test_copy_to fails from SFTP->SFTP sometimes

[Jelmer Vernooij]

On armel, this happens for Debian builds:

======================================================================
ERROR: bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPSiblingAbsoluteServer)
----------------------------------------------------------------------
Traceback (most recent call last):
testtools.testresult.real._StringException: lost connection during test 'bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPSiblingAbsoluteServer)'
======================================================================
ERROR: bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPHomeDirServer)
----------------------------------------------------------------------
Traceback (most recent call last):
testtools.testresult.real._StringException: lost connection during test 'bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPHomeDirServer)'
======================================================================
ERROR: bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPAbsoluteServer)
----------------------------------------------------------------------
Traceback (most recent call last):
testtools.testresult.real._StringException: lost connection during test 'bzrlib.tests.per_transport.TransportTests.test_copy_to(SFTPTransport,SFTPAbsoluteServer)'
----------------------------------------------------------------------

Paramiko 2.0, bzr r6619

[Vincent Ladeuil (vila)]

https://buildd.debian.org/status/fetch.php?pkg=bzr&arch=armel&ver=2.7.0%2Bbzr6619-3&stamp=148025

also mentions:

Ran 11805 tests in 1689.812s

FAILED (errors=3, known_failure_count=21)

and 'lost connection' is subunit reporting that the subprocess died.

So I'd rather suspect some OOM killer ?

[Jelmer Vernooij (jelmer)]

On Thu, Dec 08, 2016 at 09:23:19AM -0000, Vincent Ladeuil wrote:
> https://buildd.debian.org/status/fetch.php?pkg=bzr&arch=armel&ver=2.7.0%2Bbzr6619-3&stamp=148025
>
> also mentions:
>
> Ran 11805 tests in 1689.812s
>
> FAILED (errors=3, known_failure_count=21)
>
> and 'lost connection' is subunit reporting that the subprocess died.
>
> So I'd rather suspect some OOM killer ?
There should be quite a bit of RAM on these machines, they're used to
built much bigger packages than bzr. There's also no indication of
them being OOM-killed anywhere AFAICT.

It seems a little too suspicious that it's this specific test
(test_copy_to), and for SFTP only, that causes each of the 3 subunit workers
to die.

[Martin Packman (gz)]

For now have just skipped the SFTPTransport scenarios for this test.