Test azure

README.md

@@ -6,6 +6,7 @@
 
 - [flux-in-slurm](tutorial/flux-in-slurm): Bring up a Flux instance (in user-space) in a Slurm Allocation - both in Kubernetes ([video](https://youtu.be/8ZkSLV0m7To?si=WqWKCe2jvRuTXvlJ)) 
 - [Flux on AWS](tutorial/aws): Deploy an entire Flux Framework cluster to "bare metal" instances on AWS with (essentially) two `make` commands - one to build with packer, and one to deploy with Terraform ([video](https://youtu.be/LJh-ab6fAqE?si=dIzScA530N7lXs_7))
+- [Flux on Azure](tutorial/azure): Deploy Flux Framework on Azure with Infiniband
 - [HPCIC Tutorial 2024](https://youtu.be/Dt4CSZWSEJE?si=b2O7lQrJixcKh-EJ)
 
 ## What is this?

tutorial/.gitignore

@@ -0,0 +1,3 @@
+.terraform.lock.hcl
+.env
+.terraform

tutorial/azure/Makefile

@@ -0,0 +1,26 @@
+.PHONY: all
+all: init fmt validate apply
+
+.PHONY: init
+init: 
+	terraform init	
+
+.PHONY: fmt
+fmt: 
+	terraform fmt
+
+.PHONY: validate
+validate: 
+	terraform validate
+
+.PHONY: apply
+apply: 
+	terraform apply
+
+.PHONY: apply-approved
+apply-approved: 
+	terraform apply --auto-approve
+
+.PHONY: destroy
+destroy: 
+	terraform destroy

tutorial/azure/README.md

@@ -0,0 +1,121 @@
+# Flux on Azure
+
+## Usage
+
+### 1. Build Images
+
+Note that you should [build](build) the images first. Follow the instructions in the README there.
+
+### 2. Deploy Terraform
+
+Check the [start-script.sh](start-script.sh) and variables at the top of [main.tf](main.tf). You'll need to export the image full identifier to the environment:
+
+```bash
+export TF_VAR_vm_image_storage_reference=/subscriptions/xxxxxxx/resourceGroups/xxxxx/providers/Microsoft.Compute/images/flux-framework
+```
+
+Note that I needed to clone this and do from the cloud shell in the Azure portal.
+
+```bash
+git clone https://github.com/converged-computing/flux-tutorials
+cd flux-tutorials/tutorial/azure
+```
+
+and then:
+
+```bash
+make
+```
+
+The shell can be buggy - if it seems like it's hanging, it's that terraform is waiting for you to enter "yes." You can type it (despite not seeing it) and press enter and it works every time... 50% of the time. :) I added a command to the Makefile to get around this:
+
+```bash
+make apply-approved
+```
+
+You can also run each command separately:
+
+```bash
+# Terraform init
+make init
+
+# Terraform validate
+make validate
+
+# Create
+make apply
+
+# Destroy
+make destroy
+```
+
+When it's done, save the public and private key to local files:
+
+```bash
+terraform output -json public_key | jq -r > id_azure.pub
+terraform output -json private_key | jq -r > id_azure
+chmod 600 id_azure*
+```
+
+Then get the instance ip addresses from the command line (or portal), and ssh in!
+
+```bash
+ip_address=$(az vmss list-instance-public-ips -g terraform-testing -n flux | jq -r .[0].ipAddress)
+ssh -i ./id_azure azureuser@${ip_address}
+```
+
+To get a difference instance, just use the index (e.g., index 1 is the second instance)
+
+```bash
+follower_address=$(az vmss list-instance-public-ips -g terraform-testing -n flux | jq -r .[1].ipAddress)
+ssh -i ./id_azure azureuser@${follower_address}
+```
+
+### 3. Checks
+
+Check the cluster status, the overlay status, and try running a job:
+
+```bash
+$ flux resource list
+```
+```bash
+$ flux run -N 2 hostname
+```
+
+### 4. Cleanup
+
+This should work (but see [debugging](#debugging)).
+
+```bash
+make destroy
+```
+
+But if not, you can either delete the resource group from the console, or the command line:
+
+```bash
+az group delete --name terraform-testing
+```
+
+Note that this current build does not have flux-pmix, which might lead to issues with MPI. It's an issue of the VM base being compiled with a libpmix.so that has a different ABI than what flux is expecting. I will be looking into it.
+
+### Debugging
+
+Depending on your environment, terraform (e.g., `make` or `make destroy` doesn't always work. I get this error from the Azure Cloud Shell:
+
+```console
+terraform destroy
+random_pet.id: Refreshing state... [id=usable-grouper]
+random_string.fqdn: Refreshing state... [id=lhppiw]
+╷
+│ Error: building account: could not acquire access token to parse claims: running Azure CLI: exit status 1: ERROR: Failed to connect to MSI. Please make sure MSI is configured correctly.
+│ Get Token request returned: <Response [400]>
+│ 
+│   with provider["registry.terraform.io/hashicorp/azurerm"],
+│   on main.tf line 28, in provider "azurerm":
+│   28: provider "azurerm" {
+│ 
+╵
+make: *** [Makefile:22: destroy] Error 1
+```
+
+If I open a new cloud shell, it seems to magically go away. But you can also interact with the `az` tool (that does seem to to work) or issue commands via clicking directly in the portal.

tutorial/azure/build/Makefile

@@ -0,0 +1,18 @@
+.PHONY: all
+all: init fmt validate build
+
+.PHONY: init
+init: 
+	packer init .
+
+.PHONY: fmt
+fmt: 
+	packer fmt .
+
+.PHONY: validate
+validate: 
+	packer validate .
+
+.PHONY: build
+build: 
+	packer build flux-build.pkr.hcl

tutorial/azure/build/README.md

@@ -0,0 +1,38 @@
+# Build Packer Images
+
+Note that I needed to do this build from a cloud shell, so clone and then:
+
+```bash
+git clone https://github.com/converged-computing/flux-tutorials
+flux-tutorials/tutorial/azure/build
+```
+
+And install packer
+
+```bash
+wget https://releases.hashicorp.com/packer/1.11.2/packer_1.11.2_linux_amd64.zip
+unzip packer_1.11.2_linux_amd64.zip
+mkdir -p ./bin
+mv ./packer ./bin/
+export PATH=$(pwd)/bin:$PATH
+```
+
+Get your account information for azure as follows:
+
+```bash
+az account show 
+```
+
+And export variables in the following format. Note that the resource group needs to actually exist - I created mine in the console UI.
+
+```bash
+export AZURE_SUBSCRIPTION_ID=xxxxxxxxx
+export AZURE_TENANT_ID=xxxxxxxxxxx
+export AZURE_RESOURCE_GROUP_NAME=packer-testing
+```
+
+Then build!
+
+```bash
+make
+```