ab6e8d28aa
Over the past couple of months we've been using redb instead of sled as the default filesystem-based database in PS and DS. I am confident that we can get rid of sled completely now, and just keep redb. Change-Id: I11fa1e4453e280253855f8eade990b37eb6965ae Reviewed-on: https://cl.tvl.fyi/c/depot/+/12567 Reviewed-by: yuka <yuka@yuka.dev> Reviewed-by: flokli <flokli@flokli.de> Tested-by: BuildkiteCI Autosubmit: Ilan Joselevich <personal@ilanjoselevich.com> |
||
---|---|---|
.. | ||
tests | ||
default.nix | ||
README.md | ||
tvix-init.go |
tvix/boot
This directory provides tooling to boot VMs with /nix/store provided by virtiofs.
In the tests/
subdirectory, there's some integration tests.
//tvix/boot:runVM
A script spinning up a tvix-store virtiofs
daemon, then starting a cloud-
hypervisor VM.
The cloud-hypervisor VM is using a (semi-)minimal kernel image with virtiofs support, and a custom initrd (using u-root). It supports various command line options, to be able to do VM tests, act as an interactive shell or exec a binary from a closure.
It supports the following env vars:
CH_NUM_CPUS=1
controls the number of CPUs available to the VMCH_MEM_SIZE=512M
controls the memory availabe to the VMCH_CMDLINE=
controls the kernel cmdline (which can be used to control the boot)
Usage
First, ensure you have tvix-store
in $PATH
, as that's what run-tvix-vm
expects:
Assuming you ran cargo build --profile=release-with-debug
before, and are in
the tvix
directory:
export PATH=$PATH:$PWD/target/release-with-debug
Now, spin up tvix-daemon, connecting to some (local) backends:
tvix-store --otlp=false daemon \
--blob-service-addr=objectstore+file://$PWD/blobs \
--directory-service-addr=redb://$PWD/directories.redb \
--path-info-service-addr=redb://$PWD/pathinfo.redb &
Copy some data into tvix-store (we use nar-bridge
for this for now):
mg run //tvix:nar-bridge -- --otlp=false &
rm -Rf ~/.cache/nix; nix copy --to http://localhost:9000\?compression\=none $(mg build //third_party/nixpkgs:hello)
pkill nar-bridge
By default, the tvix-store virtiofs
command used in the runVM
script
connects to a running tvix-store daemon
via gRPC - in which case you want to
keep tvix-store daemon
running.
In case you want to have tvix-store virtiofs
open the stores directly, kill
tvix-store daemon
too, and export the addresses from above:
pkill tvix-store
export BLOB_SERVICE_ADDR=objectstore+file://$PWD/blobs
export DIRECTORY_SERVICE_ADDR=redb://$PWD/directories.redb
export PATH_INFO_SERVICE_ADDR=redb://$PWD/pathinfo.redb
Interactive shell
Run the VM like this:
CH_CMDLINE=tvix.shell mg run //tvix/boot:runVM --
You'll get dropped into an interactive shell, from which you can do things with the store:
______ _ ____ _ __
/_ __/ __(_) __ / _/___ (_) /_
/ / | | / / / |/_/ / // __ \/ / __/
/ / | |/ / /> < _/ // / / / / /_
/_/ |___/_/_/|_| /___/_/ /_/_/\__/
/# ls -la /nix/store/
dr-xr-xr-x root 0 0 Jan 1 00:00 .
dr-xr-xr-x root 0 989 Jan 1 00:00 aw2fw9ag10wr9pf0qk4nk5sxi0q0bn56-glibc-2.37-8
dr-xr-xr-x root 0 3 Jan 1 00:00 jbwb8d8l28lg9z0xzl784wyb9vlbwss6-xgcc-12.3.0-libgcc
dr-xr-xr-x root 0 82 Jan 1 00:00 k8ivghpggjrq1n49xp8sj116i4sh8lia-libidn2-2.3.4
dr-xr-xr-x root 0 141 Jan 1 00:00 mdi7lvrn2mx7rfzv3fdq3v5yw8swiks6-hello-2.12.1
dr-xr-xr-x root 0 5 Jan 1 00:00 s2gi8pfjszy6rq3ydx0z1vwbbskw994i-libunistring-1.1
Once you exit the shell, the VM will power off itself.
Execute a specific binary
Run the VM like this:
hello_cmd=$(mg build //third_party/nixpkgs:hello)/bin/hello
CH_CMDLINE=tvix.run=$hello_cmd mg run //tvix/boot:runVM --
Observe it executing the file (and closure) from the tvix-store:
[ 0.277486] Run /init as init process
______ _ ____ _ __
/_ __/ __(_) __ / _/___ (_) /_
/ / | | / / / |/_/ / // __ \/ / __/
/ / | |/ / /> < _/ // / / / / /_
/_/ |___/_/_/|_| /___/_/ /_/_/\__/
Hello, world!
2023/09/24 21:10:19 Nothing left to be done, powering off.
[ 0.299122] ACPI: PM: Preparing to enter system sleep state S5
[ 0.299422] reboot: Power down
Boot a NixOS system closure
It's also possible to boot a system closure. To do this, tvix-init honors the
init= cmdline option, and will switch_root
to it.
Make sure to first copy that system closure into tvix-store,
using a similar nix copy
comamnd as above.
CH_CMDLINE=init=/nix/store/…-nixos-system-…/init mg run //tvix/boot:runVM --
______ _ ____ _ __
/_ __/ __(_) __ / _/___ (_) /_
/ / | | / / / |/_/ / // __ \/ / __/
/ / | |/ / /> < _/ // / / / / /_
/_/ |___/_/_/|_| /___/_/ /_/_/\__/
2023/09/24 21:16:43 switch_root: moving mounts
2023/09/24 21:16:43 switch_root: Skipping "/run" as the dir does not exist
2023/09/24 21:16:43 switch_root: Changing directory
2023/09/24 21:16:43 switch_root: Moving /
2023/09/24 21:16:43 switch_root: Changing root!
2023/09/24 21:16:43 switch_root: Deleting old /
2023/09/24 21:16:43 switch_root: executing init
<<< NixOS Stage 2 >>>
[ 0.322096] booting system configuration /nix/store/g657sdxinpqfcdv0162zmb8vv9b5c4c5-nixos-system-client-23.11.git.82102fc37da
running activation script...
setting up /etc...
starting systemd...
[ 0.980740] systemd[1]: systemd 253.6 running in system mode (+PAM +AUDIT -SELINUX +APPARMOR +IMA +SMACK +SECCOMP +GCRYPT -GNUTLS +OPENSSL +ACL +BLKID +CURL +ELFUTILS +FIDO2 +IDN2 -IDN +IPTC +KMOD +LIBCRYPTSETUP +LIBFDISK +PCRE2 -PWQUALITY +P11KIT -QRENCODE +TPM2 +BZIP2 +LZ4 +XZ +ZLIB +ZSTD +BPF_FRAMEWORK -XKBCOMMON +UTMP -SYSVINIT default-hierarchy=unified)
This effectively replaces the NixOS Stage 1 entirely.