test(patroni): end-to-end failover, recovery & data-integrity suite

.github/workflows/ci.yml

@@ -459,6 +459,28 @@ jobs:
           PG_HARDSTORAGE_TEST_PG_MAJOR: ${{ matrix.pg }}
         run: make test-integration
 
+  test-patroni:
+    # Heavy Patroni failover / recovery / data-integrity suite: each test
+    # stands up a real 3-node Spilo/Patroni cluster, so the whole set runs
+    # well past the per-PR integration budget and lives in its own lane with
+    # a generous timeout. The tests are PostgreSQL-version independent (the
+    # Spilo image pins its own PG), so a single cell suffices rather than the
+    # 4-way matrix.
+    name: test (patroni failover)
+    runs-on: ubuntu-24.04
+    needs: [vet]
+    timeout-minutes: 45
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version: ${{ env.GO_VERSION }}
+          cache: true
+      # Docker is preinstalled on ubuntu-24.04 runners; the topology brings
+      # up etcd + 3 Spilo nodes via `docker compose`.
+      - name: patroni failover suite
+        run: make test-patroni
+
   lint:
     name: golangci-lint
     runs-on: ubuntu-24.04

Makefile

@@ -340,6 +340,17 @@ $(HS_TMPDIR):
 test-integration: | $(HS_TMPDIR)
 	go test -tags=integration -race -count=1 -timeout=$(INTEGRATION_TIMEOUT) $(GO_PKGS)
 
+# Heavy Patroni failover / recovery / data-integrity suite: each test
+# stands up a real 3-node Spilo/Patroni cluster, so the whole set runs far
+# longer than the per-PR integration budget and lives behind an extra
+# `patroni` build tag (see internal/testkit/topology/*patroni_failover*).
+# Runs in its own CI lane, once (the tests are PG-version independent).
+# No -race: these are Docker-orchestration tests with little in-process
+# concurrency, and -race roughly doubles their already-long wallclock.
+PATRONI_TIMEOUT ?= 30m
+test-patroni: | $(HS_TMPDIR)
+	go test -tags='integration patroni' -count=1 -timeout=$(PATRONI_TIMEOUT) ./internal/testkit/topology/...
+
 # Mutation harness. Loops over internal/testkit/mutation/Registry and
 # runs `go test -tags=<mutation-tag>` against each affected package,
 # asserting the suite catches the deliberate regression. A failure