This is an archive of the discontinued Mercurial Phabricator instance.

Differential D7962

lfs: fix the stall and corruption issue when concurrently uploading blobs
ClosedPublic

Authored by mharbison72 on Jan 21 2020, 1:37 PM.

Details

Reviewers
marmoute
Group Reviewers
hg-reviewers
Commits
rHG43eea17ae7b3: lfs: fix the stall and corruption issue when concurrently uploading blobs
Summary

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

Diff Detail

Repository
rHG Mercurial
Lint
Lint Skipped
Unit
Unit Tests Skipped

Event Timeline

mharbison72 created this revision.Jan 21 2020, 1:37 PM
mharbison72 added a comment.Jan 21 2020, 1:43 PM

Even though this talks about fixing corruption, I don't *think* we need this in the next release. We've lived with this problem for 2 years now. But somebody who really understands how all of the URL request handling and keepalive stuff works should probably take a look before then. I still don't know that this *can't* happen without using workers (or if it is possible in other usage contexts), so we may still want to shore things up there.

marmoute accepted this revision.Feb 5 2020, 6:19 PM
marmoute added a subscriber: marmoute.

The patch is a bit obscur, but the explanation is strong.

mharbison72 added a commit: rHG43eea17ae7b3: lfs: fix the stall and corruption issue when concurrently uploading blobs.Feb 6 2020, 5:20 PM
This revision was not accepted when it landed; it landed in state Needs Review.
Closed by commit rHG43eea17ae7b3: lfs: fix the stall and corruption issue when concurrently uploading blobs (authored by mharbison72). · Explain Why
This revision was automatically updated to reflect the committed changes.

Revision Contents
Changeset List

PathPackages
Mhgext/lfs/blobstore.py (31 lines)

Diff 19496

hgext/lfs/blobstore.py

import socket import socket
from mercurial.i18n import _ from mercurial.i18n import _
from mercurial.pycompat import getattr from mercurial.pycompat import getattr
from mercurial import ( from mercurial import (
encoding, encoding,
error, error,
httpconnection as httpconnectionmod,
node, node,
pathutil, pathutil,
pycompat, pycompat,
url as urlmod, url as urlmod,
util, util,
vfs as vfsmod, vfs as vfsmod,
worker, worker,
) )
def walk(self, path=None, onerror=None): def walk(self, path=None, onerror=None):
return (b'', [], []) return (b'', [], [])
def write(self, oid, data): def write(self, oid, data):
pass pass
class lfsuploadfile(object): class lfsuploadfile(httpconnectionmod.httpsendfile):
"""a file-like object that supports __len__ and read. """a file-like object that supports keepalive.
""" """
def __init__(self, fp): def __init__(self, ui, filename):
self._fp = fp super(lfsuploadfile, self).__init__(ui, filename, b'rb')
fp.seek(0, os.SEEK_END) self.read = self._data.read
self._len = fp.tell()
fp.seek(0) def _makeprogress(self):
return None # progress is handled by the worker client
def __len__(self):
return self._len
def read(self, size):
if self._fp is None:
return b''
return self._fp.read(size)
def close(self):
if self._fp is not None:
self._fp.close()
self._fp = None
class local(object): class local(object):
"""Local blobstore for large file contents. """Local blobstore for large file contents.
This blobstore is used both as a cache and as a staging area for large blobs This blobstore is used both as a cache and as a staging area for large blobs
to be uploaded to the remote blobstore. to be uploaded to the remote blobstore.
""" """
hint=_(b'run hg verify'), hint=_(b'run hg verify'),
) )
for k, v in headers: for k, v in headers:
request.add_header(pycompat.strurl(k), pycompat.strurl(v)) request.add_header(pycompat.strurl(k), pycompat.strurl(v))
try: try:
if action == b'upload': if action == b'upload':
request.data = lfsuploadfile(localstore.open(oid)) request.data = lfsuploadfile(self.ui, localstore.path(oid))
request.get_method = lambda: 'PUT' request.get_method = lambda: 'PUT'
request.add_header('Content-Type', 'application/octet-stream') request.add_header('Content-Type', 'application/octet-stream')
request.add_header('Content-Length', len(request.data)) request.add_header('Content-Length', request.data.length)
with contextlib.closing(self.urlopener.open(request)) as res: with contextlib.closing(self.urlopener.open(request)) as res:
contentlength = res.info().get(b"content-length") contentlength = res.info().get(b"content-length")
ui = self.ui # Shorten debug lines ui = self.ui # Shorten debug lines
if self.ui.debugflag: if self.ui.debugflag:
ui.debug(b'Status: %d\n' % res.status) ui.debug(b'Status: %d\n' % res.status)
# lfs-test-server and hg serve return headers in different # lfs-test-server and hg serve return headers in different
# order # order