canonical-ci-engineering team mailing list archive

[James Henstridge <james.henstridge@xxxxxxxxxxxxx>]

On 14 December 2015 at 09:26, Michi Henning <michi.henning@xxxxxxxxxxxxx> wrote:
>
> On 12 Dec 2015, at 18:17 , James Henstridge <james.henstridge@xxxxxxxxxxxxx>
> wrote:
>
> So, looking at the xvfb-run man page, it sends the X server logs to
> /dev/null by default:
>
>       -e file, --error-file=file
>              Store output from xauth  and  Xvfb  in  file.   The  default
> is
>              /dev/null.
>
> We should adjust the test to do something useful with those logs (e.g.
> print them out if the test fails).  Presumably the cause of the test
> failure will be obvious once we can see that.
>
>
> Did that. Here is what comes out:
>
> 8: Test command:
> /tmp/buildd/thumbnailer-2.3+16.04.20151102.2bzr314pkg0vivid283/tests/qml/run_test.sh
> "/tmp/buildd/thumbnailer-2.3+16.04.20151102.2bzr314pkg0vivid283/obj-arm-linux-gnueabihf/stderr.log"
> "/tmp/buildd/thumbnailer-2.3+16.04.20151102.2bzr314pkg0vivid283/obj-arm-linux-gnueabihf/plugins"
> 8: Test timeout computed to be: 1500
> 8: QXcbConnection: Could not connect to display :109
> 8: Aborted
> 8: _XSERVTransSocketUNIXCreateListener: ...SocketCreateListener() failed
> 8: _XSERVTransMakeAllCOTSServerListeners: server already running
> 8: (EE)
> 8: Fatal server error:
> 8: (EE) Cannot establish any listening sockets - Make sure an X server isn't
> already running(EE)
> 8: _XSERVTransSocketUNIXCreateListener: ...SocketCreateListener() failed
> 8: _XSERVTransMakeAllCOTSServerListeners: server already running
> 8: (EE)

Here's the relevant code from xvfb-run script:

SERVERNUM=99

# Find a free server number by looking at .X*-lock files in /tmp.
find_free_servernum() {
    # Sadly, the "local" keyword is not POSIX.  Leave the next line commented in
    # the hope Debian Policy eventually changes to allow it in /bin/sh scripts
    # anyway.
    #local i

    i=$SERVERNUM
    while [ -f /tmp/.X$i-lock ]; do
        i=$(($i + 1))
    done
    echo $i
}


So to get these results, there must have been /tmp/.X99-lock to
/tmp/.X108-lock files on the system, and the /tmp/.X11-unix/X109
socket either existed or couldn't be created due to permission issues.

The code looks like it could be prone to race conditions if other jobs
were trying to start of Xvfb servers at the same time, but that
wouldn't explain the repeated failures.  It seems more likely that
some previous test run (or multiple runs) has left garbage behind
under /tmp.

James.