Merge pull request 'Phase 5 (3/4): every test on a bus of its own, at the release the mesh runs (hq ADR 0237)' (#97) from feat/a-suite-that-cannot-flake into main

internal/inventory/hold_test.go

@@ -69,16 +69,19 @@ func TestAWaiterPinsNoConnectionAndGivesUp(t *testing.T) {
 			done <- err
 		}()
 	}
-	// While they wait, the pool lends nothing to them for longer than a look.
-	pinned := 0
-	for range 10 {
-		time.Sleep(60 * time.Millisecond)
-		if n := int(pool.Stat().AcquiredConns() - base); n > pinned {
-			pinned = n
+	// While they wait, the pool lends nothing to them for longer than a look: **pinned is held throughout**,
+	// so at some moment of the wait fewer than all of them hold one. Asked of the least seen over many
+	// looks, not of the most: a look that lands while every waiter is mid-query is a busy machine, not a
+	// pinned connection, and on a loaded one (the suite's packages at once, the race detector) it did land.
+	least := waiters
+	for range 100 {
+		time.Sleep(10 * time.Millisecond)
+		if n := int(pool.Stat().AcquiredConns() - base); n < least {
+			least = n
 		}
 	}
-	if pinned >= waiters {
-		t.Fatalf("%d connections were held by %d waiters", pinned, waiters)
+	if least >= waiters {
+		t.Fatalf("%d waiters held %d connections at every look across their wait", waiters, least)
 	}
 	for range waiters {
 		if err := <-done; !errors.Is(err, ErrNodeBusy) || !strings.Contains(err.Error(), "anchor") {

internal/inventory/store_as_run_test.go

@@ -0,0 +1,49 @@
+package inventory
+
+import (
+	"encoding/json"
+	"os"
+	"strings"
+	"testing"
+)
+
+// **The store the tests run is the store the mesh runs** (novox/hq ADR 0227 rule 9, ADR 0237). In a merge
+// check the build seat raises the store of the major release the facts snapshot says the mesh's store
+// runs; this holds the two equal there, so a test passing here is a test passing on that release. Without
+// a snapshot — on a workstation — it says what it ran against and compares nothing.
+func TestTheStoreTheTestsRunIsTheStoreTheMeshRuns(t *testing.T) {
+	inv := ForTest(t)
+	ran, err := inv.ServerVersionForTest(t)
+	if err != nil {
+		t.Fatal(err)
+	}
+	path := os.Getenv("MESH_FACTS")
+	if path == "" {
+		t.Logf("the tests ran the store at %s; with no MESH_FACTS, nothing to compare it with", ran)
+		return
+	}
+	body, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var f struct {
+		Versions struct {
+			Store string `json:"store"`
+		} `json:"versions"`
+	}
+	if err := json.Unmarshal(body, &f); err != nil {
+		t.Fatal(err)
+	}
+	major := func(v string) string { m, _, _ := strings.Cut(strings.TrimSpace(v), "."); return m }
+	if f.Versions.Store != "" && major(ran) != major(f.Versions.Store) {
+		t.Errorf("the tests ran the store at %s and the mesh's store runs %s", ran, f.Versions.Store)
+	}
+}
+
+// ServerVersionForTest is the store's own word for its release.
+func (i *Inventory) ServerVersionForTest(t *testing.T) (string, error) {
+	t.Helper()
+	var v string
+	err := i.store.Pool().QueryRow(t.Context(), `show server_version`).Scan(&v)
+	return v, err
+}