post_check: run-to-completion units that exited 0 are ok, not warnings

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KqhjP7nud5gdBg25zCiRk9

deploy.sh

@@ -7,6 +7,7 @@
 #         ./deploy.sh --diff [conf]        # what would change in /etc
 #         ./deploy.sh --post-check [conf]  # read-only live checks (units, br0, LAN, NAT, dnsmasq, radios, uplink); exit 1 on any warning
 #                                    # POSTCHECK_WAIT=<s>: seconds retried probes may wait in total (default 15; 0 = single attempt)
+#         ./deploy.sh --classify-unit <ActiveState> <Result> <ExecMainStatus>  # prints ok|finished|wait|warn (used by post_check; for tests)
 #         ./deploy.sh render <file>        # preview a templated file on stdout
 #         ./deploy.sh --redact             # filter stdin, masking secret values (used by --diff; for tests)
 # Idempotent. Netplan (wlan0 + Starlink NIC = NM-managed WANs; eth0 is a LAN
@@ -131,6 +132,15 @@ if [ "$ENABLE_FRIGATE" = 1 ]; then
 fi
 }
 
+classify_unit() {  # classify_unit <ActiveState> <Result> <ExecMainStatus> -> ok | finished | wait | warn
+    case "$1" in
+        active) echo ok ;;
+        activating|reloading) echo wait ;;
+        inactive) if [ "${2:-}" = success ] && [ "${3:-}" = 0 ]; then echo finished; else echo warn; fi ;;
+        *) echo warn ;;
+    esac
+}
+
 post_check() {  # read-only live checks; every probe is guarded, warns instead of aborting
     set_units
     echo "== post-deploy check =="
@@ -146,10 +156,19 @@ post_check() {  # read-only live checks; every probe is guarded, warns instead o
         pending+=("$u")
     done
     still=("${pending[@]}")
+    local -A verdict=() detail=()
+    local k v as res st ls
     while :; do
         local next=()
         for u in "${still[@]}"; do
-            systemctl is-active --quiet "$u" 2>/dev/null || next+=("$u")
+            as= res= st= ls=
+            while IFS='=' read -r k v; do
+                case "$k" in LoadState) ls=$v ;; ActiveState) as=$v ;; Result) res=$v ;; ExecMainStatus) st=$v ;; esac
+            done < <(systemctl show -p LoadState,ActiveState,Result,ExecMainStatus "$u" 2>/dev/null || true)
+            verdict[$u]=$(classify_unit "$as" "$res" "$st")
+            [ "$ls" = loaded ] || { verdict[$u]=warn; as="${as:-?}(LoadState=${ls:-?})"; }
+            detail[$u]="ActiveState=${as:-?} Result=${res:-?}"
+            case "${verdict[$u]}" in ok|finished) ;; *) next+=("$u") ;; esac
         done
         still=("${next[@]}")
         [ ${#still[@]} -eq 0 ] && break
@@ -157,8 +176,11 @@ post_check() {  # read-only live checks; every probe is guarded, warns instead o
         sleep 1
     done
     for u in "${pending[@]}"; do
-        if [[ " ${still[*]:-} " == *" $u "* ]]; then warn "post-deploy: unit $u is not active"
+        case "${verdict[$u]:-warn}" in
-        else echo "   ok    unit $u active"; fi
+            ok) echo "   ok    unit $u active" ;;
+            finished) echo "   ok    unit $u finished (exit 0)" ;;
+            *) warn "post-deploy: unit $u is not active (${detail[$u]:-state unknown})" ;;
+        esac
     done
     # b. bridge address
     has_br0_addr() { local o; o=$(ip -4 -br addr show dev br0 2>/dev/null) || true; [[ $o == *"$GATEWAY_IP/$GATEWAY_PREFIX"* ]]; }
@@ -240,6 +262,9 @@ case "${1:-}" in
     --redact)   # ./deploy.sh --redact — stdin through redact_secrets (no config, no root)
         redact_secrets
         exit 0 ;;
+    --classify-unit)   # ./deploy.sh --classify-unit <ActiveState> <Result> <ExecMainStatus>
+        classify_unit "${2:-}" "${3:-}" "${4:-}"
+        exit 0 ;;
     --post-check)   # ./deploy.sh --post-check [conf] — live checks, no root, no install; exit 1 on any warning
         load_conf "${2:-$VANLINK_CONF}"
         post_check

tests/check-templates.sh

@@ -184,4 +184,13 @@ elif echo "$pc" | grep -q 'van-nvme-watch'; then bad "--post-check mentions van-
 else ok "--post-check skips van-nvme-watch when ENABLE_NVME_WATCH=0"; fi
 if grep -q 'USB hub' deploy.sh; then bad "stale 'USB hub' text remains in deploy.sh"; else ok "no stale 'USB hub' text in deploy.sh"; fi
 
+# 12. unit-state classification used by post_check (run-to-completion units are not failures)
+for row in "active success 0:ok" "inactive success 0:finished" "failed exit-code 1:warn" \
+           "activating success 0:wait" "inactive exit-code 1:warn" "inactive success 1:warn"; do
+    args=${row%%:*}; want=${row##*:}
+    # shellcheck disable=SC2086
+    got=$(./deploy.sh --classify-unit $args 2>&1) || true
+    if [ "$got" = "$want" ]; then ok "--classify-unit $args -> $want"; else bad "--classify-unit $args gave '$got', want '$want'"; fi
+done
+
 exit "$fail"