The workflow was hitting the metrics endpoint too soon/from the host. I switched the metrics check to run inside the loader container (which now has curl installed) against http://postgres:9911/metrics. This avoids host-network timing issues.

Ensured the loader installs curl so that internal metrics check works.
Files changed:

demo/postgres-survive-kill9/docker-compose.yml: loader now installs curl alongside the postgres client.
demo/postgres-survive-kill9/ci-smoke.sh: metrics wait uses docker compose exec -T loader curl http://postgres:9911/metrics and still waits up to ~4 minutes for readiness/tables/metrics.

Author: Shane Wall

demo/postgres-survive-kill9/ci-smoke.sh

@@ -40,7 +40,7 @@ wait_pgbench_tables() {
 wait_metrics() {
   echo "Waiting for metrics endpoint..."
   for _ in {1..120}; do
-    if curl -sf http://localhost:9911/metrics | grep -q diffkeeper_recovery_total; then
+    if docker compose exec -T loader curl -sf http://postgres:9911/metrics | grep -q diffkeeper_recovery_total; then
       return
     fi
     sleep 2

demo/postgres-survive-kill9/docker-compose.yml

@@ -27,7 +27,7 @@ services:
       -c "
         set -euo pipefail;
         apt-get update -qq;
-        apt-get install -y -qq postgresql-client postgresql-contrib;
+        apt-get install -y -qq postgresql-client postgresql-contrib curl;
         until pg_isready -h "$$PGHOST" -U "$$PGUSER"; do sleep 1; done;
         createdb "$$PGDATABASE" || true;
         if ! psql -h "$$PGHOST" -U "$$PGUSER" -d "$$PGDATABASE" -tAc \"SELECT 1 FROM pg_tables WHERE tablename='pgbench_accounts'\" | grep -q 1; then