20260713 unify installer scripts and RTR registration

2026-07-13 12:46:40 +08:00
parent b66b425512
commit 9f4f4cf069
20 changed files with 346 additions and 100 deletions
--- a/deploy/docker-installer/scripts/register_rtr_monitor.sh
+++ b/deploy/docker-installer/scripts/register_rtr_monitor.sh
@@ -0,0 +1,149 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/common.sh"
+
+REPORT_DIR=""
+DRY_RUN=0
+CANDIDATE_ENV=""
+CONFIG_FILE=""
+ORIGINAL_ENV_FILE=""
+ENV_BACKUP=""
+REGISTRATION_APPLIED=0
+
+usage() {
+  cat <<'EOF'
+Usage: ./scripts/register_rtr_monitor.sh --report-dir <absolute-host-path> [--dry-run]
+
+Registers one local RTR report directory for the artifact metrics sidecar.
+The directory must be on the same host as this installer and may contain
+rtr-source-*.json, rtr-runtime-*.json, and rtr-clients-*.json report files.
+
+This command recreates only artifact-metrics, Prometheus, and Grafana. It does
+not restart or recreate the ours-rp-soak container.
+EOF
+}
+
+cleanup() {
+  [[ -n "$CANDIDATE_ENV" && -f "$CANDIDATE_ENV" ]] && rm -f "$CANDIDATE_ENV"
+  [[ -n "$CONFIG_FILE" && -f "$CONFIG_FILE" ]] && rm -f "$CONFIG_FILE"
+  return 0
+}
+trap cleanup EXIT
+
+restore_failed_registration() {
+  if (( REGISTRATION_APPLIED == 0 )) || [[ -z "$ENV_BACKUP" || ! -f "$ENV_BACKUP" ]]; then
+    return 0
+  fi
+
+  warn "registration failed after .env replacement; restoring previous metrics configuration"
+  cp -p "$ENV_BACKUP" "$ORIGINAL_ENV_FILE"
+  ENV_FILE="$ORIGINAL_ENV_FILE"
+  load_env || warn "unable to reload restored .env"
+  compose_cmd --profile sidecar up -d --force-recreate artifact-metrics \
+    || warn "unable to recreate artifact-metrics with restored configuration"
+}
+
+fail_after_apply() {
+  restore_failed_registration
+  die "$*"
+}
+
+while (( $# > 0 )); do
+  case "$1" in
+    --report-dir)
+      [[ $# -ge 2 ]] || die "--report-dir requires a value"
+      REPORT_DIR="$2"
+      shift 2
+      ;;
+    --dry-run)
+      DRY_RUN=1
+      shift
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      die "unknown argument: $1"
+      ;;
+  esac
+done
+
+require_cmd docker
+require_cmd curl
+require_cmd grep
+require_cmd jq
+validate_rtr_registration_dir "$REPORT_DIR"
+load_env
+assert_arch_compatibility
+
+ORIGINAL_ENV_FILE="$ENV_FILE"
+REPORT_FILE_COUNT="$(rtr_report_file_count "$REPORT_DIR")"
+LATEST_REPORT_FILE="$(latest_rtr_report_file "$REPORT_DIR")"
+if (( REPORT_FILE_COUNT == 0 )); then
+  warn "no recognized RTR report JSON files are currently present in $REPORT_DIR"
+fi
+
+CANDIDATE_ENV="$(mktemp "$INSTALLER_ROOT/.env.rtr-register.XXXXXX")"
+cp -p "$ORIGINAL_ENV_FILE" "$CANDIDATE_ENV"
+replace_env_key_file "$CANDIDATE_ENV" "RTR_REPORT_DIR" "$REPORT_DIR"
+
+ENV_FILE="$CANDIDATE_ENV"
+load_env
+CONFIG_FILE="$(mktemp "$INSTALLER_ROOT/.rtr-compose-config.XXXXXX")"
+compose_cmd --profile sidecar --profile monitor config --format json > "$CONFIG_FILE"
+jq -e --arg source "$REPORT_DIR" --arg target "$RTR_REPORT_CONTAINER_DIR" '
+  .services["artifact-metrics"].volumes
+  | any(.[]; .type == "bind" and .source == $source and .target == $target and .read_only == true)
+' "$CONFIG_FILE" >/dev/null || die "candidate compose configuration is missing expected read-only RTR bind mount"
+
+if (( DRY_RUN == 1 )); then
+  printf 'dry_run=ok\n'
+  printf 'report_dir=%s\n' "$REPORT_DIR"
+  printf 'recognized_report_files=%s\n' "$REPORT_FILE_COUNT"
+  printf 'latest_report=%s\n' "${LATEST_REPORT_FILE:--}"
+  printf 'expected_mount_source=%s\n' "$REPORT_DIR"
+  printf 'expected_mount_target=%s\n' "$RTR_REPORT_CONTAINER_DIR"
+  exit 0
+fi
+
+ENV_FILE="$ORIGINAL_ENV_FILE"
+load_env
+soak_container_before="$(compose_cmd ps -q ours-rp-soak || true)"
+ENV_BACKUP="${ORIGINAL_ENV_FILE}.rtr-register-$(date -u +%Y%m%dT%H%M%SZ).bak"
+cp -p "$ORIGINAL_ENV_FILE" "$ENV_BACKUP"
+mv "$CANDIDATE_ENV" "$ORIGINAL_ENV_FILE"
+CANDIDATE_ENV=""
+REGISTRATION_APPLIED=1
+
+load_env
+compose_cmd --profile sidecar up -d --force-recreate artifact-metrics \
+  || fail_after_apply "unable to recreate artifact-metrics"
+compose_cmd --profile sidecar --profile monitor up -d prometheus grafana \
+  || fail_after_apply "unable to start Prometheus and Grafana"
+
+soak_container_after="$(compose_cmd ps -q ours-rp-soak || true)"
+if [[ -n "$soak_container_before" && "$soak_container_before" != "$soak_container_after" ]]; then
+  fail_after_apply "ours-rp-soak container changed during RTR registration; investigate before continuing"
+fi
+
+wait_for_endpoint "http://127.0.0.1:${METRICS_PORT:-9556}/metrics" 60 \
+  || fail_after_apply "artifact-metrics did not become ready"
+wait_for_endpoint "http://127.0.0.1:${PROMETHEUS_PORT:-9090}/-/ready" 60 \
+  || fail_after_apply "Prometheus did not become ready"
+wait_for_endpoint "http://127.0.0.1:${GRAFANA_PORT:-3000}/api/health" 60 \
+  || fail_after_apply "Grafana did not become ready"
+
+if (( REPORT_FILE_COUNT > 0 )); then
+  curl -fsS "http://127.0.0.1:${METRICS_PORT:-9556}/metrics" | grep -q '^ours_rp_rtr_' \
+    || fail_after_apply "artifact-metrics exposes no RTR metrics after registration"
+fi
+
+printf 'registration=ok\n'
+printf 'report_dir=%s\n' "$REPORT_DIR"
+printf 'recognized_report_files=%s\n' "$REPORT_FILE_COUNT"
+printf 'latest_report=%s\n' "${LATEST_REPORT_FILE:--}"
+printf 'env_backup=%s\n' "$ENV_BACKUP"
+printf 'ours_rp_soak_container_unchanged=%s\n' "${soak_container_before:-not-running}"