Repository navigation
Releases
Cross Platform Action 1.8.0
Added
- Add support for OmniOS r151060
- Add support for NetBSD on SPARC64 (#41)
- Add support for NetBSD 11.1
Changed
- Bump the bundled QEMU to 11.1.3
Cross Platform Action 1.7.0
Cross Platform Action 1.6.0
Added
- Add support for FreeBSD 14.5 (#168)
- Add support for NetBSD 9.5
Fixed
- DragonFly BSD guests no longer occasionally start a job without working
connectivity (#164).
The image configured its interface withdhclient, which daemonizes before
the DHCP exchange finishes, so the boot never waited for the network and the
first step could run while the guest still had no route off its own subnet.
The image now configures the interface statically, from the lease user mode
networking hands out identically on every boot
Cross Platform Action 1.5.0
Changed
- NetBSD images download around 45% smaller -- 11.0 x86-64 goes from 497 MiB to
268 MiB -- which is most of what a NetBSD job spends on setup
(#151) - Every guest becomes usable sooner, by up to 8 seconds, because the action no
longer waits on a readiness probe it sent before the guest was listening - NetBSD guests are logged into without a credential, so no SSH key is generated
and no resources disk is built to carry one. A custom image supplied through
image_urlstill gets both, since it may expect them
Added
- Add support for Haiku R1/beta6
(#165) - A
variantinput, selecting a named configuration of a platform. Defaults to
default, which boots exactly as before, so no existing workflow changes.
See Variants - The
microvmvariant for NetBSD onx86-64, which reaches a usable guest in
roughly half the time. It is opt-in because it changes the hardware the guest
sees: the root disk becomesld0rather thansd0, there is no PCI bus to
inspect, anduname -vreports aMICROVMkernel. Asking for it where it
cannot be booted is an error, not a slow boot - Log a breakdown of how long each phase of setting up the VM took, together
with how long the VM took to become reachable over SSH
Fixed
- 5-level paging (LA57) is no longer exposed to the guests on x86-64
(#158).
FreeBSD 13.0 enables 5-level paging whenever the CPU reports it and panics
in the trampoline that switches to it, so every job that landed on an Intel
runner from Ice Lake onwards failed to boot - STIBP always-on mode is no longer exposed to the guests on x86-64
(#158). Some
of the AMD runners report it without the STIBP and IBRS bits that normally
come with it, which made DragonFly BSD writeIA32_SPEC_CTRLand take a
general protection fault while booting
Security
- The guest's SSH port is only forwarded to the runner's loopback address. It was
previously bound to every interface, making the guest reachable from anything
that could reach the runner
Cross Platform Action 1.4.0
Added
- Add support for FreeBSD on RISC-V 64 (
riscv64) (#36) - Add support for NetBSD on VAX, running on the SIMH simulator
- Add support for NetBSD 11.0 (#154)
Changed
- OpenBSD on ARM64 now boots with the same EDK II UEFI firmware (
uefi.fd) as
every other ARM64 guest, instead of a separate Linaro UEFI build. The QEMU
machine type for OpenBSD on ARM64 is nowvirt,acpi=off, which makes the
kernel fall back to the device tree instead of hanging during ACPI attach.
This removes the dependency on the Linaro release server, which no longer
exists
Fixed
- Don't log the
Tearing down VMgroup whenshutdown_vmisfalse, since
the VM is not being torn down in that case
(#109) - The "Start VM" step could keep running for hours when the VM failed to boot
(#158).
Waiting for the VM to become ready is now bounded by an actual wall clock
timeout, instead of by a number of connection attempts, and a single SSH
connection attempt is bounded byConnectTimeout. - The post job step now reports when the console log of the VM is missing or
empty, instead of silently printing nothing
(#158) - A reboot of a guest that dies instead of rebooting now fails after roughly
20 seconds, instead of after 13 minutes. The SSH session that issues the
reboot is bounded byServerAliveInterval, so it no longer blocks for the
full TCP retransmission time, and waiting for the VM to come back up is
given up on as soon as the guest announces a kernel panic on its serial
console - Advanced Matrix Extensions (AMX) is no longer exposed to the guests on
x86-64, so which runner a job happens to get no longer decides whether the
VM boots
(#158).
Kernels released before AMX existed fault as soon as userland starts, which
affected NetBSD 9.4 and 10.1 and FreeBSD 12.4 on the runners that have it
Cross Platform Action 1.3.0
Added
-
Add support for FreeBSD 15.1 (#153)
-
Add a
--environment-variablespost-flag to thecpa.shcustom shell for
forwarding additional environment variables to the VM on a per-step basis,
without having to declare them on theStart VMstep via the
environment_variablesinput
(#145):- name: Run a command with extra environment variables shell: cpa.sh {0} --environment-variables MY_ENV1 MY_ENV2 env: MY_ENV1: value1 MY_ENV2: value2 run: echo "$MY_ENV1 $MY_ENV2"
Cross Platform Action 1.2.0
Cross Platform Action 1.1.0
Added
-
Add support for DragonFly BSD (#19)
-
Add support for MidnightBSD (#102)
-
Add support for FreeBSD 14.4 (#122)
-
Add support for OmniOS r151058
-
New syntax for multiple steps (#83).
Instead of invoking the action multiple times it's now possible to use a
custom shell when running commands:jobs: custom-shell: runs-on: ubuntu-latest defaults: run: shell: cpa.sh {0} steps: - name: Start VM uses: cross-platform-actions/action@master with: operating_system: freebsd architecture: x86-64 version: '15.0' - name: Run command using custom shell run: '[ "`uname`" = FreeBSD ]'
Each custom-shell step automatically synchronizes files in both
directions: runner-to-vm before the step runs and vm-to-runner after.
Pass--sync-files DIRECTIONafter the file argument to change this
(both(default),none(skip sync),runner-to-vm, or
vm-to-runner), or usecpa.sh --sync-filesstandalone to sync on
demand without running a command:- name: Sync files from runner to VM run: cpa.sh --sync-files runner-to-vm
-
Reboot mode for rebooting the VM and waiting for it to come back up
(#103,
#118).
cpa.sh --rebootissues the reboot and blocks until the VM is reachable
again:- name: Reboot VM run: cpa.sh --reboot
Security
- Bump builders to releases that use immutable releases, providing
integrity verification for downloaded artifacts
(#140)
Deprecated
-
The
runinput parameter has been deprecated and is now optional. Use the
custom shell (shell: cpa.sh {0}) in subsequent steps to run commands in
the virtual machine instead. -
The
shutdown_vminput parameter has been deprecated and will be removed
in a future release. There is no replacement. When unset, it now defaults
totrueif therunparameter is provided (preserving the legacy
behavior) andfalseotherwise (so the VM stays alive across subsequent
custom-shell steps without needing to specify it).
Cross Platform Action 1.0.0
Fixed
- Fix #108: Fix file ownership on Haiku after rsync, resolving git
safe.directoryerrors
Changed
- Breaking: Update the requirement of Node for running this action from
version 20 to 24.
Removed
- Breaking: Remove support for running on macOS runners. Only Linux runners
(e.g.ubuntu-latest) are now supported. This was deprecated in v0.25.0. - Breaking: Remove the Xhyve hypervisor and the
hypervisorinput
parameter. QEMU is now the only supported hypervisor. These were deprecated
in v0.25.0.
Cross Platform Action 0.32.0
Added
- Add support for OmniOS