407 lines
13 KiB
Bash

#!/usr/bin/env bash
set -euo pipefail
INSTALLER_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
ENV_FILE="${ENV_FILE:-$INSTALLER_ROOT/.env}"
ENV_EXAMPLE="$INSTALLER_ROOT/.env.example"
COMPOSE_FILE="$INSTALLER_ROOT/compose/docker-compose.yml"
MANIFEST_FILE="${MANIFEST_FILE:-$INSTALLER_ROOT/PACKAGE-MANIFEST.env}"
log() {
printf '[ours-rp-installer] %s\n' "$*"
}
warn() {
printf '[ours-rp-installer][WARN] %s\n' "$*" >&2
}
die() {
printf '[ours-rp-installer][ERROR] %s\n' "$*" >&2
exit 1
}
require_cmd() {
command -v "$1" >/dev/null 2>&1 || die "missing command: $1"
}
normalize_arch() {
case "$1" in
amd64|x86_64|linux/amd64)
printf 'amd64\n'
;;
arm64|aarch64|linux/arm64)
printf 'arm64\n'
;;
*)
return 1
;;
esac
}
platform_for_arch() {
printf 'linux/%s\n' "$1"
}
env_flag_enabled() {
case "${1:-0}" in
1|true|TRUE|yes|YES|on|ON)
return 0
;;
*)
return 1
;;
esac
}
detect_host_arch() {
local raw_arch
raw_arch="$(uname -m)"
normalize_arch "$raw_arch" || die "unsupported host architecture: $raw_arch"
}
load_manifest() {
MANIFEST_PACKAGE_ARCH_RAW=""
MANIFEST_PACKAGE_PLATFORM_RAW=""
if [[ -f "$MANIFEST_FILE" ]]; then
set -a
# shellcheck disable=SC1090
source "$MANIFEST_FILE"
set +a
MANIFEST_PACKAGE_ARCH_RAW="${PACKAGE_ARCH:-${package_arch:-}}"
MANIFEST_PACKAGE_PLATFORM_RAW="${PACKAGE_PLATFORM:-${package_platform:-}}"
fi
}
effective_package_arch() {
if [[ -n "${PACKAGE_ARCH:-}" ]]; then
normalize_arch "$PACKAGE_ARCH" || die "unsupported PACKAGE_ARCH=${PACKAGE_ARCH}"
return 0
fi
if [[ -n "${PACKAGE_PLATFORM:-}" ]]; then
normalize_arch "$PACKAGE_PLATFORM" || die "unsupported PACKAGE_PLATFORM=${PACKAGE_PLATFORM}"
return 0
fi
if [[ -n "${RPKI_PLATFORM:-}" ]]; then
normalize_arch "$RPKI_PLATFORM" || die "unsupported RPKI_PLATFORM=${RPKI_PLATFORM}"
return 0
fi
if [[ -n "${MONITOR_PLATFORM:-}" ]]; then
normalize_arch "$MONITOR_PLATFORM" || die "unsupported MONITOR_PLATFORM=${MONITOR_PLATFORM}"
return 0
fi
die "unable to determine package architecture from manifest or env"
}
arch_mode() {
local host_arch package_arch
host_arch="$(detect_host_arch)"
package_arch="$(effective_package_arch)"
if [[ "$host_arch" == "$package_arch" ]]; then
printf 'native\n'
return 0
fi
if env_flag_enabled "${ALLOW_CROSS_ARCH:-0}"; then
printf 'cross-arch-enabled\n'
return 0
fi
printf 'mismatch-blocked\n'
}
assert_arch_compatibility() {
local host_arch package_arch package_platform mode
host_arch="$(detect_host_arch)"
package_arch="$(effective_package_arch)"
package_platform="${PACKAGE_PLATFORM:-${RPKI_PLATFORM:-$(platform_for_arch "$package_arch")}}"
mode="$(arch_mode)"
case "$mode" in
native)
return 0
;;
cross-arch-enabled)
log "cross-arch execution explicitly enabled: host_arch=$host_arch package_arch=$package_arch package_platform=$package_platform"
return 0
;;
mismatch-blocked)
die "package architecture mismatch: host_arch=$host_arch package_arch=$package_arch package_platform=$package_platform. Set ALLOW_CROSS_ARCH=1 in $ENV_FILE to allow explicit QEMU/binfmt emulation."
;;
*)
die "unknown arch compatibility mode: $mode"
;;
esac
}
load_env() {
load_manifest
if [[ ! -f "$ENV_FILE" ]]; then
[[ -f "$ENV_EXAMPLE" ]] || die "missing $ENV_EXAMPLE"
cp "$ENV_EXAMPLE" "$ENV_FILE"
log "created .env from .env.example"
fi
set -a
# shellcheck disable=SC1090
source "$ENV_FILE"
set +a
if [[ -n "${MANIFEST_PACKAGE_ARCH_RAW:-}" ]]; then
manifest_package_arch="$(normalize_arch "$MANIFEST_PACKAGE_ARCH_RAW")" || die "unsupported manifest package arch: $MANIFEST_PACKAGE_ARCH_RAW"
if [[ -n "${PACKAGE_ARCH:-}" ]]; then
env_package_arch="$(normalize_arch "$PACKAGE_ARCH")" || die "unsupported env package arch: $PACKAGE_ARCH"
[[ "$env_package_arch" == "$manifest_package_arch" ]] || die "env PACKAGE_ARCH=$env_package_arch mismatches manifest PACKAGE_ARCH=$manifest_package_arch"
fi
PACKAGE_ARCH="$manifest_package_arch"
else
if [[ -z "${PACKAGE_ARCH:-}" ]]; then
PACKAGE_ARCH="$(effective_package_arch)"
else
PACKAGE_ARCH="$(normalize_arch "$PACKAGE_ARCH")"
fi
fi
if [[ -n "${MANIFEST_PACKAGE_PLATFORM_RAW:-}" ]]; then
manifest_package_platform="$MANIFEST_PACKAGE_PLATFORM_RAW"
if [[ -n "${PACKAGE_PLATFORM:-}" && "$PACKAGE_PLATFORM" != "$manifest_package_platform" ]]; then
die "env PACKAGE_PLATFORM=$PACKAGE_PLATFORM mismatches manifest PACKAGE_PLATFORM=$manifest_package_platform"
fi
PACKAGE_PLATFORM="$manifest_package_platform"
else
PACKAGE_PLATFORM="${PACKAGE_PLATFORM:-$(platform_for_arch "$PACKAGE_ARCH")}"
fi
HOST_DATA_DIR="${HOST_DATA_DIR:-/var/lib/ours-rp-${PACKAGE_ARCH}-installer}"
COMPOSE_PROJECT_NAME="${COMPOSE_PROJECT_NAME:-ours-rp-${PACKAGE_ARCH}-installer}"
RPKI_IMAGE="${RPKI_IMAGE:-ours-rp-runtime-${PACKAGE_ARCH}:dev}"
RPKI_PLATFORM="${RPKI_PLATFORM:-$PACKAGE_PLATFORM}"
MONITOR_PLATFORM="${MONITOR_PLATFORM:-$PACKAGE_PLATFORM}"
PROMETHEUS_IMAGE="${PROMETHEUS_IMAGE:-prom/prometheus:v2.55.1}"
GRAFANA_IMAGE="${GRAFANA_IMAGE:-grafana/grafana:11.3.1}"
METRICS_INSTANCE="${METRICS_INSTANCE:-${PACKAGE_ARCH}-installer}"
ALLOW_CROSS_ARCH="${ALLOW_CROSS_ARCH:-0}"
FIRST_RUN_WAIT_TIMEOUT_SECS="${FIRST_RUN_WAIT_TIMEOUT_SECS:-7200}"
}
compose_cmd() {
docker compose --env-file "$ENV_FILE" -f "$COMPOSE_FILE" -p "$COMPOSE_PROJECT_NAME" "$@"
}
create_data_dirs() {
load_env
mkdir -p \
"$HOST_DATA_DIR/state" \
"$HOST_DATA_DIR/runs" \
"$HOST_DATA_DIR/logs" \
"$HOST_DATA_DIR/tmp" \
"$HOST_DATA_DIR/prometheus" \
"$HOST_DATA_DIR/grafana"
chmod 755 "$HOST_DATA_DIR" "$HOST_DATA_DIR/state" "$HOST_DATA_DIR/runs" "$HOST_DATA_DIR/logs" "$HOST_DATA_DIR/tmp" || true
chmod 777 "$HOST_DATA_DIR/prometheus" "$HOST_DATA_DIR/grafana" || true
}
latest_run_dir() {
load_env
find "$HOST_DATA_DIR/runs" -maxdepth 1 -mindepth 1 -type d -name 'run_*' 2>/dev/null | sort | tail -1
}
latest_success_run_dir() {
load_env
find "$HOST_DATA_DIR/runs" -maxdepth 2 -type f -path '*/run-summary.json' 2>/dev/null \
| while read -r summary; do
if jq -e '.status == "success"' "$summary" >/dev/null 2>&1; then
dirname "$summary"
fi
done | sort | tail -1
}
has_success_run() {
[[ -n "$(latest_success_run_dir)" ]]
}
print_run_summary() {
local run_dir="$1"
local summary="$run_dir/run-summary.json"
local meta="$run_dir/run-meta.json"
local timing="$run_dir/stage-timing.json"
local process_time="$run_dir/process-time.txt"
local vrps_file="$run_dir/vrps.csv"
local vaps_file="$run_dir/vaps.csv"
local status="unknown"
local sync_mode="unknown"
local wall_ms="null"
local validation_ms="null"
local repo_sync_ms="null"
local max_rss_kb="null"
local publication_points="null"
local vrps="null"
local vaps="null"
local warnings="null"
[[ -f "$summary" ]] || {
warn "missing run-summary.json in $run_dir"
return 1
}
status="$(jq -r '.status // "unknown"' "$summary" 2>/dev/null || echo unknown)"
wall_ms="$(jq -r '.wallMs // .wall_ms // "null"' "$summary" 2>/dev/null || echo null)"
warnings="$(jq -r '.warningCount // .warnings // "null"' "$summary" 2>/dev/null || echo null)"
if [[ -f "$meta" ]]; then
sync_mode="$(jq -r '.sync_mode // .syncMode // "unknown"' "$meta" 2>/dev/null || echo unknown)"
status="$(jq -r --arg fallback "$status" '.status // $fallback' "$meta" 2>/dev/null || echo "$status")"
fi
if [[ -f "$timing" ]]; then
validation_ms="$(jq -r '.validation_ms // "null"' "$timing" 2>/dev/null || echo null)"
repo_sync_ms="$(jq -r '.repo_sync_ms_total // "null"' "$timing" 2>/dev/null || echo null)"
publication_points="$(jq -r '.publication_points // "null"' "$timing" 2>/dev/null || echo null)"
fi
if [[ -f "$process_time" ]]; then
max_rss_kb="$(awk -F': ' '/Maximum resident set size/ {print $2; found=1} END {if (!found) print "null"}' "$process_time")"
fi
if [[ -f "$vrps_file" ]]; then
vrps="$(( $(wc -l < "$vrps_file") > 0 ? $(wc -l < "$vrps_file") - 1 : 0 ))"
fi
if [[ -f "$vaps_file" ]]; then
vaps="$(( $(wc -l < "$vaps_file") > 0 ? $(wc -l < "$vaps_file") - 1 : 0 ))"
fi
jq -n \
--arg run "$(basename "$run_dir")" \
--arg status "$status" \
--arg syncMode "$sync_mode" \
--argjson wallMs "$wall_ms" \
--argjson validationMs "$validation_ms" \
--argjson repoSyncMs "$repo_sync_ms" \
--argjson maxRssKb "$max_rss_kb" \
--argjson vrps "$vrps" \
--argjson vaps "$vaps" \
--argjson publicationPoints "$publication_points" \
--argjson warnings "$warnings" \
'{run:$run,status:$status,syncMode:$syncMode,wallMs:$wallMs,validationMs:$validationMs,repoSyncMs:$repoSyncMs,maxRssKb:$maxRssKb,vrps:$vrps,vaps:$vaps,publicationPoints:$publicationPoints,warnings:$warnings}'
}
wait_for_new_success_run() {
local before_latest="$1"
local timeout_secs="$2"
local start_epoch now run_dir summary meta status meta_status
start_epoch="$(date +%s)"
while true; do
run_dir="$(latest_run_dir || true)"
if [[ -n "$run_dir" && "$run_dir" != "$before_latest" ]]; then
summary="$run_dir/run-summary.json"
meta="$run_dir/run-meta.json"
if [[ -f "$summary" ]]; then
status="$(jq -r '.status // "unknown"' "$summary" 2>/dev/null || echo unknown)"
if [[ "$status" == "success" ]]; then
meta_status="unknown"
if [[ -f "$meta" ]]; then
meta_status="$(jq -r '.status // "unknown"' "$meta" 2>/dev/null || echo unknown)"
fi
if [[ "$meta_status" == "success" ]]; then
print_run_summary "$run_dir" || true
return 0
fi
fi
if [[ "$status" == "failed" || "$status" == "error" ]]; then
print_run_summary "$run_dir" || true
die "run failed: $run_dir"
fi
fi
fi
now="$(date +%s)"
if (( now - start_epoch > timeout_secs )); then
die "timed out waiting for first successful run after ${timeout_secs}s"
fi
sleep 10
done
}
docker_compose_available() {
docker compose version >/dev/null 2>&1
}
install_docker_if_missing() {
if command -v docker >/dev/null 2>&1 && docker_compose_available && command -v jq >/dev/null 2>&1 && command -v rsync >/dev/null 2>&1 && command -v curl >/dev/null 2>&1; then
log "docker and docker compose are already installed"
return 0
fi
if [[ "${SKIP_DEP_INSTALL:-0}" == "1" ]]; then
die "docker/docker compose missing and SKIP_DEP_INSTALL=1"
fi
if ! command -v apt-get >/dev/null 2>&1; then
die "docker/docker compose missing; automatic install currently supports apt-get only"
fi
log "installing missing runtime packages via apt"
apt-get update
DEBIAN_FRONTEND=noninteractive apt-get install -y ca-certificates curl jq rsync gzip tar docker.io
if ! docker_compose_available; then
if apt-cache show docker-compose-v2 >/dev/null 2>&1; then
DEBIAN_FRONTEND=noninteractive apt-get install -y docker-compose-v2
elif apt-cache show docker-compose-plugin >/dev/null 2>&1; then
DEBIAN_FRONTEND=noninteractive apt-get install -y docker-compose-plugin
elif apt-cache show docker-compose >/dev/null 2>&1; then
DEBIAN_FRONTEND=noninteractive apt-get install -y docker-compose
fi
fi
systemctl enable --now docker >/dev/null 2>&1 || true
docker_compose_available || die "docker compose is still unavailable after install"
}
load_installer_images() {
require_cmd docker
shopt -s nullglob
local image
local found=0
for image in "$INSTALLER_ROOT"/images/*.tar "$INSTALLER_ROOT"/images/*.tar.gz; do
found=1
log "loading docker image: $image"
if [[ "$image" == *.gz ]]; then
gzip -dc "$image" | docker load
else
docker load -i "$image"
fi
done
shopt -u nullglob
(( found == 1 )) || warn "no image tar found under $INSTALLER_ROOT/images"
}
ensure_binfmt_if_needed() {
require_cmd docker
load_env
local host_arch package_arch
host_arch="$(detect_host_arch)"
package_arch="$(effective_package_arch)"
if [[ "$host_arch" == "$package_arch" ]]; then
return 0
fi
assert_arch_compatibility
log "host arch is $host_arch; enabling binfmt/qemu for package arch $package_arch"
docker run --rm --privileged tonistiigi/binfmt --install "$package_arch"
}
verify_runtime_image() {
load_env
require_cmd docker
local help_path
help_path="$(mktemp "${TMPDIR:-/tmp}/ours-rp-rpki-help.XXXXXX")"
log "verifying runtime image $RPKI_IMAGE on $RPKI_PLATFORM"
docker image inspect "$RPKI_IMAGE" >/dev/null
docker run --rm --platform "$RPKI_PLATFORM" "$RPKI_IMAGE" /opt/ours-rp/bin/rpki --help >"$help_path"
head -5 "$help_path" || true
rm -f "$help_path"
}
verify_image_platform() {
local image="$1"
local expected_platform="$2"
local role="$3"
local actual_platform
docker image inspect "$image" >/dev/null
actual_platform="$(docker image inspect --format '{{.Os}}/{{.Architecture}}' "$image" 2>/dev/null || echo unknown)"
[[ "$actual_platform" == "$expected_platform" ]] || die "$role image platform mismatch: image=$image expected=$expected_platform actual=$actual_platform"
}
verify_monitor_images() {
load_env
require_cmd docker
verify_image_platform "$PROMETHEUS_IMAGE" "$MONITOR_PLATFORM" "prometheus"
verify_image_platform "$GRAFANA_IMAGE" "$MONITOR_PLATFORM" "grafana"
}
endpoint_ok() {
local url="$1"
curl -fsS --max-time 5 "$url" >/dev/null 2>&1
}