ServerKit is a lightweight, modern server control panel for managing web applications, databases, and services on your VPS or dedicated server. Built with Python/Flask backend and React frontend.
Python astracloudflarecontrol-panelcyberpaneldnsfablehosting-control-panelhosting-panelkimi-k3server-managementserverkitubuntu-web-panelwebserver
fix(install): close data-loss re-run + no-domain abort, harden pkg/venv/docker paths (I1-I10,I12,I20,I21)
Juan Denisf2e7715479f4
diff --git a/install.sh b/install.shindex 9c34f36b..a4676c66 100644--- a/install.sh+++ b/install.sh@@ -271,32 +271,68 @@ APT_EOF}refresh_pkg_index() {+ # Best-effort by design (mirrors scripts/lib/pkg.sh pkg_refresh): a failed+ # index refresh — flaky mirror, momentary lock — must never abort the+ # install, so every arm is guarded and the function always returns 0.+ # The --refresh flag is dnf-only; classic yum has no such flag, so its arm+ # is a plain makecache (same as pkg.sh).case "$PKG_MGR" in- apt) apt-get update -y >/dev/null 2>&1 ;;- dnf) dnf makecache --refresh >/dev/null 2>&1 ;;- yum) yum makecache --refresh >/dev/null 2>&1 ;;- zypper) zypper --non-interactive refresh >/dev/null 2>&1 ;;- pacman) pacman -Sy --noconfirm >/dev/null 2>&1 ;;- apk) apk update >/dev/null 2>&1 ;;+ apt) apt-get update -y >/dev/null 2>&1 || true ;;+ dnf) dnf makecache --refresh >/dev/null 2>&1 || true ;;+ yum) yum makecache >/dev/null 2>&1 || true ;;+ zypper) zypper --non-interactive refresh >/dev/null 2>&1 || true ;;+ pacman) pacman -Sy --noconfirm >/dev/null 2>&1 || true ;;+ apk) apk update >/dev/null 2>&1 || true ;;esac+ return 0}+# Warn-and-continue package install. Output is captured so a failure can be+# reported with the manager's last lines — and that capture carries an+# `|| rc=$?` guard because a bare `out=$(apt-get ...)` assignment is itself+# an abort point under `set -e`: the old body died on the assignment, before+# it could print anything at all. Always returns 0; callers that *require* a+# package must probe the resulting state (command -v, locate_python, ...)+# rather than this exit code.pkg_add() {- local out rc+ local out="" rc=0case "$PKG_MGR" in- apt) out=$(apt-get install -y "$@" 2>&1) ;;- dnf) out=$(dnf install -y "$@" 2>&1) ;;- yum) out=$(yum install -y "$@" 2>&1) ;;- zypper) out=$(zypper --non-interactive install "$@" 2>&1) ;;- pacman) out=$(pacman -S --noconfirm "$@" 2>&1) ;;- apk) out=$(apk add "$@" 2>&1) ;;+ apt) out=$(apt-get install -y "$@" 2>&1) || rc=$? ;;+ dnf) out=$(dnf install -y "$@" 2>&1) || rc=$? ;;+ yum) out=$(yum install -y "$@" 2>&1) || rc=$? ;;+ zypper) out=$(zypper --non-interactive install "$@" 2>&1) || rc=$? ;;+ pacman) out=$(pacman -S --noconfirm "$@" 2>&1) || rc=$? ;;+ apk) out=$(apk add "$@" 2>&1) || rc=$? ;;+ *) warn "No supported package manager — cannot install: $*"; return 0 ;;esac- rc=$?- if [ $rc -ne 0 ]; then+ if [ "$rc" -ne 0 ]; thenwarn "Could not install: $* (exit $rc)"- printf '%s\n' "$out" | tail -5 >&2- return $rc+ printf '%s\n' "$out" | tail -5 >&2 || truefi+ return 0+}++# ---------------------------------------------------------------------------+# RHEL-family (Rocky/Alma/RHEL/CentOS 9): upgrade openssh + openssl TOGETHER,+# up front, before any other dnf transaction runs.+#+# Rocky 9 images ship openssh linked against openssl-libs 3.0.x while the+# updates stream carries openssl-libs 3.5.x plus a matching openssh rebuild.+# Any later `dnf install` that pulls dependencies can transitively upgrade+# openssl-libs; if openssh is not upgraded in the SAME transaction, every new+# sshd fork dies with "OpenSSL version mismatch" — remote installs over SSH+# then cut themselves off mid-run. Upgrading both together lets openssh's+# scriptlet restart sshd against the matched libssl (KillMode=process keeps+# the live SSH session alive). Do NOT rewrite this with --exclude=openssl+# shapes; they fail on python3-libs symbol requirements. Best-effort: a box+# without the updates repo (or already current) is a clean no-op.+# ---------------------------------------------------------------------------+upgrade_rhel_crypto_stack() {+ [ "$OS_FAMILY" = "rhel" ] || return 0+ command -v dnf &>/dev/null || return 0+ step "Upgrading openssh/openssl together (avoids sshd 'OpenSSL version mismatch')..."+ dnf upgrade -y openssh openssh-server openssh-clients openssl openssl-libs openssl-devel 2>/dev/null || true+ return 0}# ---------------------------------------------------------------------------@@ -338,16 +374,36 @@ ver_in_range() {printf '%s\n%s' "$1" "$PYTHON_MAX" | sort -C -V}+# True when <python> can actually create virtual environments. Debian/Ubuntu+# minimal images split the venv module (ensurepip) into pythonX.Y-venv, so an+# otherwise-valid interpreter still dies much later at `python -m venv` with+# "ensurepip is not available" — probe up front instead. (I8)+py_venv_ok() {+ "$1" -m venv --help >/dev/null 2>&1+}+locate_python() {# Prefer an explicit minor version, newest first, then bare python3. This# mirrors scripts/update.sh so a box that has python3.11 (Debian 12) but a- # too-new/too-old default python3 is still recognized. Sets PYTHON_BIN and- # returns 0 on success; returns 1 (without aborting) when nothing fits.+ # too-new/too-old default python3 is still recognized. A candidate must+ # also be venv-capable (py_venv_ok); on Debian-family boxes the missing+ # pythonX.Y-venv package is installed on demand before the candidate is+ # given up on. Sets PYTHON_BIN and returns 0 on success; returns 1+ # (without aborting) when nothing fits.local c vfor c in python3.12 python3.11 python3; doif command -v "$c" &>/dev/null; thenv=$("$c" -c 'import sys;print(".".join(map(str,sys.version_info[:2])))' 2>/dev/null || true)if [ -n "$v" ] && ver_in_range "$v"; then+ if ! py_venv_ok "$c" && [ "$OS_FAMILY" = "debian" ]; then+ step "Installing python${v}-venv (the venv module is missing)..."+ pkg_add "python${v}-venv"+ py_venv_ok "$c" || pkg_add python3-venv+ fi+ if ! py_venv_ok "$c"; then+ warn "$c (Python $v) cannot create virtualenvs (no venv/ensurepip) — skipping it."+ continue+ fiPYTHON_BIN="$c"good "Using $c (Python $v)"return 0@@ -382,31 +438,37 @@ provision_python() {# Prefer the distro's own package over a source compile. Debian 12 ships# python3.11; Ubuntu 24.04 ships python3.12 (older Ubuntu falls back to the# deadsnakes PPA); Fedora/RHEL provide 3.12 or 3.11 in their repos. A slow,- # fragile source build is now strictly the last resort.+ # fragile source build is now strictly the last resort. pkg_add is+ # warn-and-continue (always returns 0), so each attempt is followed by a+ # state probe (command -v) to decide whether the next fallback is needed.if [ "$OS_FAMILY" = "debian" ]; thenif [ "${ID:-}" = "ubuntu" ]; then- if ! pkg_add python3.12 python3.12-venv python3.12-dev; then+ pkg_add python3.12 python3.12-venv python3.12-dev+ if ! command -v python3.12 &>/dev/null; thenstep "Adding deadsnakes PPA for Python 3.12..."pkg_add software-properties-common- add-apt-repository -y ppa:deadsnakes/ppa+ add-apt-repository -y ppa:deadsnakes/ppa || truerefresh_pkg_indexpkg_add python3.12 python3.12-venv python3.12-devfielse# Debian (and Debian-like): python3.11 lives in the main repo.refresh_pkg_index- pkg_add python3.11 python3.11-venv python3.11-dev || \- pkg_add python3.12 python3.12-venv python3.12-dev || true+ pkg_add python3.11 python3.11-venv python3.11-dev+ command -v python3.11 &>/dev/null || \+ pkg_add python3.12 python3.12-venv python3.12-devfielif [ "$OS_FAMILY" = "fedora" ] || [ "$OS_FAMILY" = "rhel" ]; then- pkg_add python3.12 python3.12-devel || pkg_add python3.11 python3.11-devel || true+ pkg_add python3.12 python3.12-devel+ command -v python3.12 &>/dev/null || pkg_add python3.11 python3.11-develelif [ "$OS_FAMILY" = "suse" ]; then- pkg_add python311 python311-devel || pkg_add python312 python312-devel || true+ pkg_add python311 python311-devel+ command -v python3.11 &>/dev/null || pkg_add python312 python312-develelif [ "$OS_FAMILY" = "arch" ]; then# Rolling release — `python` is already 3.11+.- pkg_add python || true+ pkg_add pythonelif [ "$OS_FAMILY" = "alpine" ]; then- pkg_add python3 python3-dev || true+ pkg_add python3 python3-devfi# Did a distro package give us something usable?@@ -436,6 +498,22 @@ provision_python() {# ---------------------------------------------------------------------------# Docker engine + compose plugin# ---------------------------------------------------------------------------+# Add Docker's upstream .repo file, handling both config-manager generations:+# classic dnf4 uses `--add-repo URL`, while dnf5 (Fedora 41+) removed that+# flag in favour of `addrepo --from-repofile=URL`. Warn-and-continue: when+# neither works the docker-ce install below fails too (also non-fatally). (I10)+docker_repo_add() {+ local url="$1"+ if dnf config-manager --add-repo "$url" >/dev/null 2>&1; then+ return 0+ fi+ if dnf config-manager addrepo --from-repofile="$url" >/dev/null 2>&1; then+ return 0+ fi+ warn "Could not add the Docker repository ($url)."+ return 0+}+provision_docker() {phase "Docker"@@ -448,12 +526,12 @@ provision_docker() {case "$OS_FAMILY" infedora)pkg_add dnf-plugins-core- dnf config-manager --add-repo https://download.docker.com/linux/fedora/docker-ce.repo+ docker_repo_add https://download.docker.com/linux/fedora/docker-ce.repopkg_add docker-ce docker-ce-cli containerd.io docker-compose-plugin docker-buildx-plugin;;rhel)pkg_add dnf-plugins-core- dnf config-manager --add-repo https://download.docker.com/linux/rhel/docker-ce.repo+ docker_repo_add https://download.docker.com/linux/rhel/docker-ce.repopkg_add docker-ce docker-ce-cli containerd.io docker-compose-plugin docker-buildx-plugin;;suse)@@ -488,8 +566,21 @@ ensure_compose_plugin() {returnfistep "Installing Docker Compose plugin..."+ # Fallback chain (I9): when Docker came from the distro repo (docker.io),+ # Docker's own repo was never configured, so 'docker-compose-plugin' does+ # not exist there — Ubuntu ships the same plugin as 'docker-compose-v2'.+ # Probe after each attempt; if neither lands, warn and continue (the box+ # may have compose v1, or the operator can add it later).pkg_add docker-compose-plugin- good "Docker Compose plugin installed."+ if ! docker compose version &>/dev/null; then+ pkg_add docker-compose-v2+ fi+ if docker compose version &>/dev/null; then+ good "Docker Compose plugin installed."+ else+ warn "Could not install the Docker Compose plugin — 'docker compose' is unavailable."+ warn "Managed app deployments need it; install docker-compose-plugin or docker-compose-v2 manually."+ fi}# ---------------------------------------------------------------------------@@ -539,6 +630,47 @@ provision_node() {good "Node.js $(node --version) installed."}+# ---------------------------------------------------------------------------+# Live-state carry: keep .env + backend/instance/ across a slot rewrite+# ---------------------------------------------------------------------------+# fetch_release and sync_source both rebuild $FIRST_SLOT with an rm -rf. On a+# re-run over an existing install that tree holds irreplaceable state: the+# generated .env (SECRET_KEY / JWT_SECRET_KEY / SERVERKIT_ENCRYPTION_KEY —+# losing that key orphans every encrypted secret) and backend/instance/ (the+# SQLite database). Copy them aside before the rewrite and restore them after,+# mirroring scripts/update.sh's deploy_source carry-forward. Both helpers are+# best-effort and always return 0 — preservation must never abort an install.+stash_live_state() {+ # Echoes a stash directory holding the live state, or nothing when there+ # is nothing to preserve (fresh install) or no stash could be created.+ local src="$1" stash+ if [ ! -f "$src/.env" ] && [ ! -d "$src/backend/instance" ]; then+ return 0+ fi+ stash="$(mktemp -d 2>/dev/null || true)"+ [ -n "$stash" ] || return 0+ cp -a "$src/.env" "$stash/.env" 2>/dev/null || true+ cp -a "$src/backend/instance" "$stash/instance" 2>/dev/null || true+ printf '%s' "$stash"+ return 0+}++restore_live_state() {+ local stash="$1" dest="$2"+ [ -n "$stash" ] || return 0+ if [ -f "$stash/.env" ]; then+ cp -a "$stash/.env" "$dest/.env" 2>/dev/null || true+ fi+ if [ -d "$stash/instance" ]; then+ mkdir -p "$dest/backend" 2>/dev/null || true+ rm -rf "$dest/backend/instance" 2>/dev/null || true+ cp -a "$stash/instance" "$dest/backend/instance" 2>/dev/null || true+ fi+ rm -rf "$stash" 2>/dev/null || true+ good "Preserved the existing .env and database across the rewrite."+ return 0+}+# ---------------------------------------------------------------------------# Release tarball path# ---------------------------------------------------------------------------@@ -608,9 +740,19 @@ fetch_release() {[ -d "$unpacked" ] || halt "Release tarball layout is unrecognized (expected serverkit/ or opt/serverkit/)."ensure_install_layout++ # A re-run over a live install must never destroy its secrets or database+ # (the old bare rm -rf below did exactly that). Stash them, rewrite,+ # restore. (I4)+ local live_dir stash+ live_dir="$(readlink -f "$INSTALL_DIR" 2>/dev/null || echo "$FIRST_SLOT")"+ [ -d "$live_dir" ] || live_dir="$FIRST_SLOT"+ stash="$(stash_live_state "$live_dir")"+rm -rf "$FIRST_SLOT"cp -a "$unpacked" "$FIRST_SLOT"rm -rf "$tmp_dir"+ restore_live_state "$stash" "$FIRST_SLOT"[ -L "$INSTALL_DIR" ] || ln -s "$FIRST_SLOT" "$INSTALL_DIR"@@ -625,6 +767,16 @@ fetch_release() {sync_source() {phase "Source Code"+ # Source installs clone with git, but minimal images don't guarantee it+ # and the preflight never checked — install it on demand, and halt only+ # if that fails too (a source install cannot proceed without git). (I7)+ if ! command -v git &>/dev/null; then+ step "Installing git..."+ pkg_add git+ command -v git &>/dev/null || \+ halt "git is required for a source install but could not be installed."+ fi+ensure_install_layoutif [ -d "$FIRST_SLOT/.git" ]; then@@ -637,8 +789,14 @@ sync_source() {fielsestep "Cloning ServerKit..."+ # Same live-state carry as fetch_release: the non-git tree replaced+ # here may still be a live install (e.g. one installed from a release+ # tarball) holding the generated .env and the database. (I4)+ local stash+ stash="$(stash_live_state "$FIRST_SLOT")"rm -rf "$FIRST_SLOT"git clone --depth 1 "https://github.com/${GITHUB_REPO}.git" "$FIRST_SLOT"+ restore_live_state "$stash" "$FIRST_SLOT"cd "$FIRST_SLOT"fi[ -L "$INSTALL_DIR" ] || ln -sfn "$FIRST_SLOT" "$INSTALL_DIR"@@ -687,6 +845,18 @@ build_virtualenv() {# If the release shipped a pre-built venv at the expected path, use it.if [ "$INSTALL_FROM_RELEASE" = "1" ] && [ -f "$FIRST_SLOT/venv/bin/activate" ] && [ -x "$FIRST_SLOT/venv/bin/python" ]; thenstep "Using pre-built virtual environment from release..."+ # In the default layout $INSTALL_DIR is a symlink to $FIRST_SLOT, so+ # $VENV_DIR *is* $FIRST_SLOT/venv — the rm -rf below would delete the+ # very venv it is about to "copy", and the cp would then fail. When+ # source and destination resolve to the same directory the venv is+ # already in place; verify and keep it. (I3)+ local src_venv dst_venv+ src_venv="$(readlink -f "$FIRST_SLOT/venv" 2>/dev/null || echo "$FIRST_SLOT/venv")"+ dst_venv="$(readlink -f "$VENV_DIR" 2>/dev/null || echo "$VENV_DIR")"+ if [ "$src_venv" = "$dst_venv" ]; then+ good "Virtual environment already in place from the release."+ return+ firm -rf "$VENV_DIR"cp -a "$FIRST_SLOT/venv" "$VENV_DIR"good "Virtual environment installed from release."@@ -839,9 +1009,12 @@ configure_firewall() {step "Opening HTTP/HTTPS (80, 443) via $backend..."firewall_open "$backend" 80/tcp 443/tcpif [ "${FW_DRY_RUN:-0}" != "1" ] && command -v state_set >/dev/null 2>&1; then- state_set firewall_backend "$backend"- state_append firewall_ports 80/tcp- state_append firewall_ports 443/tcp+ # Best-effort, exactly like update.sh's ensure_firewall: failing to+ # RECORD the opened ports must not abort an install whose firewall+ # work already succeeded. (I21)+ state_set firewall_backend "$backend" || true+ state_append firewall_ports 80/tcp || true+ state_append firewall_ports 443/tcp || truefigood "Firewall configured ($backend): 80/tcp and 443/tcp open."}@@ -960,18 +1133,28 @@ choose_ssl_mode() {# Install the nginx config matching the chosen SSL mode.# ---------------------------------------------------------------------------install_nginx_config_for_mode() {+ # Dirs are overridable so the unit tests can exercise this against+ # fixtures (same pattern as harden_global_tls / update.sh's NGINX_DIR).+ local nginx_dir="${SERVERKIT_NGINX_DIR:-/etc/nginx}"+ local config_dir="${SERVERKIT_CONFIG_DIR:-/etc/serverkit}"if [ "$SSL_MODE" = "secure" ]; thensed -i "s|/etc/letsencrypt/live/YOUR_DOMAIN/|/etc/letsencrypt/live/$PANEL_DOMAIN/|g" \- /etc/nginx/sites-available/serverkit.conf- ln -sf /etc/nginx/sites-available/serverkit.conf /etc/nginx/sites-enabled/serverkit.conf+ "$nginx_dir/sites-available/serverkit.conf"+ ln -sf "$nginx_dir/sites-available/serverkit.conf" "$nginx_dir/sites-enabled/serverkit.conf"else- ln -sf /etc/nginx/sites-available/serverkit-insecure.conf /etc/nginx/sites-enabled/serverkit.conf+ ln -sf "$nginx_dir/sites-available/serverkit-insecure.conf" "$nginx_dir/sites-enabled/serverkit.conf"fi- mkdir -p /etc/serverkit- printf '%s\n' "$SSL_MODE" > /etc/serverkit/ssl-mode+ mkdir -p "$config_dir"+ printf '%s\n' "$SSL_MODE" > "$config_dir/ssl-mode"# Persist the domain so update.sh can re-apply the cert path on upgrades- # without depending on the (commented-out) .env public URL.- [ -n "$PANEL_DOMAIN" ] && printf '%s\n' "$PANEL_DOMAIN" > /etc/serverkit/panel-domain+ # without depending on the (commented-out) .env public URL. An empty+ # domain (the default no-domain install) is simply not persisted — and as+ # the LAST statement of the function this must be an `if`, not a+ # `[ -n ] && ...` list: the list form returned 1 on an empty domain and+ # set -e killed every no-domain install right here. (I1)+ if [ -n "$PANEL_DOMAIN" ]; then+ printf '%s\n' "$PANEL_DOMAIN" > "$config_dir/panel-domain"+ fi}# Point the static-frontend `root` at the real install dir. The shipped nginx@@ -1184,10 +1367,26 @@ await_health() {snapshot_existing() {local activeactive="$(readlink -f "$INSTALL_DIR" 2>/dev/null || echo "$INSTALL_DIR")"- if [ -d "$active" ] && [ ! -d "$active.backup" ]; then+ [ -d "$active" ] || return 0+ if [ ! -d "$active.backup" ]; thenstep "Backing up the existing installation..."cp -a "$active" "$active.backup"good "Backup saved to $active.backup"+ return 0+ fi+ # A .backup left by an earlier run is what revert_install restores — a+ # stale one would silently roll the box back to a months-old tree.+ # Refresh it: build the new copy first and swap only on success, so a+ # failed copy never costs us the backup we already have. (I20)+ step "Refreshing the existing installation backup..."+ rm -rf "$active.backup.new" 2>/dev/null || true+ if cp -a "$active" "$active.backup.new" 2>/dev/null; then+ rm -rf "$active.backup"+ mv "$active.backup.new" "$active.backup"+ good "Backup refreshed at $active.backup"+ else+ rm -rf "$active.backup.new" 2>/dev/null || true+ warn "Could not refresh $active.backup — keeping the previous backup."fi}@@ -1250,28 +1449,54 @@ ping_telemetry() {# Ask for a panel domain when running interactively# ---------------------------------------------------------------------------prompt_for_domain() {- [ -z "$PANEL_DOMAIN" ] && [ -t 0 ] || return 0+ # Nothing to ask when a domain is already set, or when there is no+ # interactive stdin to ask on (curl | bash). SERVERKIT_FORCE_PROMPT=1+ # lets the unit tests drive the prompt from a pipe.+ [ -z "$PANEL_DOMAIN" ] || return 0+ if [ ! -t 0 ] && [ "${SERVERKIT_FORCE_PROMPT:-0}" != "1" ]; then+ return 0+ fiprintf '\n'printf '%sEnter your panel domain (e.g. panel.example.com)%s\n' "$BLD" "$RST"printf 'Leave blank to access via IP (no SSL attempt)\n'printf '%sTip:%s set PANEL_DOMAIN=... in the environment to skip this prompt\n' "$BLD" "$RST"printf ' set SERVERKIT_SKIP_SSL=1 to disable HTTPS entirely\n'printf '> '- read -r PANEL_DOMAIN+ read -r PANEL_DOMAIN || truePANEL_DOMAIN=$(printf '%s' "$PANEL_DOMAIN" | tr -d ' ')- [ -n "$PANEL_DOMAIN" ] && PANEL_PORT="80"+ # A blank answer (plain Enter) is the documented default — and as the+ # LAST statement of the function this must be an `if`, not a+ # `[ -n ] && ...` list: the list form returned 1 on a blank answer and+ # set -e killed every interactive no-domain install right here. (I2)+ if [ -n "$PANEL_DOMAIN" ]; then+ PANEL_PORT="80"+ fi}# ---------------------------------------------------------------------------# Orchestration# ---------------------------------------------------------------------------+# Decide whether a fresh install should default to a release tarball. An+# EXISTING install must never flip to release mode by default: fetch_release+# rewrites the whole slot. The old check tested $INSTALL_DIR/backend/src — a+# path that never existed (the tree is backend/app) — so every re-run over a+# live install picked release mode and rm -rf'd it, .env and database+# included. Probe the real tree and the generated .env instead. (I4)+should_default_to_release() {+ [ "$BUILD_FROM_SOURCE" != "1" ] || return 1+ [ -z "${SERVERKIT_VERSION:-}" ] || return 1+ [ ! -d "$INSTALL_DIR/backend/app" ] || return 1+ [ ! -f "$INSTALL_DIR/.env" ] || return 1+ return 0+}+main() {STARTED_AT=$(date +%s)masthead- # Default to a release install unless a source tree is already present or+ # Default to a release install unless an install is already present or# the caller forced a source build / pinned a version.- if [ "$BUILD_FROM_SOURCE" != "1" ] && [ ! -d "$INSTALL_DIR/backend/src" ] && [ -z "${SERVERKIT_VERSION:-}" ]; then+ if should_default_to_release; thenINSTALL_FROM_RELEASE=1fi@@ -1283,6 +1508,10 @@ main() {prompt_for_domainsnapshot_existing+ # RHEL 9 family: keep sshd alive across the dnf work below (see the+ # function's comment for the full openssl/openssh mismatch story).+ upgrade_rhel_crypto_stack+provision_dockerensure_compose_pluginprovision_nodediff --git a/scripts/test/test_install.sh b/scripts/test/test_install.shindex a6c4a419..e0129d84 100644--- a/scripts/test/test_install.sh+++ b/scripts/test/test_install.sh@@ -309,6 +309,407 @@ elseok "docker-compose.yml is frontend-only and routes /api to the host backend (backend:host-gateway)"fi+# --------------------------------------------------------------------------+# T13 — I1: the default no-domain install must survive install_nginx_config_+# for_mode. `[ -n "$PANEL_DOMAIN" ] && printf ...` as the function's LAST+# statement returned 1 on an empty domain and set -e killed every curl-pipe+# install at the nginx phase (shipped broken since v1.6.25).+# --------------------------------------------------------------------------+t="$WORK/t13"; mkdir -p "$t/nginx/sites-available" "$t/nginx/sites-enabled" "$t/cfg"+printf 'server { listen 80; }\n' > "$t/nginx/sites-available/serverkit-insecure.conf"+printf 'ssl_certificate /etc/letsencrypt/live/YOUR_DOMAIN/fullchain.pem;\n' \+ > "$t/nginx/sites-available/serverkit.conf"+if ( set -Eeuo pipefail+ SERVERKIT_NGINX_DIR="$t/nginx" SERVERKIT_CONFIG_DIR="$t/cfg" \+ PANEL_DOMAIN="" SSL_MODE=insecure install_nginx_config_for_mode ) >/dev/null 2>&1; then+ if [ "$(cat "$t/cfg/ssl-mode")" = "insecure" ] && [ ! -f "$t/cfg/panel-domain" ]; then+ ok "install_nginx_config_for_mode survives an empty PANEL_DOMAIN (the v1.6.25 no-domain abort)"+ else+ bad "install_nginx_config_for_mode wrote the wrong mode/domain state for a no-domain install"+ fi+else+ bad "install_nginx_config_for_mode DIED with an empty PANEL_DOMAIN under set -e (I1 regression)"+fi+if ( set -Eeuo pipefail+ SERVERKIT_NGINX_DIR="$t/nginx" SERVERKIT_CONFIG_DIR="$t/cfg" \+ PANEL_DOMAIN="panel.example.com" SSL_MODE=secure install_nginx_config_for_mode ) >/dev/null 2>&1 \+ && grep -q 'live/panel.example.com/' "$t/nginx/sites-available/serverkit.conf" \+ && [ "$(cat "$t/cfg/panel-domain")" = "panel.example.com" ]; then+ ok "install_nginx_config_for_mode (secure) applies the cert path and persists the domain"+else+ bad "install_nginx_config_for_mode (secure) did not apply the cert path / persist the domain"+fi++# --------------------------------------------------------------------------+# T14 — I2: plain Enter at the interactive domain prompt must not abort.+# Same end-of-function `[ -n ] && ...` shape as I1. SERVERKIT_FORCE_PROMPT=1+# lets the test drive the prompt from a pipe instead of a tty.+# --------------------------------------------------------------------------+if printf '\n' | ( set -Eeuo pipefail+ PANEL_DOMAIN="" SERVERKIT_FORCE_PROMPT=1 prompt_for_domain ) >/dev/null 2>&1; then+ ok "prompt_for_domain survives a blank answer (plain Enter, the no-domain default)"+else+ bad "prompt_for_domain DIED on a blank answer under set -e (I2 regression)"+fi+res="$(printf 'panel.example.com\n' | (+ set -Eeuo pipefail+ PANEL_DOMAIN=""; PANEL_PORT=""; SERVERKIT_FORCE_PROMPT=1+ prompt_for_domain >/dev/null 2>&1+ printf '%s|%s' "$PANEL_DOMAIN" "$PANEL_PORT"+))"+if [ "$res" = "panel.example.com|80" ]; then+ ok "prompt_for_domain accepts a typed domain and sets the port"+else+ bad "prompt_for_domain returned [$res], expected panel.example.com|80"+fi++# --------------------------------------------------------------------------+# T15 — I4 (detection half): should_default_to_release must see an EXISTING+# install. The old check probed backend/src (never exists — the tree is+# backend/app), so every re-run picked release mode and fetch_release+# rm -rf'd the live install, .env + SQLite DB included.+# --------------------------------------------------------------------------+t="$WORK/t15"+mkdir -p "$t/fresh" "$t/existing/backend/app" "$t/envonly"+printf 'SECRET_KEY=x\n' > "$t/envonly/.env"+det_ok=1+( set -Eeuo pipefail; INSTALL_DIR="$t/fresh"; BUILD_FROM_SOURCE=0; SERVERKIT_VERSION=""+ should_default_to_release ) || { bad "fresh box should default to a release install"; det_ok=0; }+( set -Eeuo pipefail; INSTALL_DIR="$t/existing"; BUILD_FROM_SOURCE=0; SERVERKIT_VERSION=""+ should_default_to_release ) && { bad "existing backend/app tree must NOT default to release (data loss)"; det_ok=0; }+( set -Eeuo pipefail; INSTALL_DIR="$t/envonly"; BUILD_FROM_SOURCE=0; SERVERKIT_VERSION=""+ should_default_to_release ) && { bad "existing .env must NOT default to release (data loss)"; det_ok=0; }+( set -Eeuo pipefail; INSTALL_DIR="$t/fresh"; BUILD_FROM_SOURCE=1; SERVERKIT_VERSION=""+ should_default_to_release ) && { bad "BUILD_FROM_SOURCE=1 must force the source path"; det_ok=0; }+[ "$det_ok" = "1" ] && ok "should_default_to_release detects an existing install (backend/app or .env)"++# --------------------------------------------------------------------------+# T16 — I4 (preservation half): live state survives a slot rewrite+# byte-identical. First the helpers in isolation, then the real fetch_release+# offline-tarball path over a symlinked blue/green layout (Linux CI; skipped+# where symlinks are unsupported, same convention as test_update.sh T4).+# --------------------------------------------------------------------------+t="$WORK/t16"; mkdir -p "$t/slot/backend/instance"+printf 'SECRET_KEY=sentinel-abc123\nSERVERKIT_ENCRYPTION_KEY=k\n' > "$t/slot/.env"+printf 'SQLITE-SENTINEL-BYTES\n' > "$t/slot/backend/instance/serverkit.db"+cp "$t/slot/.env" "$t/env.orig"+cp "$t/slot/backend/instance/serverkit.db" "$t/db.orig"+if (+ set -Eeuo pipefail+ stash="$(stash_live_state "$t/slot")"+ [ -n "$stash" ]+ rm -rf "$t/slot" # the rewrite+ mkdir -p "$t/slot/backend" # fresh tree: no .env, no instance/+ restore_live_state "$stash" "$t/slot"+ cmp -s "$t/slot/.env" "$t/env.orig"+ cmp -s "$t/slot/backend/instance/serverkit.db" "$t/db.orig"+) >/dev/null 2>&1; then+ ok "stash/restore_live_state carries .env + instance DB byte-identical across a rewrite"+else+ bad "stash/restore_live_state lost or altered the live state"+fi++t="$WORK/t16r"; slotA="$t/sk-a"; slotB="$t/sk-b"; inst="$t/sk"+mkdir -p "$slotA/backend/instance"+printf 'SECRET_KEY=keep-me\n' > "$slotA/.env"+printf 'DB-SENTINEL\n' > "$slotA/backend/instance/serverkit.db"+cp "$slotA/.env" "$t/env.orig"; cp "$slotA/backend/instance/serverkit.db" "$t/db.orig"+ln -sfn "$slotA" "$inst" 2>/dev/null || true+if [ ! -L "$inst" ]; then+ skip "fetch_release live-state carry — symlinks unsupported here (runs on Linux CI)"+else+ stage="$t/stage"; mkdir -p "$stage/serverkit/backend/app" "$stage/serverkit/scripts"+ printf '#!/bin/sh\n' > "$stage/serverkit/serverkit"+ printf '9.9.9\n' > "$stage/serverkit/VERSION"+ tar czf "$t/rel.tar.gz" -C "$stage" serverkit+ if (+ set -Eeuo pipefail+ INSTALL_DIR="$inst"; DIR_A="$slotA"; DIR_B="$slotB"; FIRST_SLOT="$slotA"+ SERVERKIT_OFFLINE_TARBALL="$t/rel.tar.gz"+ fetch_release+ cmp -s "$slotA/.env" "$t/env.orig"+ cmp -s "$slotA/backend/instance/serverkit.db" "$t/db.orig"+ [ -d "$slotA/backend/app" ]+ ) >/dev/null 2>&1; then+ ok "fetch_release re-run preserves .env + database byte-identical (the data-loss bug)"+ else+ bad "fetch_release re-run destroyed or altered .env / the database (I4 regression)"+ fi+fi++# --------------------------------------------------------------------------+# T17 — I20: snapshot_existing must refresh a stale .backup instead of+# silently keeping a months-old tree (which revert_install would then+# restore over a much newer install).+# --------------------------------------------------------------------------+t="$WORK/t17"; mkdir -p "$t/inst"+printf 'v1\n' > "$t/inst/marker"+( set -Eeuo pipefail; INSTALL_DIR="$t/inst"; snapshot_existing ) >/dev/null 2>&1+if [ -f "$t/inst.backup/marker" ] && grep -q v1 "$t/inst.backup/marker"; then+ ok "snapshot_existing creates the first backup"+else+ bad "snapshot_existing did not create a backup"+fi+printf 'v2\n' > "$t/inst/marker"+( set -Eeuo pipefail; INSTALL_DIR="$t/inst"; snapshot_existing ) >/dev/null 2>&1+if grep -q v2 "$t/inst.backup/marker" 2>/dev/null && [ ! -d "$t/inst.backup.new" ]; then+ ok "snapshot_existing refreshes a stale .backup on re-run (no months-old rollback source)"+else+ bad "snapshot_existing kept the stale backup (I20 regression) or left .backup.new behind"+fi++# --------------------------------------------------------------------------+# T18 — I3: the release-venv install must not delete its own copy source. In+# the default layout $VENV_DIR resolves to $FIRST_SLOT/venv, so the old+# rm -rf + cp deleted the venv then failed to copy it. Same-path → keep in+# place; distinct path → still copies.+# --------------------------------------------------------------------------+t="$WORK/t18"; mkdir -p "$t/slot/venv/bin" "$t/elsewhere"+: > "$t/slot/venv/bin/activate"+printf '#!/bin/sh\n' > "$t/slot/venv/bin/python"; chmod +x "$t/slot/venv/bin/python"+if (+ set -Eeuo pipefail+ INSTALL_FROM_RELEASE=1; FIRST_SLOT="$t/slot"; VENV_DIR="$t/slot/venv"+ build_virtualenv+ [ -f "$t/slot/venv/bin/activate" ]+) >/dev/null 2>&1; then+ ok "build_virtualenv keeps the venv when source and destination are the same directory"+else+ bad "build_virtualenv deleted its own venv on the same-path layout (I3 regression)"+fi+if (+ set -Eeuo pipefail+ INSTALL_FROM_RELEASE=1; FIRST_SLOT="$t/slot"; VENV_DIR="$t/elsewhere/venv"+ build_virtualenv+ [ -f "$t/elsewhere/venv/bin/activate" ] && [ -f "$t/slot/venv/bin/activate" ]+) >/dev/null 2>&1; then+ ok "build_virtualenv still copies the release venv to a distinct VENV_DIR"+else+ bad "build_virtualenv failed to copy the release venv to a custom VENV_DIR"+fi+ln -sfn "$t/slot" "$t/link" 2>/dev/null || true+if [ ! -L "$t/link" ]; then+ skip "build_virtualenv symlinked default layout — symlinks unsupported here (runs on Linux CI)"+else+ if (+ set -Eeuo pipefail+ INSTALL_FROM_RELEASE=1; FIRST_SLOT="$t/slot"; VENV_DIR="$t/link/venv"+ build_virtualenv+ [ -x "$t/slot/venv/bin/python" ]+ ) >/dev/null 2>&1; then+ ok "build_virtualenv survives the default symlinked layout (/opt/serverkit → slot)"+ else+ bad "build_virtualenv broke on the symlinked layout (I3 regression)"+ fi+fi++# --------------------------------------------------------------------------+# T19 — I6/I12: pkg_add promises warn-and-continue, so a failing package+# manager must produce a warning and exit 0 — the old body died on the+# out=$(...) assignment under set -e, printing NOTHING. refresh_pkg_index is+# best-effort too, and its yum arm must not use the dnf-only --refresh flag.+# --------------------------------------------------------------------------+t="$WORK/t19"; mkdir -p "$t/bin"+printf '#!/usr/bin/env bash\necho "E: Unable to locate package"\nexit 100\n' > "$t/bin/apt-get"+chmod +x "$t/bin/apt-get"+out="$( set -Eeuo pipefail; PATH="$t/bin:$PATH"; PKG_MGR=apt pkg_add nosuchpkg 2>&1 )"+rc=$?+if [ "$rc" -eq 0 ] && printf '%s' "$out" | grep -q 'Could not install: nosuchpkg'; then+ ok "pkg_add warns and returns 0 when the package manager fails (errexit-immune body)"+else+ bad "pkg_add rc=$rc (expected 0) or printed no warning: [$out]"+fi+if ( set -Eeuo pipefail; PATH="$t/bin:$PATH"; PKG_MGR=apt refresh_pkg_index ) >/dev/null 2>&1; then+ ok "refresh_pkg_index is best-effort (a failed index refresh never aborts)"+else+ bad "refresh_pkg_index propagated a failure under set -e (I6 regression)"+fi+if ! grep -qE '^[^#]*yum makecache --refresh' "$INSTALL_SH"; then+ ok "refresh_pkg_index yum arm avoids the dnf-only --refresh flag (I12)"+else+ bad "the yum arm uses 'makecache --refresh' again — that flag is dnf-only (I12 regression)"+fi++# --------------------------------------------------------------------------+# T20 — I5: the RHEL-9 upfront openssh+openssl paired upgrade (drops sshd+# 'OpenSSL version mismatch' on Rocky 9) must run the exact paired+# transaction, survive a failing dnf, and no-op on other families.+# --------------------------------------------------------------------------+t="$WORK/t20"; mkdir -p "$t/bin"+DNF_LOG="$t/dnf.log"; : > "$DNF_LOG"+cat > "$t/bin/dnf" <<EOF+#!/usr/bin/env bash+printf '%s\n' "\$*" >> "$DNF_LOG"+exit 1+EOF+chmod +x "$t/bin/dnf"+if ( set -Eeuo pipefail; PATH="$t/bin:$PATH"; OS_FAMILY=rhel upgrade_rhel_crypto_stack ) >/dev/null 2>&1 \+ && grep -q 'upgrade -y openssh openssh-server openssh-clients openssl openssl-libs openssl-devel' "$DNF_LOG"; then+ ok "upgrade_rhel_crypto_stack upgrades openssh+openssl in ONE transaction and survives a failing dnf"+else+ bad "upgrade_rhel_crypto_stack missing/wrong dnf transaction or aborted (I5 regression)"+fi+: > "$DNF_LOG"+( set -Eeuo pipefail; PATH="$t/bin:$PATH"; OS_FAMILY=debian upgrade_rhel_crypto_stack ) >/dev/null 2>&1+if [ ! -s "$DNF_LOG" ]; then+ ok "upgrade_rhel_crypto_stack is a no-op outside the RHEL family"+else+ bad "upgrade_rhel_crypto_stack ran dnf on a non-RHEL family"+fi++# --------------------------------------------------------------------------+# T21 — I7: sync_source needs git but nothing installed it. Without git and+# with the on-demand install failing, sync_source must halt LOUDLY (clear+# message) before ever attempting the clone. Curated PATH: a failing apt-get+# plus a real tail; no git anywhere.+# --------------------------------------------------------------------------+t="$WORK/t21"; mkdir -p "$t/bin"+printf '#!/usr/bin/env bash\nexit 1\n' > "$t/bin/apt-get"; chmod +x "$t/bin/apt-get"+printf '#!/usr/bin/env bash\nexec %s "$@"\n' "$(command -v tail)" > "$t/bin/tail"; chmod +x "$t/bin/tail"+out="$( set -Eeuo pipefail; PATH="$t/bin"; PKG_MGR=apt sync_source 2>&1 )"+rc=$?+if [ "$rc" -ne 0 ] && printf '%s' "$out" | grep -q 'git is required'; then+ ok "sync_source installs git on demand and halts clearly when that fails (never a cryptic clone error)"+else+ bad "sync_source rc=$rc without the 'git is required' halt: [$out]"+fi++# --------------------------------------------------------------------------+# T22 — I8: locate_python must reject an interpreter whose venv module is+# broken (Debian/Ubuntu minimal without pythonX.Y-venv → 'ensurepip is not+# available' much later), and on Debian-family it must first try to install+# the matching pythonX.Y-venv package.+# --------------------------------------------------------------------------+t="$WORK/t22"; PYX="$t/py"; mkdir -p "$PYX" "$t/bin"+mkvless() { # mkvless <name> <major.minor> — version OK, `-m venv` broken+ {+ printf '#!/usr/bin/env bash\n'+ printf 'if [ "${1:-}" = "-c" ]; then printf "%s"; exit 0; fi\n' "$2"+ printf 'if [ "${1:-}" = "-m" ]; then exit 1; fi\n'+ printf 'exit 0\n'+ } > "$PYX/$1"+ chmod +x "$PYX/$1"+}+mkvless python3.12 3.13 # out of range — rejected before the venv probe+mkvless python3.11 3.11 # in range but venv-less — must be rejected+mkvless python3 3.10 # too old+if ( set -Eeuo pipefail; PATH="$PYX:$PATH"; locate_python ) >/dev/null 2>&1; then+ bad "locate_python accepted a venv-less interpreter (I8 regression)"+else+ ok "locate_python rejects an interpreter whose '-m venv' is broken"+fi+APT_LOG="$t/apt.log"; : > "$APT_LOG"+cat > "$t/bin/apt-get" <<EOF+#!/usr/bin/env bash+printf '%s\n' "\$*" >> "$APT_LOG"+exit 1+EOF+chmod +x "$t/bin/apt-get"+( set -Eeuo pipefail; PATH="$t/bin:$PYX:$PATH"; OS_FAMILY=debian PKG_MGR=apt locate_python ) >/dev/null 2>&1+if grep -q 'install -y python3.11-venv' "$APT_LOG"; then+ ok "locate_python (Debian family) tries to install the matching pythonX.Y-venv package"+else+ bad "locate_python never attempted the python3.11-venv fallback; apt saw: $(tr '\n' ';' < "$APT_LOG")"+fi++# --------------------------------------------------------------------------+# T23 — I9: distro-repo Docker (docker.io) has no Docker-repo+# 'docker-compose-plugin'; ensure_compose_plugin must fall back to Ubuntu's+# 'docker-compose-v2' and, when neither lands, warn and continue instead of+# aborting the install.+# --------------------------------------------------------------------------+t="$WORK/t23"; mkdir -p "$t/bin"+APT_LOG="$t/apt.log"; : > "$APT_LOG"+printf '#!/usr/bin/env bash\nexit 1\n' > "$t/bin/docker"; chmod +x "$t/bin/docker"+cat > "$t/bin/apt-get" <<EOF+#!/usr/bin/env bash+printf '%s\n' "\$*" >> "$APT_LOG"+exit 100+EOF+chmod +x "$t/bin/apt-get"+out="$( set -Eeuo pipefail; PATH="$t/bin:$PATH"; PKG_MGR=apt ensure_compose_plugin 2>&1 )"+rc=$?+if [ "$rc" -eq 0 ] && grep -q 'install -y docker-compose-plugin' "$APT_LOG" \+ && grep -q 'install -y docker-compose-v2' "$APT_LOG" \+ && printf '%s' "$out" | grep -q 'docker compose'; then+ ok "ensure_compose_plugin tries plugin → docker-compose-v2 → warn-and-continue (never aborts)"+else+ bad "ensure_compose_plugin rc=$rc; apt saw: $(tr '\n' ';' < "$APT_LOG")"+fi++# --------------------------------------------------------------------------+# T24 — I10: dnf5 (Fedora 41+) removed `config-manager --add-repo URL`;+# docker_repo_add must fall back to the dnf5 `addrepo --from-repofile=` form,+# and stay warn-and-continue when both forms fail.+# --------------------------------------------------------------------------+t="$WORK/t24"; mkdir -p "$t/bin"+DNF_LOG="$t/dnf.log"; : > "$DNF_LOG"+cat > "$t/bin/dnf" <<EOF+#!/usr/bin/env bash+printf '%s\n' "\$*" >> "$DNF_LOG"+case "\$*" in+ "config-manager --add-repo "*) exit 1 ;; # dnf5: flag removed+ "config-manager addrepo "*) exit 0 ;;+esac+exit 0+EOF+chmod +x "$t/bin/dnf"+if ( set -Eeuo pipefail; PATH="$t/bin:$PATH"+ docker_repo_add https://example.invalid/docker-ce.repo ) >/dev/null 2>&1 \+ && grep -q 'config-manager addrepo --from-repofile=https://example.invalid/docker-ce.repo' "$DNF_LOG"; then+ ok "docker_repo_add falls back to the dnf5 addrepo syntax when --add-repo fails"+else+ bad "docker_repo_add never tried the dnf5 syntax; dnf saw: $(tr '\n' ';' < "$DNF_LOG")"+fi+printf '#!/usr/bin/env bash\nexit 1\n' > "$t/bin/dnf"; chmod +x "$t/bin/dnf"+if ( set -Eeuo pipefail; PATH="$t/bin:$PATH"+ docker_repo_add https://example.invalid/docker-ce.repo ) >/dev/null 2>&1; then+ ok "docker_repo_add warns and continues when both config-manager forms fail"+else+ bad "docker_repo_add aborted when both config-manager forms failed (must warn-and-continue)"+fi++# --------------------------------------------------------------------------+# T25 — I21: a failing state_set/state_append (e.g. unwritable+# install-state.json) must not abort the firewall phase whose actual firewall+# work already succeeded. The state file is pointed under a regular FILE so+# state.sh's python genuinely fails; ufw is stubbed so firewall_open runs.+# --------------------------------------------------------------------------+if command -v python3 >/dev/null 2>&1; then+ t="$WORK/t25"; mkdir -p "$t/bin"+ printf '#!/usr/bin/env bash\nexit 0\n' > "$t/bin/ufw"; chmod +x "$t/bin/ufw"+ : > "$t/blocker" # regular file — state.json can't live under it+ out="$(+ set -Eeuo pipefail+ export PATH="$t/bin:$PATH"+ export SERVERKIT_STATE_FILE="$t/blocker/state.json"+ FIREWALL_BACKEND=ufw configure_firewall 2>&1+ )"+ rc=$?+ if [ "$rc" -eq 0 ] && printf '%s' "$out" | grep -q 'Firewall configured'; then+ ok "configure_firewall survives failing state_set/state_append (records are best-effort)"+ else+ bad "configure_firewall rc=$rc when state writes fail (I21 regression): [$out]"+ fi+else+ skip "configure_firewall state-failure guard — python3 unavailable here"+fi++# --------------------------------------------------------------------------+# T26 — regression guards baked into the source (same spirit as+# test_update.sh T26): the audited failure shapes must never creep back in.+# --------------------------------------------------------------------------+guards_ok=1+if grep -qE '^[^#]*backend/src' "$INSTALL_SH"; then+ bad "install.sh probes backend/src again — that path never exists (I4 detection regression)"+ guards_ok=0+fi+if ! grep -qE '^[[:space:]]*upgrade_rhel_crypto_stack$' "$INSTALL_SH"; then+ bad "main() no longer calls upgrade_rhel_crypto_stack (I5 regression)"+ guards_ok=0+fi+[ "$guards_ok" = "1" ] && ok "source guards: no backend/src probe; main runs the RHEL crypto upgrade"+# --------------------------------------------------------------------------printf '\n%d passed, %d failed, %d skipped\n\n' "$PASS" "$FAIL" "$SKIP"[ "$FAIL" -eq 0 ]