feat: allow different versions of bacalhau to be installed

gracefully handle version upgrades, ensure the attached disks re-attach

ops/tf/main.tf

@@ -32,6 +32,9 @@ module "requester_instance" {
   bacalhau_repo_disk_size = var.bacalhau_repo_disk_size
   bacalhau_otel_collector_endpoint = var.bacalhau_otel_collector_endpoint
   bacalhau_auth_token = random_string.bacalhau_auth_token.result
+
+  bacalhau_install_version = var.bacalhau_install_version
+  bacalhau_install_branch = var.bacalhau_install_branch
 }
 
 module "compute_instance" {
@@ -60,4 +63,7 @@ module "compute_instance" {
   bacalhau_local_disk_size = var.bacalhau_local_disk_size
   bacalhau_otel_collector_endpoint = var.bacalhau_otel_collector_endpoint
   bacalhau_auth_token = random_string.bacalhau_auth_token.result
+
+  bacalhau_install_version = var.bacalhau_install_version
+  bacalhau_install_branch = var.bacalhau_install_branch
 }

ops/tf/modules/cloud-init/cloud-init.yml

@@ -1,11 +1,3 @@
-#cloud-config
-packages:
-  - jq
-
-runcmd:
-  - bash /etc/install-bacalhau.sh release v1.2.1
-  - systemctl restart bacalhau
-
 write_files:
   # bacalhau config
   - path: /etc/config.yaml

ops/tf/modules/gcp/compute_instances/compute/main.tf

@@ -6,10 +6,11 @@ resource "google_compute_instance" "compute" {
   zone         = var.zone
 
   metadata = {
-    startup-script = local.bacalhau_start_script
     user-data = data.cloudinit_config.compute_cloud_init.rendered
   }
 
+  metadata_startup_script = local.bacalhau_start_script
+
   boot_disk {
     initialize_params {
       image = var.boot_image
@@ -99,8 +100,12 @@ locals {
   //
   // templating the bacalhau start script
   //
+
+  // inject custom bacalhau install based on variables.
+  bacalhau_install_cmd_content = var.bacalhau_install_version != "" ? "release ${var.bacalhau_install_version}" : var.bacalhau_install_branch != "" ? "branch ${var.bacalhau_install_branch}" : ""
   bacalhau_start_script = templatefile("${path.module}/../../../instance_files/start.sh", {
     node_type = "compute"
+    bacalhau_version_cmd = local.bacalhau_install_cmd_content
     // Add more arguments as needed
   })
 
@@ -144,13 +149,13 @@ data "cloudinit_config" "compute_cloud_init" {
     content_type = "text/cloud-config"
 
     content = templatefile("${path.module}/../../../cloud-init/cloud-init.yml", {
-      bacalhau_config_file  : base64encode(local.compute_config_content),
-      bacalhau_service_file : base64encode(local.bacalhau_service_content),
-      bacalhau_authn_policy_file : base64encode(local.bacalhau_authn_policy_content)
-      bacalhau_authz_policy_file : base64encode(local.bacalhau_authz_policy_content)
-      otel_config_file      : base64encode(local.otel_config_content)
-      otel_service_file     : base64encode(local.otel_service_content),
-      requester_ip          : var.requester_ip,
+      bacalhau_config_file        : base64encode(local.compute_config_content)
+      bacalhau_service_file       : base64encode(local.bacalhau_service_content)
+      bacalhau_authn_policy_file  : base64encode(local.bacalhau_authn_policy_content)
+      bacalhau_authz_policy_file  : base64encode(local.bacalhau_authz_policy_content)
+      otel_config_file            : base64encode(local.otel_config_content)
+      otel_service_file           : base64encode(local.otel_service_content)
+      requester_ip                : var.requester_ip
     })
   }
 }

ops/tf/modules/gcp/compute_instances/compute/variables.tf

@@ -70,7 +70,6 @@ variable "bacalhau_local_disk_size" {
   type        = number
 }
 
-
 variable "bacalhau_otel_collector_endpoint" {
   description = "The opentelemetry collector endpoint to send metrics to"
   type = string
@@ -80,3 +79,15 @@ variable "bacalhau_auth_token" {
   description = "Auth token for bacalhau api"
   type = string
 }
+
+variable "bacalhau_install_version" {
+  description = "The version or branch of bacalhau to install. If empty https://get.bacalhau.org/install.sh will be used to install"
+  type = string
+  default = ""
+}
+
+variable "bacalhau_install_branch" {
+  description = "The branch of bacalhau to install. If empty default to https://get.bacalhau.org/install.sh"
+  type = string
+  default = ""
+}

ops/tf/modules/gcp/compute_instances/requester/main.tf

@@ -5,9 +5,11 @@ resource "google_compute_instance" "requester" {
   zone         = var.zone
 
   metadata = {
-    startup-script = local.bacalhau_start_script
     user-data = data.cloudinit_config.requester_cloud_init.rendered
   }
+
+  metadata_startup_script = local.bacalhau_start_script
+
   boot_disk {
     initialize_params {
       image = var.boot_image
@@ -78,8 +80,12 @@ locals {
   //
   // templating the bacalhau start script
   //
+
+  // inject custom bacalhau install based on variables.
+  bacalhau_install_cmd_content = var.bacalhau_install_version != "" ? "release ${var.bacalhau_install_version}" : var.bacalhau_install_branch != "" ? "branch ${var.bacalhau_install_branch}" : ""
   bacalhau_start_script = templatefile("${path.module}/../../../instance_files/start.sh", {
     node_type = "requester"
+    bacalhau_version_cmd = local.bacalhau_install_cmd_content
     // Add more arguments as needed
   })
 
@@ -122,12 +128,12 @@ data "cloudinit_config" "requester_cloud_init" {
     content_type = "text/cloud-config"
 
     content = templatefile("${path.module}/../../../cloud-init/cloud-init.yml", {
-      bacalhau_config_file      : base64encode(local.requester_config_content)
-      bacalhau_service_file     : base64encode(local.bacalhau_service_content)
-      bacalhau_authn_policy_file : base64encode(local.bacalhau_authn_policy_content)
-      bacalhau_authz_policy_file : base64encode(local.bacalhau_authz_policy_content)
-      otel_config_file          : base64encode(local.otel_config_content)
-      otel_service_file         : base64encode(local.otel_service_content)
+      bacalhau_config_file        : base64encode(local.requester_config_content)
+      bacalhau_service_file       : base64encode(local.bacalhau_service_content)
+      bacalhau_authn_policy_file  : base64encode(local.bacalhau_authn_policy_content)
+      bacalhau_authz_policy_file  : base64encode(local.bacalhau_authz_policy_content)
+      otel_config_file            : base64encode(local.otel_config_content)
+      otel_service_file           : base64encode(local.otel_service_content)
     })
   }
 }

ops/tf/modules/gcp/compute_instances/requester/variables.tf

@@ -69,3 +69,15 @@ variable "bacalhau_auth_token" {
   description = "Auth token for bacalhau api"
   type = string
 }
+
+variable "bacalhau_install_version" {
+  description = "The version or branch of bacalhau to install. If empty https://get.bacalhau.org/install.sh will be used to install"
+  type = string
+  default = ""
+}
+
+variable "bacalhau_install_branch" {
+  description = "The branch of bacalhau to install. If empty default to https://get.bacalhau.org/install.sh"
+  type = string
+  default = ""
+}

ops/tf/modules/instance_files/start.sh

@@ -1,6 +1,9 @@
 #!/bin/bash
 
+set -x
+
 NODE_TYPE="${node_type}"
+BACALHAU_VERSION_CMD="${bacalhau_version_cmd}"
 
 # mount or format repo disk
 function setup-bacalhau-repo-disk() {
@@ -49,6 +52,11 @@ function install-otel-collector() {
     sudo mv otelcol-contrib /usr/local/bin/otelcol
 }
 
+function install-bacalhau() {
+    echo "Installing bacalhau"
+    bash /etc/install-bacalhau.sh $BACALHAU_VERSION_CMD
+}
+
 # reload service files and enable services
 function setup-services() {
   echo "Loading systemctl services..."
@@ -76,6 +84,10 @@ function start() {
     setup-bacalhau-local-disk
   fi
 
+  if [ "$BACALHAU_VERSION_CMD" != "" ]; then
+    install-bacalhau
+  fi
+
   # TODO move this into the VMI, maybe?
   install-otel-collector
   setup-bacalhau-config

ops/tf/variables.tf

@@ -65,3 +65,16 @@ variable "bacalhau_otel_collector_endpoint" {
   description = "The opentelemetry collector endpoint to send metrics to"
   type = string
 }
+
+
+variable "bacalhau_install_version" {
+  description = "The version of bacalhau to install. If empty default to https://get.bacalhau.org/install.sh"
+  type = string
+  default = ""
+}
+
+variable "bacalhau_install_branch" {
+  description = "The branch of bacalhau to install. If empty default to https://get.bacalhau.org/install.sh"
+  type = string
+  default = ""
+}

ops/tf/vars.tfvars

@@ -7,7 +7,7 @@ requester_machine_type ="e2-standard-8"
 compute_machine_type = "e2-standard-8"
 compute_count = 2
 
-gcp_boot_image = "projects/forrest-dev-407420/global/images/bacalhau-ubuntu-2004-lts-test"
+gcp_boot_image = "projects/forrest-dev-407420/global/images/bacalhau-ubuntu-2204-lts-test-latest"
 gcp_project_id = "forrest-dev-407420"
 gcp_region = "us-west1"
 gcp_zone = "us-west1-b"
@@ -18,4 +18,7 @@ aws_secret_access_key = ""
 
 bacalhau_accept_networked_jobs = true
 
-bacalhau_otel_collector_endpoint = ""
+bacalhau_otel_collector_endpoint = ""
+
+bacalhau_install_branch = "main"
+bacalhau_install_version = ""