Based on UTAH code and on the assumptions made due to the debugging I created a test script that tries to trigger the cleanup being concurrent to the undefine calls.
I aran 600 loops - so far nothing triggered, so we miss some aspect of the original issue.
Maybe the reason/way qemu dies is different?
Maybe we need a case where libvirt writes to qemu on some of those streams?
Not sure yet, for now I enable a debugging log of libvirt on s1lp4 to maybe get better insight this happens next time.
@paride - please ping me when it happens next time so we can check the logs
@paride - if anything comes to your mind how this guest abort could be different let me know
Based on UTAH code and on the assumptions made due to the debugging I created a test script that tries to trigger the cleanup being concurrent to the undefine calls.
I aran 600 loops - so far nothing triggered, so we miss some aspect of the original issue.
Maybe the reason/way qemu dies is different?
Maybe we need a case where libvirt writes to qemu on some of those streams?
Not sure yet, for now I enable a debugging log of libvirt on s1lp4 to maybe get better insight this happens next time.
@paride - please ping me when it happens next time so we can check the logs
@paride - if anything comes to your mind how this guest abort could be different let me know