[fix](ci) cancel superseded buildall revisions

.github/workflows/comment-to-trigger-teamcity.yml

@@ -259,6 +259,17 @@ jobs:
           exit 1
         fi
 
+    - name: "Cancel superseded buildall builds"
+      if: ${{ fromJSON(steps.parse.outputs.comment_trigger) && steps.parse.outputs.COMMENT_TRIGGER_TYPE == 'buildall' }}
+      env:
+        PULL_REQUEST_NUM: ${{ steps.parse.outputs.PULL_REQUEST_NUM }}
+        COMMIT_ID_FROM_TRIGGER: ${{ steps.parse.outputs.COMMIT_ID_FROM_TRIGGER }}
+      run: |
+        source ./regression-test/pipeline/common/teamcity-utils.sh
+        cancel_superseded_buildall_builds \
+          "${PULL_REQUEST_NUM}" \
+          "${COMMIT_ID_FROM_TRIGGER}"
+
     - name: "skip check coverage"
       if: ${{ fromJSON(steps.parse.outputs.comment_skip_coverage) }}
       run: |

regression-test/pipeline/common/teamcity-utils.sh

@@ -101,6 +101,146 @@ get_commit_id_of_build() {
     fi
 }
 
+is_buildall_pipeline() {
+    local build_type_id="$1"
+    local pipeline_id
+
+    # check_coverage_fe is triggered by the buildall downstream chain but does
+    # not have a standalone PR comment command, so it is not in the map above.
+    if [[ "${build_type_id}" == "Doris_Coverage_CheckCoverageFe" ]]; then
+        return 0
+    fi
+
+    for pipeline_id in "${comment_to_pipeline[@]}"; do
+        if [[ "${build_type_id}" == "${pipeline_id}" ]]; then
+            return 0
+        fi
+    done
+    return 1
+}
+
+cancel_superseded_buildall_builds() {
+    # Before a new run buildall fan-out starts, cancel every queued/running
+    # buildall pipeline for the same PR that is still bound to an older head.
+    local pull_request_num="${PULL_REQUEST_NUM:-$1}"
+    local current_commit_id="${COMMIT_ID_FROM_TRIGGER:-$2}"
+    if [[ ! "${pull_request_num}" =~ ^[0-9]+$ ||
+        ! "${current_commit_id}" =~ ^[0-9a-f]{40}$ ]]; then
+        echo "Usage: cancel_superseded_buildall_builds PULL_REQUEST_NUM COMMIT_ID_FROM_TRIGGER"
+        return 1
+    fi
+
+    local teamcity_url="http://43.132.222.7:8111"
+    local branch_name="pull/${pull_request_num}"
+    local fields='build(id,state,buildTypeId,branchName,revisions(revision(version)))'
+    local queued_json
+    local running_json
+    local active_json
+
+    if ! queued_json=$(
+        curl --fail --silent --show-error --retry 4 \
+            -u OneMoreChance:OneMoreChance \
+            -H "Accept: application/json" \
+            --get \
+            --data-urlencode "locator=count:1000" \
+            --data-urlencode "fields=${fields}" \
+            "${teamcity_url}/app/rest/buildQueue"
+    ); then
+        echo "ERROR: failed to list queued TeamCity builds before run buildall"
+        return 1
+    fi
+
+    if ! running_json=$(
+        curl --fail --silent --show-error --retry 4 \
+            -u OneMoreChance:OneMoreChance \
+            -H "Accept: application/json" \
+            --get \
+            --data-urlencode "locator=branch:(name:${branch_name}),state:running,defaultFilter:false,count:1000" \
+            --data-urlencode "fields=${fields}" \
+            "${teamcity_url}/app/rest/builds"
+    ); then
+        echo "ERROR: failed to list running TeamCity builds for PR ${pull_request_num}"
+        return 1
+    fi
+
+    if ! active_json=$(jq -s \
+        '[.[0].build[]?, .[1].build[]?] | unique_by(.id)' \
+        <<<"${queued_json}"$'\n'"${running_json}"); then
+        echo "ERROR: failed to parse active TeamCity builds for PR ${pull_request_num}"
+        return 1
+    fi
+
+    local build_json
+    local build_id
+    local build_state
+    local build_type_id
+    local build_branch
+    local build_commit_id
+    local cancel_endpoint
+    local cancel_payload
+    local http_code
+    local canceled_count=0
+
+    while IFS= read -r build_json; do
+        build_id=$(jq -er '.id' <<<"${build_json}") || return 1
+        build_state=$(jq -r '.state // empty' <<<"${build_json}")
+        build_type_id=$(jq -r '.buildTypeId // empty' <<<"${build_json}")
+        build_branch=$(jq -r '.branchName // empty' <<<"${build_json}")
+        build_commit_id=$(jq -r '[.revisions.revision[]?.version][0] // empty' <<<"${build_json}")
+
+        [[ "${build_branch}" == "${branch_name}" ]] || continue
+        is_buildall_pipeline "${build_type_id}" || continue
+        if [[ ! "${build_commit_id}" =~ ^[0-9a-f]{40}$ ]]; then
+            echo "WARNING: keep build ${build_id}; no immutable revision binding was returned"
+            continue
+        fi
+        [[ "${build_commit_id}" != "${current_commit_id}" ]] || continue
+
+        if [[ "${TEAMCITY_CANCEL_DRY_RUN:-false}" == "true" ]]; then
+            echo "INFO: would cancel superseded ${build_state} build ${build_id} (${build_type_id}) revision ${build_commit_id}"
+            continue
+        fi
+
+        if [[ "${build_state}" == "queued" ]]; then
+            cancel_endpoint="buildQueue/id:${build_id}"
+        elif [[ "${build_state}" == "running" ]]; then
+            cancel_endpoint="builds/id:${build_id}"
+        else
+            echo "WARNING: keep build ${build_id}; unsupported active state ${build_state}"
+            continue
+        fi
+
+        cancel_payload=$(jq -cn \
+            --arg comment "Superseded by PR #${pull_request_num} revision ${current_commit_id}; old revision ${build_commit_id}" \
+            '{comment: $comment, readdIntoQueue: false}')
+        http_code=$(
+            curl --silent --show-error --retry 4 \
+                -o /dev/null -w '%{http_code}' \
+                -u OneMoreChance:OneMoreChance \
+                -H "Content-Type: application/json" \
+                -H "Accept: application/json" \
+                --data-binary "${cancel_payload}" \
+                "${teamcity_url}/app/rest/${cancel_endpoint}"
+        ) || return 1
+
+        case "${http_code}" in
+            2??)
+                echo "INFO: canceled superseded ${build_state} build ${build_id} (${build_type_id}) revision ${build_commit_id}"
+                ((canceled_count += 1))
+                ;;
+            404|409)
+                echo "INFO: build ${build_id} finished while cancellation was being requested"
+                ;;
+            *)
+                echo "ERROR: failed to cancel build ${build_id}; TeamCity returned HTTP ${http_code}"
+                return 1
+                ;;
+        esac
+    done < <(jq -c '.[]' <<<"${active_json}")
+
+    echo "INFO: superseded buildall cancellation completed for PR ${pull_request_num}; canceled=${canceled_count}"
+}
+
 get_running_build_of_pr() {
     # "获取pr在某条流水线上正在跑的build"
     local PULL_REQUEST_NUM="${PULL_REQUEST_NUM:-$1}"