Just for your information. I've just sent the mail below to the VirtualBox user mailing list. I've done all development and testing of the patches for multiple CD support in VirtualBox, basically without any problems other than the issues listed below. Cheers, FJP ---------- Forwarded Message ---------- Subject: Issues using VirtualBox for development of the Debian Installer Date: Monday 29 October 2007 Hi, I've been using VirtualBox for some time now for testing and development of the installation system for Debian GNU/Linux. In general I must say I'm very impressed, especially with the speed of the virtual machines and with the general stability. However, I do have a few issues that I hope can be resolved in future versions. If you'd like me to report any of the issues as bugs or feature requests, please let me know. System details: - Current VirtualBox Debian packages (1.5.2_OSE) - Intel Pentium D (dual core) host system running 2.6.23 kernel - Installer in VM runs 2.6.22 Debian kernel 1) Should allow the use of ctrl-alt-Fx for console switching The installer has a text version and a graphical version. Both have different virtual consoles active: one main console, two debug shells and a console showing the syslog. For the graphical installer the _only_ way to switch from the main console to any of the others is to use the ctrl-alt-<function key> key combinations. As these are currently not captured inside the VM, VBox is not usable for testing the graphical version of the installer. I have seen other request to support this, so please add our vote for that! Note that both vmware and qemu do support this (qemu through its management console, which is not ideal). IMHO _all_ keys, except for the Host key, should be captured when input is being captured (i.e. capturing the mouse should be extended to also capturing the keyboard), just like vmware does. This really is an essential feature for us as being able to switch to the consoles with debug shells is essential for debugging and development. 2) NAT networking connection lost after making new snapshot NAT networking works perfectly until I make a snapshot. Immediately after that I get "temporary failure in name resolution" errors, which means that the system can no longer contact the DNS server. Attempts to connect using IP address fail with "no route to host". Retrying does not help. If I revert to that same snapshot and restart the VM, the _first_ attempt to e.g. scp a file will mostly fail. However, a second attempt will succeed. After that, networking is completely reliable again. This issue is 100% reproducible. 3) Mounting CD inside VM fails first time after changing image I boot the installer using with the VM's CD pointing to an ISO CD image (CD1). During the installation I need to switch to a different image (CD2). Even if I do that procedure completely manually, the first mount command will always fail. # umount /cdrom /* use virtual disk manager to change CD ISO image */ # mount /dev/hdc /cdrom mount: you must specify the filesystem type # mount /dev/hdc /cdrom mount: block device /dev/hdc is write-protected, mounting read-only /* use virtual disk manager to change CD ISO image */ # mount /cdrom mount: no medium found # mount /cdrom mount: block device /dev/hdc is write-protected, mounting read-only # umount /cdrom There is one strange thing: I can only reproduce this when using the Debian kernel compiled for a 486 processor, not with the one compiled for a 686 processor. I can provide the kernel configs if needed. As this problem does not occur in vmware, I think it must be an issue in VirtualBox. 4) snapshots It would be really great if you could not only go back to the "last" snapshot, but also to previous snapshots. In vmware, I used to create a "clean machine" snapshot that I would use as basis for any installation tests and I could create multiple different test situations beside eachother. I miss having that option in VirtualBox. 5) "Select boot device" timeout The time you can pressing the F12 key during boot to select a different boot device is fairly short. It would be nice if that could be configured to take use a longer timeout. Cheers, Frans Pop Debian Developer P.S. Two issues I had with 1.5.0 seem to be solved in 1.5.2: - frequent segfaults when (re)starting a VM - segfault in hwclock with Linux kernels >2.6.21 Thanks! -------------------------------------------------------
Attachment:
signature.asc
Description: This is a digitally signed message part.