|
| 1 | +package e2e |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "net/http" |
| 6 | + "os" |
| 7 | + "os/exec" |
| 8 | + "testing" |
| 9 | + "time" |
| 10 | + |
| 11 | + "github.com/stretchr/testify/assert" |
| 12 | + "github.com/stretchr/testify/require" |
| 13 | + |
| 14 | + instanceoapi "github.com/kernel/kernel-images/server/lib/oapi" |
| 15 | +) |
| 16 | + |
| 17 | +// TestReadTelemetryEvents starts a headless container with S2 credentials, |
| 18 | +// publishes a known set of events, and reads them back through |
| 19 | +// GET /telemetry/events. It exercises the full archive read path against a real |
| 20 | +// S2 stream rather than the in-memory ring buffer. |
| 21 | +// |
| 22 | +// Skips automatically when S2_BASIN, S2_ACCESS_TOKEN, or S2_STREAM are unset. |
| 23 | +func TestReadTelemetryEvents(t *testing.T) { |
| 24 | + basin := os.Getenv("S2_BASIN") |
| 25 | + accessToken := os.Getenv("S2_ACCESS_TOKEN") |
| 26 | + stream := os.Getenv("S2_STREAM") |
| 27 | + if basin == "" || accessToken == "" || stream == "" { |
| 28 | + t.Skip("S2_BASIN, S2_ACCESS_TOKEN, and S2_STREAM must be set to run this test") |
| 29 | + } |
| 30 | + |
| 31 | + if _, err := exec.LookPath("docker"); err != nil { |
| 32 | + t.Skipf("docker not available: %v", err) |
| 33 | + } |
| 34 | + |
| 35 | + ctx, cancel := context.WithTimeout(context.Background(), 3*time.Minute) |
| 36 | + defer cancel() |
| 37 | + |
| 38 | + c := NewTestContainer(t, headlessImage) |
| 39 | + require.NoError(t, c.Start(ctx, ContainerConfig{ |
| 40 | + Env: map[string]string{ |
| 41 | + "S2_BASIN": basin, |
| 42 | + "S2_ACCESS_TOKEN": accessToken, |
| 43 | + "S2_STREAM": stream, |
| 44 | + }, |
| 45 | + }), "failed to start container") |
| 46 | + defer c.Stop(ctx) |
| 47 | + |
| 48 | + require.NoError(t, c.WaitReady(ctx), "api not ready") |
| 49 | + |
| 50 | + client, err := c.APIClient() |
| 51 | + require.NoError(t, err) |
| 52 | + |
| 53 | + // Start a telemetry session. The default config enables the system and |
| 54 | + // connection categories, which is what we publish into below. |
| 55 | + startResp, err := client.PutTelemetryWithResponse(ctx, instanceoapi.PutTelemetryJSONRequestBody{}) |
| 56 | + require.NoError(t, err) |
| 57 | + require.Equal(t, http.StatusCreated, startResp.StatusCode(), "put telemetry: %s", string(startResp.Body)) |
| 58 | + |
| 59 | + // Publish a deterministic set of events across two enabled categories. |
| 60 | + const systemCount, connectionCount = 3, 2 |
| 61 | + for i := 0; i < systemCount; i++ { |
| 62 | + publishEvent(t, ctx, client, "test.system", instanceoapi.PublishEventRequestCategorySystem) |
| 63 | + } |
| 64 | + for i := 0; i < connectionCount; i++ { |
| 65 | + publishEvent(t, ctx, client, "test.connection", instanceoapi.PublishEventRequestCategoryConnection) |
| 66 | + } |
| 67 | + |
| 68 | + // Give the storage writer time to flush to S2 (batcher linger + network). |
| 69 | + time.Sleep(2 * time.Second) |
| 70 | + |
| 71 | + // Bound every read tightly: a correct handler caps the S2 read at the tail, |
| 72 | + // so these return promptly. A hang here means the read is unbounded. |
| 73 | + readCtx, readCancel := context.WithTimeout(ctx, 10*time.Second) |
| 74 | + defer readCancel() |
| 75 | + |
| 76 | + // Full read returns at least everything we published. |
| 77 | + all, err := client.ReadTelemetryEventsWithResponse(readCtx, &instanceoapi.ReadTelemetryEventsParams{}) |
| 78 | + require.NoError(t, err) |
| 79 | + require.Equal(t, http.StatusOK, all.StatusCode(), "read events: %s", string(all.Body)) |
| 80 | + require.NotNil(t, all.JSON200) |
| 81 | + assert.GreaterOrEqual(t, len(all.JSON200.Events), systemCount+connectionCount) |
| 82 | + |
| 83 | + // Category filter returns only the requested category. |
| 84 | + systemCat := []instanceoapi.TelemetryEventCategory{instanceoapi.TelemetryEventCategorySystem} |
| 85 | + systemOnly, err := client.ReadTelemetryEventsWithResponse(readCtx, &instanceoapi.ReadTelemetryEventsParams{Category: &systemCat}) |
| 86 | + require.NoError(t, err) |
| 87 | + require.Equal(t, http.StatusOK, systemOnly.StatusCode()) |
| 88 | + require.NotNil(t, systemOnly.JSON200) |
| 89 | + assert.GreaterOrEqual(t, len(systemOnly.JSON200.Events), systemCount) |
| 90 | + for _, e := range systemOnly.JSON200.Events { |
| 91 | + require.NotNil(t, e.Event.Category) |
| 92 | + assert.Equal(t, instanceoapi.TelemetryEventCategorySystem, *e.Event.Category) |
| 93 | + } |
| 94 | + |
| 95 | + // Limit caps the number of returned events. |
| 96 | + limit := 1 |
| 97 | + limited, err := client.ReadTelemetryEventsWithResponse(readCtx, &instanceoapi.ReadTelemetryEventsParams{Limit: &limit}) |
| 98 | + require.NoError(t, err) |
| 99 | + require.NotNil(t, limited.JSON200) |
| 100 | + assert.Len(t, limited.JSON200.Events, 1) |
| 101 | + |
| 102 | + // An empty window returns [] not null, or the Python SDK chokes deserializing. |
| 103 | + pastSince, pastUntil := int64(1), int64(2) |
| 104 | + empty, err := client.ReadTelemetryEventsWithResponse(readCtx, &instanceoapi.ReadTelemetryEventsParams{Since: &pastSince, Until: &pastUntil}) |
| 105 | + require.NoError(t, err) |
| 106 | + require.NotNil(t, empty.JSON200) |
| 107 | + assert.Empty(t, empty.JSON200.Events) |
| 108 | + assert.Contains(t, string(empty.Body), `"events":[]`) |
| 109 | +} |
| 110 | + |
| 111 | +func publishEvent(t *testing.T, ctx context.Context, client *instanceoapi.ClientWithResponses, eventType string, category instanceoapi.PublishEventRequestCategory) { |
| 112 | + t.Helper() |
| 113 | + resp, err := client.PublishTelemetryEventWithResponse(ctx, instanceoapi.PublishTelemetryEventJSONRequestBody{ |
| 114 | + Type: eventType, |
| 115 | + Category: &category, |
| 116 | + }) |
| 117 | + require.NoError(t, err) |
| 118 | + require.Equal(t, http.StatusOK, resp.StatusCode(), "publish %s: %s", eventType, string(resp.Body)) |
| 119 | +} |
0 commit comments