mato / docker-unikernel-runner Goto Github PK

Is this unikernel runner built specifically for MirageOS or can it be modified to run OSv unikernels?

Thanks!

Both Mirage/unix and Mirage/Solo5 on ukvm currently use a freshly-generated random (locally administered, unicast) MAC address for each run of the unikernel. If several unikernels are started and stopped in series, the Docker-allocated IP address is re-used for each run, but with a different MAC address. If this happens before the host times out its ARP entry for thid IP address then connectivity will be delayed until the ARP entry is updated, potentially causing TCP connections to the unikernel to time out.

Possible solution: Save the container-side eth0 MAC address, replace it with a randomly-generated one and pass the saved address to the unikernel. May not be 100% reliable due to the "change of address" resulting in DAD or other undesirable communication from the container-side Linux network stack.

This is not a real issue but more like a design question. Why do you use a C binary to manage the networking? Last night I was determined to get the tap interface running in the vanilla alpine container without any C binary, just with scripting, and I got it running with this script (no packages required):

IP=$(ip -o -f inet address show eth0 | awk '{print $4}')
ip address delete $IP dev eth0
tunctl -t tap0
brctl addbr br0
brctl addif br0 eth0
brctl addif br0 tap0
ifconfig br0 up
ifconfig tap0 up

/my-unikernel --interface tap0 --ipv4 $IP

Of course you also have to start the container with --device=/dev/net/tun:/dev/net/tun --cap-add=NET_ADMIN.

Is there any disadvantage doing it like that?

Thanks

Sven

Hi,
at the moment I'm not able to run my unikernel inside the docker container in unix mode (built and run with unix hypervisor). The error I constantly get when I run /runtime/runner unix /unikernel/mir-console is:

error: execv: not foundernel/mir-console failed: No such file or directory

After the first try, if I launch the same command again, I receive permanently the following error:

No AF_INET address found on veth
error: Unable to determine IP address of eth0

After that point, I have to shutdown the docker container and restart it.
Let me add that I have anyway modified the Dockerfile.runner and Dockerfile.runner-build so that the base image I'm using is an ARM debian:jessie (I had problems with the alpine image given that it doesn't run glibc but musl) and the ENTRYPOINT is /bin/sh. Technically my unikernel should be good to go considering that all the necessary libraries are located and there are no apparent issues. The followign is the output of ldd command:

 ldd /unikernel/mir-console
        linux-vdso.so.1 (0x7ef43000)
        libz.so.1 => /lib/arm-linux-gnueabihf/libz.so.1 (0x76f29000)
        libpthread.so.0 => /lib/arm-linux-gnueabihf/libpthread.so.0 (0x76f05000)
        libgmp.so.10 => /lib/libgmp.so.10 (0x76e93000)
        libm.so.6 => /lib/arm-linux-gnueabihf/libm.so.6 (0x76e1f000)
        libdl.so.2 => /lib/arm-linux-gnueabihf/libdl.so.2 (0x76e0c000)
        libc.so.6 => /lib/arm-linux-gnueabihf/libc.so.6 (0x76d1c000)
        /lib/ld-linux-armhf.so.3 (0x54b04000) 

I believe that there is something wrong with the runner, given also the inconsistent error message prompted after subsequent executions. But that's just a guess.