Bug 5207

Summary:	SSH test fails with nightly-qa-systemd
Product:	[Build System, Metadata & Runtime] OE-Core	Reporter:	Paul Eggleton <bluelightning>
Component:	core	Assignee:	Stefan Stanacar <sstncr>
Status:	VERIFIED FIXED	QA Contact:
Severity:	normal
Priority:	Medium+	CC:	alexandrux.palalau, meta.mr.watcher, meta.watcher, richard.purdie, sstncr
Version:	1.5
Target Milestone:	1.5 M5
Hardware:	x86
OS:	Multiple
Whiteboard:
OS type for building Yocto:	---	Type of Regression:	---
Verified:		Documentation change:	No (bug/feature does not impact docs)

Description Paul Eggleton 2013-09-16 18:12:58 UTC

In the nightly-qa-systemd build on the autobuilder, we build core-image-sato and I can see that dropbear is being installed by looking at log.do_rootfs and according to the QEMU log it appears to be started, however the automated tests almost all fail because the SSH test fails.

http://autobuilder.yoctoproject.org:8011/builders/nightly-qa-systemd/builds/167/steps/Running%20Sanity%20Tests/logs/stdio

Booting this image locally, this looks like it has something to do with the ethernet interface not being started correctly/early enough.

Comment 1 Ross Burton 2013-09-25 11:07:29 UTC

The situation is that under systemd you reach the login prompt about the same time that connman has just completed starting (that is, its running but hasn't probed the hardware yet).  The ping test then starts and succeeds because runqemu configures the kernel to bring up eth0 but then some time later connman takes over eth0, which means taking down the interface and bringing it back up again.

This behaviour in connman is currently intentional, although I've mailed Marcel to see if they'd be willing to compare the current configuration with the requested configuration and not take the interface down if they match.

No amount of systemd dependency-fu can resolve this because connman is "ready" once it has loaded, actually working on the interfaces happens afterwards.

One way to mitigate this would be to require N pings from the VM to ensure that the connection has had time to settle.

Comment 2 Ross Burton 2013-09-25 14:54:07 UTC

Assigning to Stefan who agreed with me and has submitted a patch to do what I suggested.

Comment 3 Stefan Stanacar 2013-09-27 07:58:55 UTC

Patch merged in master http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=41f88ee1dbda44605301dd7309926e04d146ef52

Since we used the nice levels trick on the AB we haven't seen this so I'm not sure if it will fix the problem forever (as the described by Ross the behaviour of connman is intentional), but it should help with false SSH fails somewhat.

Comment 4 Alexandru Palalau 2013-10-16 08:11:37 UTC

Switching to verified.
Master branch.
Commit ID: 2b5bac02196b8bd30d223d508f5f7d4300889fd7