Help trying to run selftest on EC2

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


...

> Anyone have any ideas about how to clear this up? Should I hack up an
> os.rename() that detects PermissionDenied, calls time.sleep() for a half
> second, and then tries again?
> 
> John
> =:->

So I hacked together this:

=== modified file 'bzrlib/tests/__init__.py'
- --- bzrlib/tests/__init__.py    2009-11-30 20:33:06 +0000
+++ bzrlib/tests/__init__.py    2009-12-02 17:31:45 +0000
@@ -4447,3 +4447,25 @@
             return result
 except ImportError:
     pass
+
+
+_orig_rename = os.rename
+def os_rename_hack(*args, **kwargs):
+    try:
+        return _orig_rename(*args, **kwargs)
+    except OSError, e:
+        if e.errno in (errno.EPERM, errno.EACCES):
+            try:
+                st = os.lstat(args[0])
+            except (OSError, IOError), e:
+                note('lstat failure %s for %s' % (errno.errorcode[e.errno],
+                    args[0]))
+                st = None
+            note('%s while trying to rename %s %r, trying again.'
+                   % (errno.errorcode[e.errno], st, args))
+            time.sleep(1.0)
+            return _orig_rename(*args, **kwargs)
+            # We don't re-raise, as we assume that
+        else:
+            raise
+os.rename = os_rename_hack

But it still fails (even with a 1.0s retry).

Does anyone know if there is a way to list open file handles for a
process? I don't really understand why renaming 'held' is failing. Since
we shouldn't have opened the info file after we renamed it from
'pending.*'. At least, I believe we don't peek until after we have
renamed it to 'releasing.*'.


Ah... I may have an idea. I think after we rename pending => held we
then peek at the info file to make sure we got the lock.
And looking at the read_info_file code, we do:

        return self._parse_info(self.transport.get(path))

And *that* means we grab a file, which we then call '.readlines()', but
never call .close(). Which means that the file gets closed by the
deconstructor, which means it is being closed asynchronously. Which
*might* not happen before we go to rename the file. I'll try a patch for
that.

It doesn't explain why renaming the pack files is failing, as that looks
pretty explicit about calling "self.write_stream.close()" which I'm
pretty sure calls all the way down to doing a blocking close. But maybe
I can at least fix this problem.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAksWpsAACgkQJdeBCYSNAAPTLACfdiWQXabF6tQX8lr0YiVgdWF2
F2MAn3YCNb1QeScDoffNBMPWxKjjFYwN
=cdxx
-----END PGP SIGNATURE-----