning.t:3:
> $ diff -N aaa
warning: don't use 'diff -N'
warning.t:4:
> $ function onwarn {}
warning: don't use 'function', use old style
[1]
$ cat > error.t <<NO_CHECK_EOF
> $ [ foo == bar ]
> NO_CHECK_EOF
$ "$check_code" error.t
error.t:1:
> $ [ foo == bar ]
[ foo == bar ] is a bashism, use [ foo = bar ] instead
[1]
$ rm error.t
$ cat > raise-format.py <<NO_CHECK_EOF
> raise SomeException, message
> # this next line is okay
> raise SomeException(arg1, arg2)
> NO_CHECK_EOF
$ "$check_code" not-existing.py raise-format.py
Skipping*not-existing.py* (glob)
raise-format.py:1:
> raise SomeException, message
don't use old-style two-argument raise, use Exception(message)
[1]
$ cat <<NO_CHECK_EOF > tab.t
> indent
> > heredoc
> NO_CHECK_EOF
$ "$check_code" tab.t
tab.t:1:
> indent
don't use tabs to indent
[1]
$ rm tab.t
$ cat > rst.py <<NO_CHECK_EOF
> """problematic rst text
>
> .. note::
> wrong
> """
>
> '''
>
> .. note::
>
> valid
>
> new text
>
> .. note::
>
> also valid
> '''
>
> """mixed
>
> .. note::
>
> good
>
> .. note::
> plus bad
> """
> NO_CHECK_EOF
$ $check_code -w rst.py
rst.py:3:
> .. note::
warning: add two newlines after '.. note::'
rst.py:26:
> .. note::
warning: add two newlines after '.. note::'
[1]
$ cat > ./map-inside-gettext.py <<NO_CHECK_EOF
> print(_("map inside gettext %s" % v))
>
> print(_("concatenating " " by " " space %s" % v))
> print(_("concatenating " + " by " + " '+' %s" % v))
>
> print(_("mapping operation in different line %s"
> % v))
>
> print(_(
> "leading spaces inside of '(' %s" % v))
> NO_CHECK_EOF
$ "$check_code" ./map-inside-gettext.py
./map-inside-gettext.py:1:
> print(_("map inside gettext %s" % v))
don't use % inside _()
./map-inside-gettext.py:3:
> print(_("concatenating " " by " " space %s" % v))
don't use % inside _()
./map-inside-gettext.py:4:
> print(_("concatenating " + " by " + " '+' %s" % v))
don't use % inside _()
./map-inside-gettext.py:6:
> print(_("mapping operation in different line %s"
don't use % inside _()
./map-inside-gettext.py:9:
> print(_(
don't use % inside _()
[1]
web templates
$ mkdir -p mercurial/templates
$ cat > mercurial/templates/example.tmpl <<NO_CHECK_EOF
> {desc}
> {desc|escape}
> {desc|firstline}
> {desc|websub}
> NO_CHECK_EOF
$ "$check_code" --warnings mercurial/templates/example.tmpl
mercurial/templates/example.tmpl:2:
> {desc|escape}
warning: follow desc keyword with either firstline or websub
[1]
'string join across lines with no space' detection
$ cat > stringjoin.py <<NO_CHECK_EOF
> foo = (' foo'
> 'bar foo.'
> 'bar foo:'
> 'bar foo@'
> 'bar foo%'
> 'bar foo*'
> 'bar foo+'
> 'bar foo-'
> 'bar')
> NO_CHECK_EOF
'missing _() in ui message' detection
$ cat > uigettext.py <<NO_CHECK_EOF
> ui.status("% 10s %05d % -3.2f %*s %%"
> # this use '\\\\' instead of '\\', because the latter in
> # heredoc on shell becomes just '\'
> '\\\\ \n \t \0'
> """12345
> """
> '''.:*+-=
> ''' "%-6d \n 123456 .:*+-= foobar")
> NO_CHECK_EOF
superfluous pass
$ cat > superfluous_pass.py <<NO_CHECK_EOF
> # correct examples
> if foo:
> pass
> else:
> # comment-only line means still need pass
> pass
> def nothing():
> pass
> class empty(object):
> pass
> if whatever:
> passvalue(value)
> # bad examples
> if foo:
> "foo"
> pass
> else: # trailing comment doesn't fool checker
> wat()
> pass
> def nothing():
> "docstring means no pass"
> pass
> class empty(object):
> """multiline
> docstring also
> means no pass"""
> pass
> NO_CHECK_EOF
(Checking multiple invalid files at once examines whether caching
translation table for repquote() works as expected or not. All files
should break rules depending on result of repquote(), in this case)
$ "$check_code" stringjoin.py uigettext.py superfluous_pass.py
stringjoin.py:1:
> foo = (' foo'
string join across lines with no space
stringjoin.py:2:
> 'bar foo.'
string join across lines with no space
stringjoin.py:3:
> 'bar foo:'
string join across lines with no space
stringjoin.py:4:
> 'bar foo@'
string join across lines with no space
stringjoin.py:5:
> 'bar foo%'
string join across lines with no space
stringjoin.py:6:
> 'bar foo*'
string join across lines with no space
stringjoin.py:7:
> 'bar foo+'
string join across lines with no space
stringjoin.py:8:
> 'bar foo-'
string join across lines with no space
uigettext.py:1:
> ui.status("% 10s %05d % -3.2f %*s %%"
missing _() in ui message (use () to hide false-positives)
superfluous_pass.py:14:
> if foo:
omit superfluous pass
superfluous_pass.py:17:
> else: # trailing comment doesn't fool checker
omit superfluous pass
superfluous_pass.py:20:
> def nothing():
omit superfluous pass
superfluous_pass.py:23:
> class empty(object):
omit superfluous pass
[1]
Check code fragments embedded in test script
$ cat > embedded-code.t <<NO_CHECK_EOF
> code fragment in doctest style
> >>> x = (1,2)
> ...
> ... x = (1,2)
>
> code fragment in heredoc style
> $ python <<EOF
> > x = (1,2)
> > EOF
>
> code fragment in file heredoc style
> $ python > file.py <<EOF
> > x = (1,2)
> > EOF
> NO_CHECK_EOF
$ "$check_code" embedded-code.t
embedded-code.t:2:
> x = (1,2)
missing whitespace after ,
embedded-code.t:4:
> x = (1,2)
missing whitespace after ,
embedded-code.t:8:
> x = (1,2)
missing whitespace after ,
embedded-code.t:13:
> x = (1,2)
missing whitespace after ,
[1]
"max warnings per file" is shared by all embedded code fragments
$ "$check_code" --per-file=3 embedded-code.t
embedded-code.t:2:
> x = (1,2)
missing whitespace after ,
embedded-code.t:4:
> x = (1,2)
missing whitespace after ,
embedded-code.t:8:
> x = (1,2)
missing whitespace after ,
(too many errors, giving up)
[1]