Hi,
On 2022-08-14 09:50:35 +0800, Xuan Zhuo wrote:
> Sorry, I didn't get any valuable information from the logs, can you tell me how
> to get such an image? Or how your [1] script is executed.
Is there specific information you'd like from the VM? I just recreated the
problem and can extract.
The last image that succeeded getting built is publically available, so you
could create a gcp VM for that, go to /usr/src/linux, git pull, make & install
the new kernel and reproduce the problem that way. The git pull will take a
bit because it's a shallow clone...
gcloud compute instances create myvm --preemptible --project your-gcp-project --image-project pg-ci-images --image pg-ci-sid-newkernel-2022-08-12t06-52 --zone us-west1-a --custom-cpu=4 --custom-memory=4 --metadata=serial-port-enable=true
If you want to log in via serial console, you'd have set a password before
rebooting.
gcloud compute connect-to-serial-port --zone us-west1-a --project=pg-ci-images-dev myvm
Executing the script requires a gcp key with the right to create instances and
images. Here's how to invoke it:
PACKER_LOG=1 GOOGLE_APPLICATION_CREDENTIALS=~/image-builder@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx \
packer build \
-var gcp_project=pg-ci-images-dev \
-var "image_date=$(date --utc +'%Y-%m-%dt%H-%M')" \
-var "task_name=sid-newkernel" \
-only 'linux.googlecompute.sid-newkernel' \
-on-error=ask \
packer/linux_debian.pkr.hcl
Of course you'd need to change the gcp_project= variable to point to a the
project you have access to and GOOGLE_APPLICATION_CREDENTIALS to point to your
gcp key.
Initially (package upgrades, kernel builds) the VM would be SSH
accessible. After building the kernel it's only accessible via serial console.
I can probably also get you the image in some other form that you prefer,
although I don't know if the problem will reproduce outside gcp. If helpful I
could upload a "broken" gcp image that you could use to
> > [1] https://github.com/anarazel/pg-vm-images/blob/main/packer/linux_debian.pkr.hcl#L225
Greetings,
Andres Freund
