| 1 | //go:build linux |
| 2 | |
| 3 | package activator |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "os/exec" |
| 9 | "sync" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/hostclaim" |
| 14 | "github.com/carlosframework/platform/internal/hostwrite" |
| 15 | "github.com/carlosframework/platform/internal/registry" |
| 16 | ) |
| 17 | |
| 18 | // The start loop of startAdmitted, driven for real (ruling H17, fix round 1). |
| 19 | // |
| 20 | // What can and cannot be driven. RoleLaunch.Start (rolelaunch_linux.go) |
| 21 | // returns ErrClaimIsolationUnavailable unconditionally, until claim Plan 9 |
| 22 | // Group D replaces the by-name refusal with the isolated-root launcher. It |
| 23 | // is a value method on a concrete type, with no seam, so nothing a test can |
| 24 | // supply runs inside it. Two of the loop's hooks run only from inside Start |
| 25 | // or after it: beforeStart (which journals the attempt's first start through |
| 26 | // OpenProbe and so opens the 30 s window) and Ready. Hence the window can |
| 27 | // not be opened, and readiness can not be offered, through this entry point |
| 28 | // today. The loop itself, ahead of and around Start, can be driven: the |
| 29 | // refusal is an ordinary failed start, and these tests use it as one. |
| 30 | // OpenProbe and Ready are tripwires: the day Start stops refusing they fail, |
| 31 | // and the skipped test below is to be written then. |
| 32 | |
| 33 | type loopRun struct { |
| 34 | mu sync.Mutex |
| 35 | reserved []time.Time |
| 36 | failed int |
| 37 | } |
| 38 | |
| 39 | func (r *loopRun) starts() []time.Time { |
| 40 | r.mu.Lock() |
| 41 | defer r.mu.Unlock() |
| 42 | return append([]time.Time(nil), r.reserved...) |
| 43 | } |
| 44 | |
| 45 | func driveLoop(t *testing.T, policy startPolicy) (*loopRun, error) { |
| 46 | t.Helper() |
| 47 | run := &loopRun{} |
| 48 | obs := hostclaim.Observed{Host: "x.example", ETag: "one", Claim: &hostclaim.Claim{Holder: &hostclaim.Holder{Location: "instances/x.example/inc", Session: 1, Generation: "one"}, Op: hostclaim.Op{Kind: "landing", ID: "op", Phase: "probing", Exec: "box"}}} |
| 49 | work, _ := (hostwrite.Config{}).Select(t.Context(), obs.Host, "spawn") |
| 50 | a := &Service{cfg: Config{HostWrites: hostwrite.Config{Enabled: true}, Claims: &ClaimRuntime{Box: "box", Read: func(context.Context, string) (hostclaim.Observed, error) { return obs, nil }, Prepare: func(context.Context, registry.Route, hostclaim.Observed, startPolicy) (ClaimStart, error) { |
| 51 | return ClaimStart{RoleLauncher: &RoleLaunch{}, CurrentAttempt: func() hostclaim.WakeRetryAttempt { return hostclaim.WakeRetryAttempt{} }, Host: obs.Host, Work: work, Identity: hostwrite.Identity{Config: "c", Release: "r"}, |
| 52 | OpenProbe: func(context.Context, time.Time) (time.Time, error) { |
| 53 | t.Fatal("OpenProbe ran: Start no longer refuses, replace the skipped window test") |
| 54 | return time.Time{}, nil |
| 55 | }, |
| 56 | Restore: func(context.Context) error { return nil }, |
| 57 | Materialize: func(context.Context) error { return nil }, |
| 58 | Build: func() *exec.Cmd { return nil }, |
| 59 | Ready: func(context.Context, *exec.Cmd) error { |
| 60 | t.Fatal("Ready ran: Start no longer refuses, replace the skipped window test") |
| 61 | return nil |
| 62 | }, |
| 63 | Reserve: func(context.Context, int) error { |
| 64 | run.mu.Lock() |
| 65 | run.reserved = append(run.reserved, time.Now()) |
| 66 | run.mu.Unlock() |
| 67 | return nil |
| 68 | }, |
| 69 | Failed: func(context.Context, int) error { run.mu.Lock(); run.failed++; run.mu.Unlock(); return nil }, |
| 70 | Healthy: func(context.Context, int, *exec.Cmd) error { return nil }, |
| 71 | Replicate: func(context.Context, string) (Rep, error) { return nil, nil }, |
| 72 | Publish: func(context.Context, hostclaim.Observed) (hostclaim.Observed, error) { return obs, nil }, |
| 73 | Retain: func(Proc, Rep, error) {}, |
| 74 | }, nil |
| 75 | }}}} |
| 76 | return run, a.startAdmitted(t.Context(), &actEntry{host: obs.Host}, obs, policy) |
| 77 | } |
| 78 | |
| 79 | // Five starts, then the loop gives up, with the delay between starts |
| 80 | // doubling: 1, 2, 4 and 8 seconds. The loop's backoff is time.NewTimer |
| 81 | // inline (claimwake.go), with no clock or sleeper to substitute, so this |
| 82 | // reads real time. It runs on a quarter of the production base |
| 83 | // (claimStartBackoff), so 0.25, 0.5, 1 and 2 s, each delay checked to |
| 84 | // within 200 ms, which keeps every step apart from its neighbours: the |
| 85 | // doubling is what is proven, on 3.75 s of waiting rather than 15. |
| 86 | func TestStartAdmittedLoopStopsAfterFiveStartsWithDoublingBackoff(t *testing.T) { |
| 87 | if claimStartBackoff != time.Second { |
| 88 | t.Fatalf("claimStartBackoff = %v: the production schedule is 1, 2, 4, 8 s", claimStartBackoff) |
| 89 | } |
| 90 | base := 250 * time.Millisecond |
| 91 | shrinkClaimStartBackoff(t, base) |
| 92 | run, err := driveLoop(t, startPolicy{MaxStarts: 5, Window: 30 * time.Second, Deadline: time.Now().Add(time.Hour), Probe: true}) |
| 93 | if !errors.Is(err, ErrClaimIsolationUnavailable) || errors.Is(err, ErrAttemptOver) { |
| 94 | t.Fatalf("the last start's refusal should end the loop: %v", err) |
| 95 | } |
| 96 | at := run.starts() |
| 97 | if len(at) != 5 || run.failed != 5 { |
| 98 | t.Fatalf("starts=%d failed=%d, want 5 and 5", len(at), run.failed) |
| 99 | } |
| 100 | for i, want := range []time.Duration{base, 2 * base, 4 * base, 8 * base} { |
| 101 | if gap := at[i+1].Sub(at[i]); gap < want || gap > want+200*time.Millisecond { |
| 102 | t.Errorf("delay before start %d = %v, want %v", i+2, gap, want) |
| 103 | } |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | // The cut deadline ends the attempt while it waits between starts: one |
| 108 | // start, then ErrAttemptOver at the deadline, not after the 1 s backoff. |
| 109 | func TestStartAdmittedLoopCutDeadlineEndsTheBackoffWait(t *testing.T) { |
| 110 | deadline := time.Now().Add(300 * time.Millisecond) |
| 111 | run, err := driveLoop(t, startPolicy{MaxStarts: 5, Window: 30 * time.Second, Deadline: deadline, Probe: true}) |
| 112 | if !errors.Is(err, ErrAttemptOver) { |
| 113 | t.Fatalf("err=%v", err) |
| 114 | } |
| 115 | if n := len(run.starts()); n != 1 { |
| 116 | t.Fatalf("starts=%d, want 1", n) |
| 117 | } |
| 118 | if late := time.Since(deadline); late > 500*time.Millisecond { |
| 119 | t.Fatalf("the attempt ended %v after the deadline", late) |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | // A deadline already passed ends the attempt before any start: Reserve is |
| 124 | // never reached. |
| 125 | func TestStartAdmittedLoopNoStartPastTheCutDeadline(t *testing.T) { |
| 126 | run, err := driveLoop(t, startPolicy{MaxStarts: 5, Window: 30 * time.Second, Deadline: time.Now().Add(-time.Second), Probe: true}) |
| 127 | if !errors.Is(err, ErrAttemptOver) || len(run.starts()) != 0 { |
| 128 | t.Fatalf("err=%v starts=%d", err, len(run.starts())) |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | func TestStartAdmittedLoopWindowAndReadinessAtTheBoundary(t *testing.T) { |
| 133 | t.Skip("blocked on claim Plan 9 Group D: RoleLaunch.Start (rolelaunch_linux.go) returns ErrClaimIsolationUnavailable " + |
| 134 | "unconditionally, so beforeStart/OpenProbe (which opens the 30 s window) and Ready never run through startAdmitted. " + |
| 135 | "When Group D replaces that refusal, drive: window expiry before a start (OpenProbe returns a time older than the window), " + |
| 136 | "readiness at opened+Window and at the cut deadline rejected with ErrAttemptOver, readiness at opened+Window-1s accepted.") |
| 137 | } |
| 138 | |
amadan