fix(ci): unblock TypeScript E2E fan-out

.github/scripts/select-shared-release-artifact.sh

@@ -10,6 +10,9 @@ usage() {
 [[ $# -ge 1 && $# -le 2 ]] || usage
 output_file="$1"
 max_wait_seconds="${2:-360}"
+producer_max_wait_seconds="${PRODUCER_MAX_WAIT_SECONDS:-420}"
+producer_artifact_grace_seconds="${PRODUCER_ARTIFACT_GRACE_SECONDS:-30}"
+poll_seconds="${SELECTOR_POLL_SECONDS:-15}"
 
 : "${GH_TOKEN:?GH_TOKEN must contain a short-lived Actions token}"
 : "${GITHUB_REPOSITORY:?GITHUB_REPOSITORY must be set}"
@@ -21,10 +24,18 @@ max_wait_seconds="${2:-360}"
 [[ "$GITHUB_SHA" =~ ^[0-9a-f]{40}$ ]] || usage
 [[ "$GITHUB_PR_HEAD_SHA" =~ ^[0-9a-f]{40}$ ]] || usage
 [[ "$max_wait_seconds" =~ ^[0-9]+$ ]] || usage
+[[ "$producer_max_wait_seconds" =~ ^[0-9]+$ ]] || usage
+[[ "$producer_artifact_grace_seconds" =~ ^[0-9]+$ ]] || usage
+[[ "$poll_seconds" =~ ^[0-9]+$ ]] || usage
 
 artifact_name="node-subtensor-release-$GITHUB_SHA"
 workflow_path=.github/workflows/runtime-checks.yml
 started=$(date -u +%s)
+producer_completed_seen=
+producer_run_id=
+producer_job_status=
+producer_job_conclusion=
+last_producer_state=
 
 write_miss() {
   {
@@ -33,6 +44,28 @@ write_miss() {
   } >> "$output_file"
 }
 
+find_latest_producer_run() {
+  local producer_runs
+  producer_runs=$(gh api \
+    -H 'Accept: application/vnd.github+json' \
+    "repos/$GITHUB_REPOSITORY/actions/runs?event=pull_request&head_sha=$GITHUB_PR_HEAD_SHA&per_page=100" \
+    2>/dev/null || true)
+  jq -er \
+    --arg repository_id "$GITHUB_REPOSITORY_ID" \
+    --arg sha "$GITHUB_PR_HEAD_SHA" \
+    --arg workflow_path "$workflow_path" '
+      [.workflow_runs[]
+        | select(.head_sha == $sha)
+        | select((.head_repository.id | tostring) == $repository_id)
+        | select(.path == $workflow_path)
+        | select(.event == "pull_request")
+        | select((.id | type) == "number" and .id > 0)]
+      | sort_by(.created_at)
+      | reverse
+      | (.[0].id // empty)
+    ' <<< "$producer_runs" 2>/dev/null || true
+}
+
 while true; do
   artifacts=$(gh api \
     -H 'Accept: application/vnd.github+json' \
@@ -94,13 +127,76 @@ while true; do
     )
   fi
 
+  if [[ -z "$producer_run_id" ]]; then
+    producer_run_id=$(find_latest_producer_run)
+  fi
+
+  # Once discovered, latch the exact release-producing job. Transient API
+  # failures must not make us forget a producer and cross back to the shorter
+  # discovery timeout.
+  if [[ -n "$producer_run_id" ]]; then
+    producer_jobs=$(gh api \
+      -H 'Accept: application/vnd.github+json' \
+      "repos/$GITHUB_REPOSITORY/actions/runs/$producer_run_id/jobs?filter=latest&per_page=100" \
+      2>/dev/null || true)
+    producer_job=$(
+      jq -cer '
+        [.jobs[] | select(.name == "build release node")]
+        | sort_by(.started_at // .created_at)
+        | reverse
+        | .[0]
+      ' <<< "$producer_jobs" 2>/dev/null || true
+    )
+    current_status=$(jq -r '.status // empty' <<< "$producer_job" 2>/dev/null || true)
+    if [[ -n "$current_status" ]]; then
+      producer_job_status=$current_status
+      producer_job_conclusion=$(jq -r '.conclusion // empty' <<< "$producer_job" 2>/dev/null || true)
+    fi
+  fi
+
+  producer_state=not-found
+  if [[ -n "$producer_run_id" ]]; then
+    producer_state=${producer_job_status:-job-pending}
+    [[ -z "$producer_job_conclusion" ]] || producer_state="$producer_state/$producer_job_conclusion"
+  fi
+  if [[ "$producer_state" != "$last_producer_state" ]]; then
+    echo "Runtime Checks release producer: $producer_state"
+    last_producer_state=$producer_state
+  fi
+
   elapsed=$(($(date -u +%s) - started))
-  if (( elapsed >= max_wait_seconds )); then
+  if [[ "$producer_job_status" == completed && "$producer_job_conclusion" == success ]]; then
+    [[ -n "$producer_completed_seen" ]] || producer_completed_seen=$(date -u +%s)
+    completed_elapsed=$(($(date -u +%s) - producer_completed_seen))
+    if (( completed_elapsed >= producer_artifact_grace_seconds )); then
+      write_miss
+      echo "Runtime Checks release producer completed without a verified artifact after ${completed_elapsed}s of API propagation grace; using the local build fallback."
+      exit 0
+    fi
+  elif [[ "$producer_job_status" == completed ]]; then
+    replacement_run_id=$(find_latest_producer_run)
+    if [[ -n "$replacement_run_id" && "$replacement_run_id" != "$producer_run_id" ]]; then
+      echo "Runtime Checks replaced producer run $producer_run_id with $replacement_run_id."
+      producer_run_id=$replacement_run_id
+      producer_job_status=
+      producer_job_conclusion=
+      producer_completed_seen=
+      continue
+    fi
+    write_miss
+    echo "Runtime Checks release producer completed (${producer_job_conclusion:-unknown}); using the local build fallback."
+    exit 0
+  elif [[ -n "$producer_run_id" ]]; then
+    producer_completed_seen=
+    if (( elapsed >= producer_max_wait_seconds )); then
+      write_miss
+      echo "Runtime Checks release producer was still ${producer_job_status:-pending} after ${elapsed}s; using the local build fallback."
+      exit 0
+    fi
+  elif (( elapsed >= max_wait_seconds )); then
     write_miss
-    echo "No exact-commit TypeScript release artifact appeared after ${elapsed}s; using the local build fallback."
+    echo "No exact-commit Runtime Checks producer appeared after ${elapsed}s; using the local build fallback."
     exit 0
   fi
-  remaining=$((max_wait_seconds - elapsed))
-  (( remaining > 10 )) || sleep "$remaining"
-  (( remaining <= 10 )) || sleep 10
+  sleep "$poll_seconds"
 done

.github/scripts/split-typescript-e2e-plan.sh

@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+if [[ $# -ne 1 ]]; then
+  echo "usage: $0 OUTPUT_FILE" >&2
+  exit 2
+fi
+
+output_file=$1
+: "${STATE_MATRIX:?STATE_MATRIX must contain the trusted state matrix}"
+: "${BUILD_MATRIX:?BUILD_MATRIX must contain the trusted build matrix}"
+
+jq -e '
+  type == "object" and
+  (.include | type == "array") and
+  all(.include[]; type == "object" and (.binary == "fast" or .binary == "release"))
+' <<< "$STATE_MATRIX" >/dev/null
+jq -e '
+  type == "object" and
+  (.include | type == "array") and
+  all(.include[]; type == "object" and (.variant == "fast" or .variant == "release"))
+' <<< "$BUILD_MATRIX" >/dev/null
+
+fast_state_matrix=$(jq -c '{include: [.include[] | select(.binary == "fast")]}' <<< "$STATE_MATRIX")
+release_state_matrix=$(jq -c '{include: [.include[] | select(.binary == "release")]}' <<< "$STATE_MATRIX")
+fast_state_count=$(jq '.include | length' <<< "$fast_state_matrix")
+release_state_count=$(jq '.include | length' <<< "$release_state_matrix")
+fast_build=$(jq -r 'any(.include[]; .variant == "fast")' <<< "$BUILD_MATRIX")
+release_build=$(jq -r 'any(.include[]; .variant == "release")' <<< "$BUILD_MATRIX")
+
+if (( fast_state_count > 0 )) && [[ "$fast_build" != true ]]; then
+  echo "fast state jobs were planned without a fast binary build" >&2
+  exit 1
+fi
+if (( release_state_count > 0 )) && [[ "$release_build" != true ]]; then
+  echo "release state jobs were planned without a release binary build" >&2
+  exit 1
+fi
+
+{
+  echo "fast_state_count=$fast_state_count"
+  echo "fast_state_matrix=$fast_state_matrix"
+  echo "release_state_count=$release_state_count"
+  echo "release_state_matrix=$release_state_matrix"
+  echo "fast_build=$fast_build"
+  echo "release_build=$release_build"
+} >> "$output_file"

.github/scripts/test-classify-typescript-e2e-changes.sh

@@ -264,9 +264,24 @@ fi
 grep -Fq 'ref: ${{ github.event_name == '\''pull_request'\'' && github.event.pull_request.base.sha || github.sha }}' "$workflow"
 grep -Fq '.trusted-e2e-filter/ts-tests/scripts/e2e-shard-plan.mjs' "$workflow"
 grep -Fq '.proposed-e2e-plan/ts-tests/e2e-shards.json' "$workflow"
+grep -Fq '.trusted-e2e-filter/.github/scripts/split-typescript-e2e-plan.sh' "$workflow"
+grep -Fq '.proposed-e2e-plan/.github/scripts/split-typescript-e2e-plan.sh' "$workflow"
 grep -Fq 'matrix: ${{ fromJSON(needs.changes.outputs.shield_matrix) }}' "$workflow"
+grep -Fq 'matrix: ${{ fromJSON(needs.changes.outputs.fast_state_matrix) }}' "$workflow"
+grep -Fq 'matrix: ${{ fromJSON(needs.changes.outputs.release_state_matrix) }}' "$workflow"
+grep -Fq 'needs: [trusted-pr, changes, build-fast]' "$workflow"
+grep -Fq 'needs: [trusted-pr, changes, build-release]' "$workflow"
+grep -Fq 'needs: [trusted-pr, changes, select-release-artifact]' "$workflow"
+grep -Fq 'needs: [trusted-pr, changes, build-fast, run-fast-e2e-tests, sharding-audit-fast]' "$workflow"
+grep -Fq 'needs: [trusted-pr, changes, build-release, run-shield-tests, sharding-audit-release]' "$workflow"
+grep -Fq 'fast_build: ${{ steps.lanes.outputs.fast_build }}' "$workflow"
+grep -Fq 'release_build: ${{ steps.lanes.outputs.release_build }}' "$workflow"
 grep -Fq 'name: Audit canonical unsharded ${{ matrix.test }}' "$workflow"
 grep -Fq 'EVM_SELECTED: ${{ needs.changes.outputs.evm }}' "$workflow"
 grep -Fq 'SHIELD_SELECTED: ${{ needs.changes.outputs.shield }}' "$workflow"
+if grep -Eq '^  build:$' "$workflow"; then
+  echo "TypeScript E2E consumers must not depend on a combined build matrix." >&2
+  exit 1
+fi
 
 echo "typescript E2E change classifier tests passed"