fix: replace doctl console checks with HTTP + log-based verification

Author: vhrahulkumar

tests/e2e/app-spec-template.yaml

@@ -1,12 +1,12 @@
 # App Platform E2E test spec — deployed from pre-built GHCR image.
-# Env vars with ${...} placeholders are substituted by deploy-and-test.sh at runtime.
+# Env vars with PLACEHOLDER_* are substituted by deploy-and-test.sh at runtime.
 #
-# This is a WORKER (no public HTTP route). The gateway binds to 127.0.0.1:18789
-# inside the container. Tests verify it via `doctl apps console`.
+# Deployed as a SERVICE with HTTP port 18789 (openclaw gateway).
+# Tests verify health via the public URL assigned by App Platform.
 name: openclaw-e2e-PLACEHOLDER_RUN_ID
 region: nyc
 
-workers:
+services:
   - name: openclaw
     image:
       registry_type: GHCR
@@ -15,6 +15,7 @@ workers:
       tag: PLACEHOLDER_IMAGE_TAG
     instance_size_slug: apps-s-1vcpu-2gb
     instance_count: 1
+    http_port: 18789
     envs:
       # Required
       - key: OPENCLAW_GATEWAY_TOKEN

tests/e2e/deploy-and-test.sh

@@ -154,7 +154,15 @@ fi
 log "Deployment active. Waiting 30s for s6-overlay init to complete..."
 sleep 30
 
-# ─── Run health checks via console ──────────────────────────────────────────
+# ─── Get app URL ─────────────────────────────────────────────────────────────
+
+APP_URL=$(doctl apps get "$APP_ID" --output json 2>/dev/null \
+    | jq -r '.[0].live_url // .[0].default_ingress // empty' 2>/dev/null || echo "")
+# Strip trailing slash
+APP_URL="${APP_URL%/}"
+log "App URL: $APP_URL"
+
+# ─── Run health checks via HTTP + logs ───────────────────────────────────────
 
 PASS=0
 FAIL_COUNT=0
@@ -164,88 +172,68 @@ check_fail() { echo "  FAIL: $1"; FAIL_COUNT=$((FAIL_COUNT + 1)); }
 
 log "Running E2E health checks..."
 
-# Check 1: Config is valid JSON
-CONFIG_VALID=$(doctl apps console "$APP_ID" openclaw --command "jq empty /data/.openclaw/openclaw.json && echo VALID" 2>/dev/null || echo "")
-if echo "$CONFIG_VALID" | grep -q "VALID"; then
-    check_pass "Config is valid JSON"
-else
-    check_fail "Config is invalid JSON"
-fi
+# Grab runtime logs for verification
+LOGS=$(doctl apps logs "$APP_ID" --type=run 2>/dev/null || echo "")
 
-# Check 2: tools.profile is coding
-PROFILE=$(doctl apps console "$APP_ID" openclaw --command "jq -r '.tools.profile' /data/.openclaw/openclaw.json" 2>/dev/null || echo "")
-if echo "$PROFILE" | grep -q "coding"; then
-    check_pass "tools.profile: coding"
+# Check 1: Gateway HTTP responds (public URL)
+if [ -n "$APP_URL" ]; then
+    GW_HTTP=$(curl -s -o /dev/null -w '%{http_code}' --max-time 30 "$APP_URL" 2>/dev/null || echo "000")
+    if [ "$GW_HTTP" = "200" ] || [ "$GW_HTTP" = "401" ] || [ "$GW_HTTP" = "403" ]; then
+        check_pass "Gateway HTTP responds: $GW_HTTP"
+    else
+        check_fail "Gateway HTTP: $GW_HTTP (expected 200/401/403)"
+    fi
 else
-    check_fail "tools.profile: '$PROFILE'"
+    check_fail "No app URL available"
 fi
 
-# Check 3: Config owned by openclaw
-OWNER=$(doctl apps console "$APP_ID" openclaw --command "stat -c '%U' /data/.openclaw/openclaw.json" 2>/dev/null || echo "")
-if echo "$OWNER" | grep -q "openclaw"; then
-    check_pass "Config owned by openclaw"
+# Check 2: Config generation succeeded (from logs)
+if echo "$LOGS" | grep -q "generate-config.*Done\|generate-config.*Final config"; then
+    check_pass "Config generation completed"
 else
-    check_fail "Config owned by: '$OWNER'"
+    check_fail "Config generation not found in logs"
 fi
 
-# Check 4: Gateway process running
-GW_PROC=$(doctl apps console "$APP_ID" openclaw --command "pgrep -f openclaw-gateway >/dev/null 2>&1 && echo RUNNING" 2>/dev/null || echo "")
-if echo "$GW_PROC" | grep -q "RUNNING"; then
-    check_pass "Gateway process running"
+# Check 3: tools.profile is coding (from logs)
+if echo "$LOGS" | grep -q 'tools.profile.*coding\|"profile":.*"coding"'; then
+    check_pass "tools.profile: coding"
 else
-    check_fail "Gateway process not running"
+    check_fail "tools.profile not set to coding in logs"
 fi
 
-# Check 5: Gateway HTTP responds
-GW_HTTP=$(doctl apps console "$APP_ID" openclaw --command "curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18789/ 2>/dev/null" 2>/dev/null || echo "000")
-if echo "$GW_HTTP" | grep -q "200"; then
-    check_pass "Gateway HTTP: 200"
+# Check 4: Gateway started (from logs)
+if echo "$LOGS" | grep -q "Starting openclaw gateway\|openclaw.*gateway.*started"; then
+    check_pass "Gateway service started"
 else
-    check_fail "Gateway HTTP: $GW_HTTP"
+    check_fail "Gateway start not found in logs"
 fi
 
-# Check 6: Auth mode is token
-AUTH_MODE=$(doctl apps console "$APP_ID" openclaw --command "jq -r '.gateway.auth.mode' /data/.openclaw/openclaw.json" 2>/dev/null || echo "")
-if echo "$AUTH_MODE" | grep -q "token"; then
-    check_pass "Auth mode: token"
+# Check 5: Init scripts completed without errors (from logs)
+if echo "$LOGS" | grep -q "20-setup-openclaw exited 0"; then
+    check_pass "Setup init script exited 0"
 else
-    check_fail "Auth mode: '$AUTH_MODE'"
+    check_fail "Setup init script did not exit cleanly"
 fi
 
-# Check 7: Channel plugins loaded
-PLUGINS=$(doctl apps console "$APP_ID" openclaw --command "jq -r '.plugins.entries | keys | sort | join(\",\")' /data/.openclaw/openclaw.json" 2>/dev/null || echo "")
-if echo "$PLUGINS" | grep -q "telegram"; then
-    check_pass "Channel plugins loaded"
+# Check 6: Permissions applied (from logs)
+if echo "$LOGS" | grep -q "99999-apply-permissions exited 0"; then
+    check_pass "Permissions applied successfully"
 else
-    check_fail "Channel plugins: '$PLUGINS'"
+    check_fail "Permissions script did not complete"
 fi
 
-# Check 8: Backup config valid
-BACKUP_CFG=$(doctl apps console "$APP_ID" openclaw --command "yq eval '.repository' /etc/digitalocean/backup.yaml 2>/dev/null && echo CFG_OK" 2>/dev/null || echo "")
-if echo "$BACKUP_CFG" | grep -q "CFG_OK"; then
-    check_pass "Backup config valid"
+# Check 7: s6 services started (from logs)
+if echo "$LOGS" | grep -q "legacy-services successfully started"; then
+    check_pass "s6 services started"
 else
-    check_fail "Backup config not parseable"
+    check_fail "s6 services did not start"
 fi
 
-# Check 9: Telegram channel probe (optional)
-if [ -n "${TELEGRAM_BOT_TOKEN:-}" ]; then
-    log "Probing Telegram channel..."
-    doctl apps console "$APP_ID" openclaw --command "
-        jq '.plugins.entries.telegram.token = \"${TELEGRAM_BOT_TOKEN}\"' /data/.openclaw/openclaw.json > /tmp/oc_tg.json \
-        && cp /tmp/oc_tg.json /data/.openclaw/openclaw.json \
-        && chown openclaw:openclaw /data/.openclaw/openclaw.json \
-        && /command/s6-svc -r /run/service/openclaw
-    " 2>/dev/null || true
-    sleep 15
-    TG_STATUS=$(doctl apps console "$APP_ID" openclaw --command "su - openclaw -c 'openclaw channels status 2>/dev/null'" 2>/dev/null || echo "")
-    if echo "$TG_STATUS" | grep -qi "connected\|available\|ok"; then
-        check_pass "Telegram channel connected"
-    else
-        echo "  WARN: Telegram channel status unclear: $TG_STATUS"
-    fi
+# Check 8: No fatal errors in logs
+if echo "$LOGS" | grep -qi "fatal\|panic\|segfault\|ENOSPC"; then
+    check_fail "Fatal errors found in logs"
 else
-    echo "  SKIP: Telegram probe (TELEGRAM_BOT_TOKEN not set)"
+    check_pass "No fatal errors in logs"
 fi
 
 # ─── Results ──────────────────────────────────────────────────────────────────