| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "testing" |
| 7 | "time" |
| 8 | |
| 9 | "github.com/carlosframework/platform/internal/activator" |
| 10 | "github.com/carlosframework/platform/internal/hostbox" |
| 11 | "github.com/carlosframework/platform/internal/hostclaim" |
| 12 | "github.com/carlosframework/platform/internal/hostguard" |
| 13 | ) |
| 14 | |
| 15 | // plan9Fix1StopAtCommitted admits a probe at probing through the real |
| 16 | // runtime, then moves the stored op to probed and committed the way the moves |
| 17 | // controller does, under KeepWriterSession and without touching the attempt. |
| 18 | // The attempt's checkpoint still names probing. |
| 19 | func plan9Fix1StopAtCommitted(t *testing.T) (*recoveryFixture, *recoveryController, *claimWakeAttempt, hostclaim.Observed) { |
| 20 | t.Helper() |
| 21 | probe := activator.StartPolicy{MaxStarts: 5, Window: 30 * time.Second, Probe: true} |
| 22 | f, c, _, _, admitted := plan9D2Landing(t, hostclaim.PhaseLandingProbing, probe, true) |
| 23 | a := c.wake.get(recoveryHost) |
| 24 | if a == nil { |
| 25 | t.Fatal("no pending attempt after admission") |
| 26 | } |
| 27 | for _, phase := range []string{hostclaim.PhaseLandingProbed, hostclaim.PhaseLandingCommitted} { |
| 28 | if _, err := hostclaim.Update(t.Context(), f.claims, f.read(), hostclaim.KeepWriterSession("m1", func(c *hostclaim.Claim) error { |
| 29 | c.Op.Phase = phase |
| 30 | return nil |
| 31 | })); err != nil { |
| 32 | t.Fatalf("move to %s: %v", phase, err) |
| 33 | } |
| 34 | } |
| 35 | if checkpoint, _ := a.snapshot(); checkpoint.Claim.Op.Phase != hostclaim.PhaseLandingProbing { |
| 36 | t.Fatalf("checkpoint moved to %q", checkpoint.Claim.Op.Phase) |
| 37 | } |
| 38 | return f, c, a, admitted |
| 39 | } |
| 40 | |
| 41 | // R1: identifying a session in order to stop it grants no write authority. |
| 42 | // A probe admitted at probing whose op the moves controller took to committed |
| 43 | // is still stopped as itself by every teardown path: root fences and stops |
| 44 | // that exact session at committed, the signed receipt verifies, and the |
| 45 | // pending lease is released so the host's next admission is not refused. |
| 46 | func TestPlan9Fix1TeardownStopsAProbeThatReachedCommitted(t *testing.T) { |
| 47 | t.Parallel() |
| 48 | for name, teardown := range map[string]func(context.Context, *recoveryController, hostclaim.Observed, hostclaim.Observed) error{ |
| 49 | "stop": func(ctx context.Context, c *recoveryController, checkpoint, _ hostclaim.Observed) error { |
| 50 | return c.wake.Stop(ctx, checkpoint) |
| 51 | }, |
| 52 | "close": func(ctx context.Context, c *recoveryController, _, _ hostclaim.Observed) error { |
| 53 | return c.wake.Close(ctx) |
| 54 | }, |
| 55 | "failed-admission": func(ctx context.Context, c *recoveryController, _, admitted hostclaim.Observed) error { |
| 56 | return c.runtime.FailedAdmission(ctx, admitted, errors.New("restore failed")) |
| 57 | }, |
| 58 | } { |
| 59 | t.Run(name, func(t *testing.T) { |
| 60 | f, c, a, admitted := plan9Fix1StopAtCommitted(t) |
| 61 | ctx := t.Context() |
| 62 | checkpoint, _ := a.snapshot() |
| 63 | if err := teardown(ctx, c, checkpoint, admitted); err != nil { |
| 64 | t.Fatalf("teardown refused the probe's own session at committed: %v", err) |
| 65 | } |
| 66 | h := admitted.Claim.Holder |
| 67 | if len(f.requests) != 1 { |
| 68 | t.Fatalf("root requests = %d, want 1", len(f.requests)) |
| 69 | } |
| 70 | req := f.requests[0] |
| 71 | if req.Recover || req.Phase != hostclaim.PhaseLandingCommitted || req.OperationID != "m1" || req.WriterSession != h.Session || req.Incarnation != h.Incarnation || req.InstallationStamp != h.Stamp || req.ObligationID != hostguard.WriterStopObligation(h.Session) || req.Box != "box-1" || req.ControllerID != c.local.ControllerID { |
| 72 | t.Fatalf("root was asked to stop another tuple: %+v", req) |
| 73 | } |
| 74 | receipts, err := f.guard.CompletedWriterStops(ctx, recoveryHost, hostguard.Captured{Incarnation: h.Incarnation, Stamp: h.Stamp}, "m1", "box-1") |
| 75 | if err != nil || len(receipts) != 1 || receipts[0].Target.Session != h.Session || receipts[0].Target.Phase != hostclaim.PhaseLandingCommitted || verifyRetirementReceipt(ctx, receipts[0]) != nil { |
| 76 | t.Fatalf("no verified writer-stop receipt for session %d: %+v %v", h.Session, receipts, err) |
| 77 | } |
| 78 | if c.wake.get(recoveryHost) != nil || !a.released || a.ctx.Err() == nil { |
| 79 | t.Fatalf("pending lease kept: pending=%v released=%v", c.wake.get(recoveryHost) != nil, a.released) |
| 80 | } |
| 81 | }) |
| 82 | } |
| 83 | } |
| 84 | |
| 85 | // A teardown never follows a session backward: a checkpoint at committed |
| 86 | // does not stop a claim now back at probing, and root is never asked. |
| 87 | func TestPlan9Fix1TeardownRefusesABackwardPhase(t *testing.T) { |
| 88 | f, c, a, _ := plan9Fix1StopAtCommitted(t) |
| 89 | ctx := t.Context() |
| 90 | committed := f.read() |
| 91 | a.mu.Lock() |
| 92 | a.observed = committed |
| 93 | a.mu.Unlock() |
| 94 | if _, err := hostclaim.Update(ctx, f.claims, committed, func(c *hostclaim.Claim) error { |
| 95 | c.Op.Phase = hostclaim.PhaseLandingProbing |
| 96 | return nil |
| 97 | }); err != nil { |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | if err := c.wake.Stop(ctx, committed); !errors.Is(err, hostbox.ErrChanged) || len(f.requests) != 0 { |
| 101 | t.Fatalf("stop followed the session backward: %v requests=%d", err, len(f.requests)) |
| 102 | } |
| 103 | if err := c.wake.Close(ctx); !errors.Is(err, hostbox.ErrChanged) || len(f.requests) != 0 || c.wake.get(recoveryHost) == nil { |
| 104 | t.Fatalf("close followed the session backward: %v requests=%d pending=%v", err, len(f.requests), c.wake.get(recoveryHost) != nil) |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | func TestPlan9Fix1StopReleasesCheckpointThatAdvancedAfterCapture(t *testing.T) { |
| 109 | f, c, a, _ := plan9Fix1StopAtCommitted(t) |
| 110 | captured, _ := a.snapshot() |
| 111 | // Another successful check advances the pending attempt after teardown |
| 112 | // captured its identity. Root still verifies and stops the same session. |
| 113 | a.mu.Lock() |
| 114 | a.observed = f.read() |
| 115 | a.mu.Unlock() |
| 116 | if err := c.wake.Stop(t.Context(), captured); err != nil { |
| 117 | t.Fatal(err) |
| 118 | } |
| 119 | if len(f.requests) != 1 || c.wake.get(recoveryHost) != nil || !a.released { |
| 120 | t.Fatalf("stopped session kept an advanced checkpoint: requests=%d pending=%v released=%v", len(f.requests), c.wake.get(recoveryHost) != nil, a.released) |
| 121 | } |
| 122 | } |
| 123 | |
amadan