.merge_train_rule: &merge_train_rule
  UNIT_TEST: "yes"
  UNIT_TEST_REPEAT: 1
  UNIT_TEST_TIMEOUT: 30
  INTEGRATION_TEST: "yes"
  INTEGRATION_TEST_SCOPE: mr
  FUNCTIONAL_TEST: "yes"
  FUNCTIONAL_TEST_SCOPE: mr-slim
  FUNCTIONAL_TEST_REPEAT: 5
  FUNCTIONAL_TEST_TIME_LIMIT: 2700
  CLUSTER_A100: ""
  CLUSTER_H100: ""
  PUBLISH: "no"

workflow:
  rules:
    # Do not trigger for forks
    - if: $CI_PROJECT_NAMESPACE != "ADLR" || ($CI_PIPELINE_SOURCE == "merge_request_event" && $CI_MERGE_REQUEST_PROJECT_PATH != "ADLR/megatron-lm")
      when: never

    # ci-branches only for schedule
    - if: $CI_COMMIT_BRANCH =~ /ci-/ && $CI_PIPELINE_SOURCE != "schedule"
      when: never

    # For schedules pipelines
    - if: $CI_PIPELINE_SOURCE == "schedule"
      auto_cancel:
        on_new_commit: none

    # For manual pipelines
    - if: $CI_PIPELINE_SOURCE == "web"

    # For push to main
    - if: $CI_PIPELINE_SOURCE == 'push' && $CI_COMMIT_REF_PROTECTED == "true"
      variables:
        UNIT_TEST: "no"
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "yes"
        FUNCTIONAL_TEST_SCOPE: mr
        FUNCTIONAL_TEST_REPEAT: 5
        FUNCTIONAL_TEST_RECORD_CHECKPOINTS: "no"
        FUNCTIONAL_TEST_TIME_LIMIT: 2700
        CLUSTER_A100: ""
        CLUSTER_H100: ""
        PUBLISH: "no"
      auto_cancel:
        on_new_commit: none

    # For merge-trains that need to be fast-tracked
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merge_train' && $CI_MERGE_REQUEST_LABELS =~ /fast-track/
      variables:
        UNIT_TEST: "yes"
        UNIT_TEST_REPEAT: 1
        UNIT_TEST_TIMEOUT: 30
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "no"
        CLUSTER_A100: ""
        CLUSTER_H100: ""
        PUBLISH: "no"

    # For normal merge-trains
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merge_train'
      variables: *merge_train_rule

    # For MRs with integration suite
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merged_result' && $CI_MERGE_REQUEST_LABELS =~ /Run tests/
      variables: *merge_train_rule

    # For MRs with nightly
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merged_result' && $CI_MERGE_REQUEST_LABELS =~ /Run nightly/
      variables:
        UNIT_TEST: "yes"
        UNIT_TEST_REPEAT: 1
        UNIT_TEST_TIMEOUT: 30
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "yes"
        FUNCTIONAL_TEST_SCOPE: nightly
        FUNCTIONAL_TEST_REPEAT: 5
        FUNCTIONAL_TEST_RECORD_CHECKPOINTS: "no"
        FUNCTIONAL_TEST_TIME_LIMIT: 2700
        CLUSTER_A100: ""
        CLUSTER_H100: ""
        PUBLISH: "no"

    # For MRs with weekly
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merged_result' && $CI_MERGE_REQUEST_LABELS =~ /Run weekly/
      variables:
        UNIT_TEST: "yes"
        UNIT_TEST_REPEAT: 1
        UNIT_TEST_TIMEOUT: 30
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "yes"
        FUNCTIONAL_TEST_SCOPE: weekly
        FUNCTIONAL_TEST_REPEAT: 1
        FUNCTIONAL_TEST_RECORD_CHECKPOINTS: "no"
        FUNCTIONAL_TEST_TIME_LIMIT: 9000
        CLUSTER_A100: ""
        CLUSTER_H100: ""
        PUBLISH: "no"

    # For MRs with heavy suite
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merged_result' && $CI_MERGE_REQUEST_LABELS =~ /Run functional tests/
      variables:
        UNIT_TEST: "yes"
        UNIT_TEST_REPEAT: 1
        UNIT_TEST_TIMEOUT: 30
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "yes"
        FUNCTIONAL_TEST_SCOPE: mr
        FUNCTIONAL_TEST_REPEAT: 5
        FUNCTIONAL_TEST_TIME_LIMIT: 2700
        CLUSTER_A100: ""
        CLUSTER_H100: ""
        PUBLISH: "no"

    # Default MRs
    - if: $CI_MERGE_REQUEST_EVENT_TYPE == 'merged_result'
      variables:
        UNIT_TEST: "yes"
        UNIT_TEST_REPEAT: 1
        UNIT_TEST_TIMEOUT: 30
        INTEGRATION_TEST: "no"
        FUNCTIONAL_TEST: "no"
        PUBLISH: "no"

    - when: never

  auto_cancel:
    on_new_commit: interruptible

stages:
  - build
  - test
  - integration_tests
  - functional_tests
  - publish

default:
  interruptible: true
  retry:
    max: 2
    when: runner_system_failure

variables:
  UNIT_TEST:
    value: "yes"
    options:
      - "yes"
      - "no"
    description: To run the funtional test suite
  UNIT_TEST_REPEAT:
    value: "1"
    description: "Number of repetitions"
  UNIT_TEST_TIMEOUT:
    value: "30"
    description: Timeout (minutes) for Unit tests (all repeats)
  INTEGRATION_TEST:
    value: "yes"
    options:
      - "yes"
      - "no"
    description: To run the integration test suite
  INTEGRATION_TEST_SCOPE:
    value: "mr"
    options:
      - "mr"
      - "nightly"
      - "weekly"
      - "pre-release"
      - "release"
    description: "Testsuite to run (only for INTEGRATION_TEST=yes)"
  INTEGRATION_TEST_TIME_LIMIT:
    value: "900"
    description: "Timeout in seconds per test"
  INTEGRATION_TEST_CASES:
    value: "all"
    description: "Comma-separated list of test_cases to run. Use 'all' to run the full suite."
  FUNCTIONAL_TEST:
    value: "yes"
    options:
      - "yes"
      - "no"
    description: To run the funtional test suite
  FUNCTIONAL_TEST_SCOPE:
    value: "mr"
    options:
      - "mr"
      - "nightly"
      - "weekly"
      - "pre-release"
      - "release"
    description: "Testsuite to run (only for FUNCTIONAL_TEST=yes)"
  FUNCTIONAL_TEST_REPEAT:
    value: "5"
    description: "Number of repetitions per test"
  FUNCTIONAL_TEST_TIME_LIMIT:
    value: "2700"
    description: "Timeout in seconds per test"
  FUNCTIONAL_TEST_CASES:
    value: "all"
    description: "Comma-separated list of test_cases to run. Use 'all' to run the full suite."
  FUNCTIONAL_TEST_NAME:
    description: "Name of functional test run (only for pre-release and release)"
    value: "$$CI_COMMIT_SHA"
  FUNCTIONAL_TEST_RECORD_CHECKPOINTS:
    value: "no"
    description: "Record golden checkpoints"
    options:
      - "yes"
      - "no"
  CLUSTER_A100:
    value: "dgxa100_dracooci"
    options:
      - "dgxa100_dracooci"
      - "dgxa100_dracooci-ord"
    description: "Cluster for A100 workloads"
  CLUSTER_H100:
    value: "dgxh100_coreweave"
    options:
      - "dgxh100_coreweave"
      - "dgxh100_eos"
    description: "Cluster for H100 workloads"
  PUBLISH:
    value: "no"
    options:
      - "yes"
      - "no"
    description: Build and publish a wheel to PyPi
  PUBLISH_COMMIT:
    value: "$$CI_COMMIT_SHA"
    description: Which commit to publish
  PUBLISH_VERSION_BUMP_BRANCH:
    value: "$$CI_COMMIT_BRANCH"
    description: Which branch to target for version bump
  PUBLISH_SCOPE:
    value: "code-freeze"
    options:
      - "code-freeze"
      - "release"
      - "review-reminder"
    description: Type of publish (freeze or final release)

  # CI wide variables
  CI_MCORE_LTS_IMAGE: ${GITLAB_ENDPOINT}:5005/adlr/megatron-lm/mcore_ci_lts
  CI_MCORE_DEV_IMAGE: ${GITLAB_ENDPOINT}:5005/adlr/megatron-lm/mcore_ci_dev
  CI_NEMO_IMAGE: ${GITLAB_ENDPOINT}:5005/adlr/megatron-lm/nemo_ci
  UTILITY_IMAGE: ${GITLAB_ENDPOINT}:5005/adlr/megatron-lm/mcore_utility
  TE_GIT_REF: ""

include:
  - .gitlab/stages/00.pre.yml
  - .gitlab/stages/01.build.yml
  - .gitlab/stages/02.test.yml
  - .gitlab/stages/03.integration-tests.yml
  - .gitlab/stages/04.functional-tests.yml
  - .gitlab/stages/05.publish.yml