feat: share named volumes across Docker and rootless Podman updates [CI]
ci/woodpecker/push/main-amd64-build Pipeline was successful
ci/woodpecker/push/main-amd64-build Pipeline was successful
This commit is contained in:
16 files changed
+473
-156
No files matched your search
@@ -7,6 +7,8 @@
|
||||
# - 镜像仓库、组件、服务名都来自本机的配置（/etc/monitor-update/<组件>.conf），清单里写什么都不会改变拉哪个仓库。
|
||||
# - 清单的 seq 必须比已安装的大（防止旧清单被重放回滚）。
|
||||
# - 业务进程（Agent、Probe）只能往数据目录里写「清单」和「请求」文件，更新由这个宿主机脚本完成，业务进程不执行任何命令。
|
||||
# Shell snippets below execute inside helper containers.
|
||||
# shellcheck disable=SC2016
|
||||
set -eu
|
||||
|
||||
CMD="${1:-}"
|
||||
@@ -14,24 +16,63 @@ COMPONENT="${2:-}"
|
||||
case "$CMD" in check|apply) ;; *) echo "usage: monitor-update <check|apply> <agent|probe>" >&2; exit 2 ;; esac
|
||||
case "$COMPONENT" in agent|probe) ;; *) echo "component must be agent or probe" >&2; exit 2 ;; esac
|
||||
|
||||
CONF="${MONITOR_UPDATE_CONF_DIR:-/etc/monitor-update}/$COMPONENT.conf"
|
||||
if [ "$(id -u)" -eq 0 ]; then
|
||||
DEFAULT_CONF_DIR=/etc/monitor-update
|
||||
else
|
||||
DEFAULT_CONF_DIR="${XDG_CONFIG_HOME:-$HOME/.config}/monitor"
|
||||
fi
|
||||
CONF="${MONITOR_UPDATE_CONF_DIR:-$DEFAULT_CONF_DIR}/$COMPONENT.conf"
|
||||
[ -r "$CONF" ] || { echo "missing $CONF" >&2; exit 2; }
|
||||
# 配置：DATA_DIR（容器里 /etc/monitor 对应的宿主机目录）、IMAGE_REPO、UNIT、PUBKEY、POLICY（auto|manual）、AUTHFILE（可选）、
|
||||
# STATE_DIR（已安装版本的记录）、ENV_FILE（容器单元读取的 IMAGE=… 文件）
|
||||
# shellcheck disable=SC1090
|
||||
. "$CONF"
|
||||
: "${DATA_DIR:?}" "${IMAGE_REPO:?}" "${UNIT:?}" "${PUBKEY:?}" "${ENV_FILE:?}"
|
||||
: "${IMAGE_REPO:?}" "${UNIT:?}" "${PUBKEY:?}" "${ENV_FILE:?}"
|
||||
POLICY="${POLICY:-manual}"
|
||||
STATE_DIR="${STATE_DIR:-/var/lib/monitor-update}"
|
||||
ENGINE="${ENGINE:-podman}"
|
||||
SERVICE_SCOPE="${SERVICE_SCOPE:-system}"
|
||||
DATA_VOLUME="${DATA_VOLUME:-}"
|
||||
STATUS_VOLUME="${STATUS_VOLUME:-monitor-$COMPONENT-status}"
|
||||
case "$ENGINE" in podman|docker) ;; *) echo "unsupported engine: $ENGINE" >&2; exit 2 ;; esac
|
||||
case "$SERVICE_SCOPE" in system|user) ;; *) echo "unsupported service scope: $SERVICE_SCOPE" >&2; exit 2 ;; esac
|
||||
service_ctl() {
|
||||
if [ "$SERVICE_SCOPE" = user ]; then systemctl --user "$@"; else systemctl "$@"; fi
|
||||
}
|
||||
pull_image() {
|
||||
if [ -n "${AUTHFILE:-}" ]; then
|
||||
if [ "$ENGINE" = podman ]; then
|
||||
"$ENGINE" pull --quiet --authfile "$AUTHFILE" "$1"
|
||||
else
|
||||
mkdir -m 700 "$WORK/docker-auth"
|
||||
cp "$AUTHFILE" "$WORK/docker-auth/config.json"
|
||||
DOCKER_CONFIG="$WORK/docker-auth" "$ENGINE" pull --quiet "$1"
|
||||
fi
|
||||
else
|
||||
"$ENGINE" pull --quiet "$1"
|
||||
fi
|
||||
}
|
||||
# 业务进程（容器里的 Agent、Probe）可写的目录只有 update/：它在里面放 offer.manifest、offer.sig、apply.request。
|
||||
# 这个脚本以 root 运行，所以：只读取普通文件（不跟符号链接）、读取之后先复制到自己的临时目录再验证和使用（防止验证之后被换掉），
|
||||
# 并且绝不往 update/ 里写任何东西；状态写在 root 拥有的 update-status/ 里（业务进程只读）。
|
||||
UPDATE_DIR="$DATA_DIR/update"
|
||||
STATUS_DIR="$DATA_DIR/update-status"
|
||||
STATUS="$STATUS_DIR/status.json"
|
||||
# 更新器在容器外运行：只读取普通文件（不跟符号链接）、读取之后先复制到自己的临时目录再验证和使用（防止验证之后被换掉），
|
||||
# 状态目录在 rootless 容器里另行只读挂载；更新器配置和已安装版本状态不挂入容器。
|
||||
if [ -n "$DATA_VOLUME" ]; then
|
||||
# Read/write volume contents via short-lived engine-managed helpers, without host mountpoint access.
|
||||
: "${HELPER_IMAGE:?}"
|
||||
"$ENGINE" volume inspect "$DATA_VOLUME" >/dev/null
|
||||
"$ENGINE" volume inspect "$STATUS_VOLUME" >/dev/null
|
||||
else
|
||||
: "${DATA_DIR:?}"
|
||||
UPDATE_DIR="$DATA_DIR/update"
|
||||
STATUS_DIR="$DATA_DIR/update-status"
|
||||
STATUS="$STATUS_DIR/status.json"
|
||||
mkdir -p "$STATUS_DIR"
|
||||
chmod 755 "$STATUS_DIR"
|
||||
fi
|
||||
INSTALLED="$STATE_DIR/$COMPONENT.installed" # 两行：seq、digest
|
||||
mkdir -p "$STATE_DIR" "$STATUS_DIR"
|
||||
chmod 755 "$STATUS_DIR"
|
||||
mkdir -p "$STATE_DIR"
|
||||
# Serialize check/apply path events for this component.
|
||||
exec 9>"$STATE_DIR/$COMPONENT.lock"
|
||||
flock -w 290 9 || { echo "another update is still running" >&2; exit 1; }
|
||||
WORK="$(mktemp -d)"
|
||||
trap 'rm -rf "$WORK"' EXIT
|
||||
OFFER_MANIFEST="$WORK/offer.manifest"
|
||||
@@ -41,6 +82,12 @@ OFFER_SIG="$WORK/offer.sig"
|
||||
fetch_offer() {
|
||||
rm -f "$OFFER_MANIFEST" "$OFFER_SIG"
|
||||
for name in offer.manifest offer.sig; do
|
||||
if [ -n "$DATA_VOLUME" ]; then
|
||||
"$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||||
-v "$DATA_VOLUME:/etc/monitor:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||||
-c 'f="/etc/monitor/update/$1"; if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f"; fi' sh "$name" > "$WORK/$name"
|
||||
continue
|
||||
fi
|
||||
f="$UPDATE_DIR/$name"
|
||||
if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f" > "$WORK/$name"; fi
|
||||
done
|
||||
@@ -48,13 +95,22 @@ fetch_offer() {
|
||||
|
||||
now() { date +%s; }
|
||||
|
||||
# 状态文件：Agent / Probe 读它并随心跳报给面板。root 写，业务进程只读。
|
||||
# 状态文件：更新器写，Agent / Probe 从只读挂载中读取并随心跳上报。
|
||||
write_status() { # state available_json error
|
||||
tmp="$(mktemp "$STATUS_DIR/.status.XXXXXX")"
|
||||
tmp="$WORK/status.json"
|
||||
printf '{"state":"%s","current":"%s","available":%s,"error":"%s","at":%s}\n' \
|
||||
"$1" "$(current_digest)" "${2:-null}" "$(printf '%s' "${3:-}" | tr -d '"\\\n' | cut -c1-200)" "$(now)" > "$tmp"
|
||||
chmod 644 "$tmp"
|
||||
mv -f "$tmp" "$STATUS"
|
||||
if [ -n "$DATA_VOLUME" ]; then
|
||||
"$ENGINE" run --pull=never --rm -i --user 0 --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||||
-v "$STATUS_VOLUME:/status" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||||
-c 'tmp=$(mktemp /status/.status.XXXXXX); cat > "$tmp"; chmod 644 "$tmp"; mv -f "$tmp" /status/status.json' < "$tmp"
|
||||
else
|
||||
chmod 644 "$tmp"
|
||||
destination="$(mktemp "$STATUS_DIR/.status.XXXXXX")"
|
||||
cp "$tmp" "$destination"
|
||||
chmod 644 "$destination"
|
||||
mv -f "$destination" "$STATUS"
|
||||
fi
|
||||
}
|
||||
|
||||
current_digest() { [ -r "$INSTALLED" ] && sed -n 2p "$INSTALLED" || echo ""; }
|
||||
@@ -82,7 +138,7 @@ verify_offer() {
|
||||
[ "$M_SEQ" -gt "$(installed_seq)" ] || { ERR="stale manifest (seq $M_SEQ <= $(installed_seq))"; return 1; }
|
||||
}
|
||||
|
||||
available_json() { printf '{"version":"%s","tag":"%s","digest":"%s","seq":%s}' "$(printf '%s' "$M_VERSION" | tr -d '"\\')" "$(printf '%s' "$M_TAG" | tr -d '"\\')" "$M_DIGEST" "$M_SEQ"; }
|
||||
available_json() { printf '{"version":"%s","tag":"%s","digest":"%s","seq":%s}' "$(printf '%s' "$M_VERSION" | tr -d '\042\134')" "$(printf '%s' "$M_TAG" | tr -d '\042\134')" "$M_DIGEST" "$M_SEQ"; }
|
||||
|
||||
do_check() {
|
||||
fetch_offer
|
||||
@@ -105,25 +161,29 @@ do_check() {
|
||||
}
|
||||
|
||||
do_apply() {
|
||||
rm -f "$UPDATE_DIR/apply.request"
|
||||
if [ -n "$DATA_VOLUME" ]; then
|
||||
"$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||||
-v "$DATA_VOLUME:/etc/monitor" --entrypoint /bin/sh "$HELPER_IMAGE" -c 'rm -f /etc/monitor/update/apply.request'
|
||||
else
|
||||
rm -f "$UPDATE_DIR/apply.request"
|
||||
fi
|
||||
fetch_offer
|
||||
verify_offer || { write_status invalid null "$ERR"; echo "refusing to apply: $ERR" >&2; return 1; }
|
||||
avail="$(available_json)"
|
||||
write_status applying "$avail" ""
|
||||
ref="$IMAGE_REPO@$M_DIGEST"
|
||||
if [ -n "${AUTHFILE:-}" ]; then export REGISTRY_AUTH_FILE="$AUTHFILE"; fi
|
||||
if ! podman pull --quiet "$ref" >/dev/null 2>&1; then
|
||||
if ! pull_image "$ref" >/dev/null 2>&1; then
|
||||
write_status failed "$avail" "pull failed"; echo "pull failed: $ref" >&2; return 1
|
||||
fi
|
||||
prev="$(cat "$ENV_FILE" 2>/dev/null || true)"
|
||||
printf 'IMAGE=%s\n' "$ref" > "$ENV_FILE.new" && mv -f "$ENV_FILE.new" "$ENV_FILE"
|
||||
systemctl restart "$UNIT" || true
|
||||
service_ctl restart "$UNIT" || true
|
||||
# 健康确认：服务在跑，并且 20 秒内没有再被重启（崩溃循环算失败）
|
||||
sleep 5
|
||||
before="$(systemctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||||
before="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||||
sleep 15
|
||||
after="$(systemctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||||
if systemctl is-active --quiet "$UNIT" && [ "$before" = "$after" ]; then
|
||||
after="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||||
if service_ctl is-active --quiet "$UNIT" && [ "$before" = "$after" ]; then
|
||||
printf '%s\n%s\n' "$M_SEQ" "$M_DIGEST" > "$INSTALLED"
|
||||
write_status up_to_date null ""
|
||||
echo "updated to $M_TAG"
|
||||
@@ -131,12 +191,20 @@ do_apply() {
|
||||
fi
|
||||
echo "new version did not become healthy, rolling back" >&2
|
||||
if [ -n "$prev" ]; then printf '%s\n' "$prev" > "$ENV_FILE"; fi
|
||||
systemctl restart "$UNIT" || true
|
||||
service_ctl restart "$UNIT" || true
|
||||
write_status rolled_back "$avail" "new version did not become healthy"
|
||||
return 1
|
||||
}
|
||||
|
||||
case "$CMD" in
|
||||
check) do_check ;;
|
||||
check)
|
||||
if [ -n "$DATA_VOLUME" ] && [ "$POLICY" != auto ]; then
|
||||
requested="$($ENGINE run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||||
-v "$DATA_VOLUME:/etc/monitor:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||||
-c '[ ! -f /etc/monitor/update/apply.request ] || printf yes')"
|
||||
if [ "$requested" = yes ]; then do_apply; else do_check; fi
|
||||
else
|
||||
do_check
|
||||
fi ;;
|
||||
apply) do_apply ;;
|
||||
esac
|
||||
Reference in new issue
Block a user