GenAIChat.yml

# The version of the Xpanse description language
version: 1.0
# The category of the service.
category: compute
# The Service provided by the ISV, the name will be shown on the console as a service.
name: terraform-ecs-genai
# The version of the service, the end-user can select the version they want to deploy.
serviceVersion: 1.0.0
# For the users may have more than one service, the @namespace can be used to separate the clusters.
description: This is an enhanced compute services by ISV-A.
namespace: ISV-A
# Icon for the service.
icon: |
  data:image/png;base64,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
# Reserved for CSP, HuaweiCloud. ap-southeast-3 refers to Singapore region.
cloudServiceProvider:
  name: HuaweiCloud
  regions:
    - name: ap-southeast-3
      site: Chinese Mainland
      area: Asia China
    - name: ap-southeast-3
      site: International
      area: Asia China
    - name: eu-west-0
      site: International
      area: Europe Pairs
    - name: eu-west-101
      site: Europe
      area: Europe Dublin
billing:
  # The supported mode of billing (`Fixed`, `Pay per Use`)
  billingModes:
    - Fixed
    - Pay per Use
  defaultBillingMode: Fixed
serviceHostingType: self
# The flavor of the service, the @category/@name/@version/@flavor can locate the specific service to be deployed.
flavors:
  serviceFlavors:
    - name: 16vCPUs-64GB-normal
      priority: 1
      # The pricing of the flavor.
      pricing:
        # Used to calculate charges when users select 'pay_per_use' as the billing mode.
        resourceUsage:
          resources:
            - deployResourceKind: vm
              count: 1
              properties:
                cloud_service_type: hws.service.type.ec2
                resource_type: hws.resource.type.vm
                resource_spec: c7t.4xlarge.4.linux
            - deployResourceKind: volume
              count: 1
              properties:
                cloud_service_type: hws.service.type.ebs
                resource_type: hws.resource.type.volume
                resource_spec: SSD
                resource_size: 1000
                size_measure_id: 17
            - deployResourceKind: publicIP
              count: 1
              properties:
                cloud_service_type: hws.service.type.vpc
                resource_type: hws.resource.type.bandwidth
                resource_spec: 12_bgp
                resource_size: 300
                size_measure_id: 15
          licensePrices:
            - regionName: any
              siteName: Chinese Mainland
              price:
                cost: 1.50
                currency: CNY
                period: hourly
            - regionName: any
              siteName: International
              price:
                cost: 0.0045
                currency: USD
                period: hourly
            - regionName: eu-west-101
              siteName: Europe
              price:
                cost: 0.0045
                currency: USD
                period: hourly
          markUpPrices:
            - regionName: any
              siteName: Chinese Mainland
              price:
                cost: 1.50
                currency: CNY
                period: hourly
            - regionName: any
              siteName: International
              price:
                cost: 0.0045
                currency: USD
                period: hourly
            - regionName: any
              siteName: Europe
              price:
                cost: 0.0045
                currency: USD
                period: hourly
        # Used to calculate charges when users do not select 'pay_per_use' as the billing mode.
        fixedPrices:
          - regionName: any
            siteName: Chinese Mainland
            price:
              cost: 360
              currency: CNY
              period: monthly
          - regionName: any
            siteName: International
            price:
              cost: 35.00
              currency: USD
              period: monthly
          - regionName: any
            siteName: Europe
            price:
              cost: 35.00
              currency: USD
              period: monthly
        isPriceOnlyForManagementLayer: false
      # Properties for the service, which can be used by the deployment.
      properties:
        flavor_id: c7t.4xlarge.4
      features:
        - High Availability
        - Maximum performance
  modificationImpact:
    isDataLost: false
    isServiceInterrupted: true
  isDowngradeAllowed: true
# The contact details of the service.
serviceProviderContactDetails:
  emails: [ "test30@test.com","test31@test.com" ]
  phones: [ "011-13422222222","022-13344444444" ]
  chats: [ "test1234","test1235" ]
  websites: [ "https://hw.com","https://hwcloud.com" ]
# End user license agreement content of the service.
eula: |
  This Acceptable Use Policy ("Policy") lists prohibited conduct and content when using the services provided by or on behalf of HUAWEI CLOUD and its affiliates. This Policy is an integral part of the HUAWEI CLOUD User Agreement ("User Agreement"). The examples and restrictions listed below are not exhaustive. We may update this Policy from time to time, and the updated Policy will be posted on the Website. By continuing to use the Services, you agree to abide by the latest version of this Policy. You acknowledge and agree that we may suspend or terminate the Services if you or your users violate this Policy. Terms used in the User Agreement have the same meanings in this Policy.

  Prohibited Conduct
  When accessing or using the Services, or allowing others to access or use the Services, you may not:
  1. Violate any local, national or international laws, regulations and rules;
  2. Infringe or violate the rights of others, including but not limited to privacy rights or intellectual property rights;
  3. Engage in, encourage, assist or allow others to engage in any illegal, unlawful, infringing, harmful or fraudulent behavior, including but not limited to any of the following activities: harming or attempting to harm minors in any way, pornography, illegal gambling, illegal VPN construction, Ponzi schemes, cyber attacks, phishing or damage, privately intercepting any system, program or data, monitoring service data or traffic without permission, engaging in virtual currency "mining" or virtual currency transactions;
  4. Transmit, provide, upload, download, use or reuse, disseminate or distribute any illegal, infringing, offensive, or harmful content or materials, including but not limited to those listed in the "Prohibited Content" below;
  5. Transmit any data, send or upload any material that contains viruses, worms, Trojan horses, time bombs, keyboard loggers, spyware, adware or any other harmful programs or similar computer code designed to adversely affect the operation or security of any computer hardware or software;
  6. Attack, interfere with, disrupt or adversely affect any service, hardware, software, system, website or network, including but not limited to accessing or attacking any service, hardware, software, system, website or network using large amounts of automated means (including robots, crawlers, scripts or similar data gathering or extraction methods);
  7. Access any part of the Service, account or system without authorization, or attempt to do so;
  8. Violate or adversely affect the security or integrity of the Services, hardware, software, systems, websites or networks;
  9. Distribute, disseminate or send unsolicited email, bulk email or other messages, promotions, advertising or solicitations (such as "spam");
  10. Fraudulent offers of goods or services, or any advertising, promotional or other materials containing false, deceptive or misleading statements.
deployment:
  deployerTool:
    # kind, Supported values are terraform, opentofu.
    kind: terraform
    # version, the required version of the deployer tool for the deployer scripts.
    version: "=1.6.0"
  serviceAvailabilityConfig:
    - displayName: Availability Zone
      varName: availability_zone
      mandatory: false
      description: The availability zone to deploy the service instance. If the value is empty, the service instance will be deployed in a random availability zone.
  # Context for deployment: the context including some kind of parameters for the deployment, such as fix_env, fix_variable, env, variable, env_env, env_variable.
  # - fix_env: Values for variable of this type are defined by the managed service provider in the OCL template. Runtime will inject it to deployer as environment variables. This variable is not visible to the end user.
  # - fix_variable: Values for variable of this type are defined by the managed service provider in the OCL template. Runtime will inject it to deployer as usual variables. This variable is not visible to the end user.
  # - env: Value for a variable of this type can be provided by end user. If marked as mandatory then end user must provide value to this variable. If marked as optional and if end user does not provided it, then the fallback value to this variable is read by runtime (it can read from other sources, e.g., OS env variables). This variable is injected as a environment variable to the deployer.
  # - variable: Value for a variable of this type can be provided by end user. . If marked as mandatory then end user must provide value to this variable. If marked as optional and if end user does not provided it, then the fallback value to this variable is read by runtime (it can read from other sources, e.g., OS env variables). This variable is injected as a regular variable to the deployer.
  # - env_env: Value to this variable is read by runtime (it can read from other sources, e.g., OS env variables) and injected as a environment variable to the deployer. End user cannot see or change this variable.
  # - env_variable: Value to this variable is read by runtime (it can read from other sources, e.g., OS env variables) and injected as a regular variable to the deployer. End user cannot see or change this variable.
  # The parameters will be used to generate the API of the managed service.
  variables:
    - name: admin_passwd
      description: The admin password of the compute instance. If the value is empty, will create a random password.
      kind: variable
      dataType: string
      mandatory: false
      valueSchema:
        minLength: 8
        maxLength: 16
        pattern: ^(?=.*?[A-Z])(?=.*?[a-z])(?=.*?[0-9])(?=.*?[#?!@$%^&*-]).{8,16}$
      modificationImpact:
        isDataLost: false
        isServiceInterrupted: true
    - name: image_name
      description: The image name of the compute instance. If the value is empty, will use the default value to create compute instance.
      kind: variable
      dataType: string
      example: "Ubuntu 22.04 server 64bit"
      mandatory: false
      value: "Ubuntu 22.04 server 64bit"
      modificationImpact:
        isDataLost: false
        isServiceInterrupted: true
    - name: vpc_name
      description: The vpc name of the compute instance. If the value is empty, will use the default value to find or create VPC.
      kind: variable
      dataType: string
      example: "ecs-vpc-default"
      mandatory: false
      value: "ecs-vpc-default"
      modificationImpact:
        isDataLost: false
        isServiceInterrupted: true
    - name: subnet_name
      description: The sub network name of the compute instance. If the value is empty, will use the default value to find or create subnet.
      kind: variable
      dataType: string
      example: "ecs-subnet-default"
      mandatory: false
      value: "ecs-subnet-default"
      modificationImpact:
        isDataLost: false
        isServiceInterrupted: true
    - name: secgroup_name
      description: The security group name of the compute instance. If the value is empty, will use the default value to find or create security group.
      kind: variable
      dataType: string
      example: "ecs-secgroup-default"
      mandatory: false
      value: "ecs-secgroup-default"
      modificationImpact:
        isDataLost: false
        isServiceInterrupted: true
  deployer: |
    variable "region" {
      type        = string
      description = "The region to deploy the compute instance."
    }
    
    variable "availability_zone" {
      type        = string
      default     = ""
      description = "The availability zone to deploy the compute instance."
    }
    
    variable "flavor_id" {
      type        = string
      default     = "c7t.4xlarge.4"
      description = "The flavor_id of the compute instance."
    }
    
    variable "image_name" {
      type        = string
      default     = "Ubuntu 22.04 server 64bit"
      description = "The image name of the compute instance."
    }
    
    variable "admin_passwd" {
      type        = string
      default     = ""
      description = "The root password of the compute instance."
    }

    variable "vpc_name" {
      type        = string
      default     = "ecs-vpc-default"
      description = "The vpc name of the compute instance."
    }
    
    variable "subnet_name" {
      type        = string
      default     = "ecs-subnet-default"
      description = "The subnet name of the compute instance."
    }

    variable "secgroup_name" {
      type        = string
      default     = "ecs-secgroup-default"
      description = "The security group name of the compute instance."
    }
    
    terraform {
      required_providers {
        huaweicloud = {
          source  = "huaweicloud/huaweicloud"
          version = "~> 1.61.0"
        }
      }
    }
    
    provider "huaweicloud" {
      region = var.region
    }

    data "huaweicloud_availability_zones" "osc-az" {}

    data "huaweicloud_vpcs" "existing" {
      name = var.vpc_name
    }

    data "huaweicloud_vpc_subnets" "existing" {
      name = var.subnet_name
    }

    data "huaweicloud_networking_secgroups" "existing" {
      name = var.secgroup_name
    }

    locals {
      availability_zone = var.availability_zone == "" ? data.huaweicloud_availability_zones.osc-az.names[0] : var.availability_zone
      admin_passwd      = var.admin_passwd == "" ? random_password.password.result : var.admin_passwd
      vpc_id            = length(data.huaweicloud_vpcs.existing.vpcs) > 0 ? data.huaweicloud_vpcs.existing.vpcs[0].id : huaweicloud_vpc.new[0].id
      subnet_id         = length(data.huaweicloud_vpc_subnets.existing.subnets)> 0 ? data.huaweicloud_vpc_subnets.existing.subnets[0].id : huaweicloud_vpc_subnet.new[0].id
      secgroup_id       = length(data.huaweicloud_networking_secgroups.existing.security_groups) > 0 ? data.huaweicloud_networking_secgroups.existing.security_groups[0].id : huaweicloud_networking_secgroup.new[0].id
    }

    resource "huaweicloud_vpc" "new" {
      count = length(data.huaweicloud_vpcs.existing.vpcs) == 0 ? 1 : 0
      name  = var.vpc_name
      cidr  = "192.168.0.0/16"
    }

    resource "huaweicloud_vpc_subnet" "new" {
      count      = length(data.huaweicloud_vpcs.existing.vpcs) == 0 ? 1 : 0
      vpc_id     = local.vpc_id
      name       = var.subnet_name
      cidr       = "192.168.10.0/24"
      gateway_ip = "192.168.10.1"
    }

    resource "huaweicloud_networking_secgroup" "new" {
      count       = length(data.huaweicloud_networking_secgroups.existing.security_groups) == 0 ? 1 : 0
      name        = var.secgroup_name
      description = "Kafka cluster security group"
    }

    resource "huaweicloud_networking_secgroup_rule" "secgroup_rule_0" {
      count             = length(data.huaweicloud_networking_secgroups.existing.security_groups) == 0 ? 1 : 0
      direction         = "ingress"
      ethertype         = "IPv4"
      protocol          = "tcp"
      port_range_min    = 22
      port_range_max    = 22
      remote_ip_prefix  = "121.37.117.211/32"
      security_group_id = local.secgroup_id
    }

    resource "huaweicloud_networking_secgroup_rule" "secgroup_rule_1" {
      count             = length(data.huaweicloud_networking_secgroups.existing.security_groups) == 0 ? 1 : 0
      direction         = "ingress"
      ethertype         = "IPv4"
      protocol          = "tcp"
      port_range_min    = 8080
      port_range_max    = 8088
      remote_ip_prefix  = "121.37.117.211/32"
      security_group_id = local.secgroup_id
    }

    resource "huaweicloud_networking_secgroup_rule" "secgroup_rule_2" {
      count             = length(data.huaweicloud_networking_secgroups.existing.security_groups) == 0 ? 1 : 0
      direction         = "ingress"
      ethertype         = "IPv4"
      protocol          = "tcp"
      port_range_min    = 9090
      port_range_max    = 9099
      remote_ip_prefix  = "121.37.117.211/32"
      security_group_id = local.secgroup_id
    }

    resource "random_id" "new" {
      byte_length = 4
    }

    resource "random_password" "password" {
      length           = 12
      upper            = true
      lower            = true
      numeric          = true
      special          = true
      min_special      = 1
      override_special = "#%@"
    }

    data "huaweicloud_images_image" "image" {
      name                  = var.image_name
      most_recent           = true
      enterprise_project_id = "0"
    }
    
    resource "huaweicloud_evs_volume" "volume" {
      name              = "volume-tf-${random_id.new.hex}"
      description       = "my volume"
      volume_type       = "SSD"
      size              = 1000
      availability_zone = local.availability_zone
      tags = {
        foo = "bar"
        key = "value"
      }
    }
    
    resource "huaweicloud_compute_volume_attach" "attached" {
      instance_id = huaweicloud_compute_instance.ecs-tf.id
      volume_id   = huaweicloud_evs_volume.volume.id
    }
    
    resource "huaweicloud_vpc_eip" "eip-tf" {
      publicip {
         type = "5_bgp"
      }
      bandwidth {
        name        = "eip-tf-${random_id.new.hex}"
        size        = 300
        share_type  = "PER"
        charge_mode = "traffic"
      }
    }
    
    resource "huaweicloud_compute_eip_associate" "associated" {
      public_ip   = huaweicloud_vpc_eip.eip-tf.address
      instance_id = huaweicloud_compute_instance.ecs-tf.id
    }
    
    resource "huaweicloud_compute_keypair" "keypair" {
      name = "keypair-xpanse-6"
    }
    
    locals {
      instance_public_ip = huaweicloud_vpc_eip.eip-tf.address
    }
    
    resource "huaweicloud_compute_instance" "ecs-tf" {
      availability_zone  = local.availability_zone
      name               = "ecs-tf-${random_id.new.hex}"
      flavor_id          = var.flavor_id
      system_disk_type   = "SSD"
      system_disk_size   = 1000
      security_group_ids = [ local.secgroup_id ]
      image_id           = data.huaweicloud_images_image.image.id
      admin_pass         = local.admin_passwd
      key_pair       = huaweicloud_compute_keypair.keypair.name   
      network {uuid = local.subnet_id}
      user_data = <<-EOF
          #!bin/bash
          sudo apt-get update
          sudo apt-get install \
          apt-transport-https \
          ca-certificates \
          curl \
          software-properties-common
          curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
          sudo add-apt-repository --yes \
          "deb [arch=amd64] https://download.docker.com/linux/ubuntu \
          $(lsb_release -cs) \
          stable"
          sudo apt-get update
          sudo apt-get --yes install docker-ce
          sudo systemctl start docker
          sudo systemctl enable docker
          curl -SL https://github.com/docker/compose/releases/download/v2.30.3/docker-compose-linux-x86_64 -o /usr/local/bin/docker-compose
          git clone https://github.com/opea-project/GenAIExamples.git
          export host_ip="${local.instance_public_ip}"
          # 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
          export HUGGINGFACEHUB_API_TOKEN="hf"
          export EMBEDDING_MODEL_ID="BAAI/bge-base-en-v1.5"
          export RERANK_MODEL_ID="BAAI/bge-reranker-base"
          export LLM_MODEL_ID="Intel/neural-chat-7b-v3-3"
          export INDEX_NAME="rag-redis"
          cd /GenAIExamples/ChatQnA/docker_compose/intel/cpu/xeon/
          source ./set_env.sh
          docker compose up -d
      EOF
    }
    
    data "local_file" "example" {
      filename = pathexpand("./keypair-xpanse-6.pem")
      depends_on = [huaweicloud_compute_keypair.keypair]
    }
        
    output "pem_content" {
      value = data.local_file.example.content
    }
          
    output "ecs-host" {
      value = huaweicloud_compute_instance.ecs-tf.access_ip_v4
    }
    
    output "ecs-public-ip" {
      value = huaweicloud_vpc_eip.eip-tf.address
    }

    output "admin_passwd" {
      value = var.admin_passwd == "" ? nonsensitive(local.admin_passwd) : local.admin_passwd
    }