fix: expose runner source fetch progress

deploy/runtime/boot/agentrun-boot-report.mjs

@@ -0,0 +1,29 @@
+const managerUrl = process.env.AGENTRUN_MGR_URL;
+const runnerJobId = process.env.AGENTRUN_RUNNER_JOB_ID;
+const phase = process.argv[2];
+const attempt = Number(process.argv[3]);
+if (!managerUrl || !runnerJobId || !phase || !Number.isSafeInteger(attempt) || attempt < 1) process.exit(2);
+
+const body = {
+  phase,
+  attempt,
+  runId: process.env.AGENTRUN_RUN_ID,
+  commandId: process.env.AGENTRUN_COMMAND_ID,
+  attemptId: process.env.AGENTRUN_ATTEMPT_ID,
+  runnerId: process.env.AGENTRUN_RUNNER_ID,
+};
+if (process.argv[4]) body.code = process.argv[4];
+if (process.argv[5]) body.summary = process.argv[5];
+const headers = { "content-type": "application/json" };
+if (process.env.AGENTRUN_API_KEY) headers.authorization = `Bearer ${process.env.AGENTRUN_API_KEY}`;
+const response = await fetch(new URL(`/api/v1/runner-jobs/${encodeURIComponent(runnerJobId)}/boot-observations`, managerUrl), {
+  method: "POST",
+  headers,
+  body: JSON.stringify(body),
+});
+const envelope = await response.json();
+if (!response.ok || envelope?.ok !== true) {
+  process.stderr.write(`${JSON.stringify({ ok: false, event: "agentrun-boot-report", status: response.status, failureKind: envelope?.failureKind ?? "infra-failed", message: envelope?.message ?? "manager boot observation failed" })}\n`);
+  process.exit(1);
+}
+process.stdout.write(`${JSON.stringify(envelope.data)}\n`);

deploy/runtime/boot/agentrun-runner-boot.sh

@@ -0,0 +1,65 @@
+#!/bin/sh
+set -eu
+
+entrypoint="${1:-src/runner/main.ts}"
+repo_url="${AGENTRUN_BOOT_REPO_URL:-http://git-mirror-http.devops-infra.svc.cluster.local/pikasTech/agentrun.git}"
+commit="${AGENTRUN_BOOT_COMMIT:-${AGENTRUN_SOURCE_COMMIT:-}}"
+app_root="${AGENTRUN_APP_ROOT:-/workspace/agentrun}"
+reporter=/opt/agentrun/deploy/runtime/boot/agentrun-boot-report.mjs
+
+if [ -z "$commit" ] || ! printf '%s' "$commit" | grep -Eq '^[0-9a-f]{40}$'; then
+  printf '{"ok":false,"event":"agentrun-boot","failureKind":"source-commit-invalid","message":"AGENTRUN_BOOT_COMMIT must be a full git SHA"}\n' >&2
+  exit 64
+fi
+
+mkdir -p "$(dirname "$app_root")"
+rm -rf "$app_root"
+mkdir -p "$app_root"
+cd "$app_root"
+git init -q
+git remote add origin "$repo_url"
+
+attempt=1
+fetch_log=/tmp/agentrun-boot-fetch.log
+while :; do
+  start_response=$(node "$reporter" source-fetch-started "$attempt" 2>/tmp/agentrun-boot-report.log || true)
+  timeout_ms=$(printf '%s' "$start_response" | node -e 'let s="";process.stdin.on("data",d=>s+=d).on("end",()=>{try{const n=JSON.parse(s).attemptTimeoutMs;process.stdout.write(Number.isFinite(n)?String(n):"")}catch{}})')
+  timeout_seconds=""
+  if [ -n "$timeout_ms" ]; then timeout_seconds=$(( (timeout_ms + 999) / 1000 )); fi
+
+  if { [ -n "$timeout_seconds" ] && timeout "${timeout_seconds}s" git fetch --depth=1 origin "$commit" >"$fetch_log" 2>&1; } || { [ -z "$timeout_seconds" ] && git fetch --depth=1 origin "$commit" >"$fetch_log" 2>&1; }; then
+    node "$reporter" source-fetch-succeeded "$attempt" >/tmp/agentrun-boot-report.log 2>&1 || true
+    break
+  fi
+
+  message=$(tail -n 20 "$fetch_log" | tr '\n' ' ' | sed 's/[[:space:]]\+/ /g')
+  failure_kind=git-mirror-fetch-failed
+  if printf '%s' "$message" | grep -Eiq 'not our ref|unadvertised|Server does not allow request'; then
+    failure_kind=git-mirror-exact-commit-unavailable
+  elif printf '%s' "$message" | grep -Eiq 'Could not resolve host|Connection timed out|Failed to connect|timed out'; then
+    failure_kind=git-mirror-network-failed
+  elif printf '%s' "$message" | grep -Eiq 'Authentication failed|Permission denied|repository.*not found'; then
+    failure_kind=git-mirror-auth-failed
+  fi
+  failure_response=$(node "$reporter" source-fetch-failed "$attempt" "$failure_kind" 2>/tmp/agentrun-boot-report.log || true)
+  action=$(printf '%s' "$failure_response" | node -e 'let s="";process.stdin.on("data",d=>s+=d).on("end",()=>{try{process.stdout.write(JSON.parse(s).action||"")}catch{}})')
+  if [ "$action" != "retry" ]; then
+    printf '{"ok":false,"event":"agentrun-boot","failureKind":"%s","repoUrl":"%s","commit":"%s","message":%s}\n' "$failure_kind" "$repo_url" "$commit" "$(node -e 'console.log(JSON.stringify(process.argv[1] || ""))' "$message")" >&2
+    exit 65
+  fi
+  backoff_ms=$(printf '%s' "$failure_response" | node -e 'let s="";process.stdin.on("data",d=>s+=d).on("end",()=>{try{const n=JSON.parse(s).backoffMs;process.stdout.write(Number.isFinite(n)?String(n):"0")}catch{process.stdout.write("0")}})')
+  sleep_seconds=$(node -e 'process.stdout.write(String(Math.max(0, Number(process.argv[1]) || 0) / 1000))' "$backoff_ms")
+  sleep "$sleep_seconds"
+  attempt=$((attempt + 1))
+done
+
+git checkout -q --detach "$commit"
+actual=$(git rev-parse HEAD)
+if [ "$actual" != "$commit" ]; then
+  printf '{"ok":false,"event":"agentrun-boot","failureKind":"source-commit-mismatch","expected":"%s","actual":"%s"}\n' "$commit" "$actual" >&2
+  exit 66
+fi
+
+ln -sfn /opt/agentrun/node_modules "$app_root/node_modules"
+printf '{"ok":true,"event":"agentrun-boot","repoUrl":"%s","commit":"%s","entrypoint":"%s","nodeModules":"/opt/agentrun/node_modules"}\n' "$repo_url" "$commit" "$entrypoint"
+exec bun "$entrypoint"

deploy/runtime/boot/agentrun-runner.sh

@@ -1,3 +1,3 @@
 #!/bin/sh
 set -eu
-exec /opt/agentrun/deploy/runtime/boot/agentrun-boot.sh src/runner/main.ts
+exec /opt/agentrun/deploy/runtime/boot/agentrun-runner-boot.sh src/runner/main.ts