|
| 1 | +package discovery |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "errors" |
| 6 | + "fmt" |
| 7 | + "log/slog" |
| 8 | + "testing" |
| 9 | + "testing/synctest" |
| 10 | + "time" |
| 11 | + |
| 12 | + "github.com/aws/aws-sdk-go-v2/aws" |
| 13 | + cwltypes "github.com/aws/aws-sdk-go-v2/service/cloudwatchlogs/types" |
| 14 | + "github.com/stretchr/testify/require" |
| 15 | + |
| 16 | + accessgraphv1alpha "github.com/gravitational/teleport/gen/proto/go/accessgraph/v1alpha" |
| 17 | + "github.com/gravitational/teleport/lib/utils/testutils/grpctest" |
| 18 | +) |
| 19 | + |
| 20 | +type eksAuditLogFetcherFixture struct { |
| 21 | + ctx context.Context |
| 22 | + cancel context.CancelFunc |
| 23 | + server kalsServer |
| 24 | + fetcherErr error |
| 25 | + cluster *accessgraphv1alpha.AWSEKSClusterV1 |
| 26 | + fakeLogFetcher *fakeCloudWatchLogFetcher |
| 27 | +} |
| 28 | + |
| 29 | +// Start the fixture. Must be called inside synctest bubble. |
| 30 | +func (f *eksAuditLogFetcherFixture) Start(t *testing.T) { |
| 31 | + t.Helper() |
| 32 | + |
| 33 | + f.ctx, f.cancel = context.WithCancel(t.Context()) |
| 34 | + tester := grpctest.NewGRPCTester[kalsRequest, kalsResponse](f.ctx) |
| 35 | + f.server = tester.NewServerStream() |
| 36 | + logger := slog.New(slog.DiscardHandler) |
| 37 | + f.fakeLogFetcher = newFakeCloudWatchLogFetcher() |
| 38 | + f.cluster = &accessgraphv1alpha.AWSEKSClusterV1{ |
| 39 | + Name: "cluster-name", |
| 40 | + Arn: "cluster-arn", |
| 41 | + } |
| 42 | + logFetcher := newEKSAuditLogFetcher(f.fakeLogFetcher, f.cluster, tester.NewClientStream(), logger) |
| 43 | + go func() { f.fetcherErr = logFetcher.Run(f.ctx) }() |
| 44 | +} |
| 45 | + |
| 46 | +// End the fixture. Must be called inside synctest bubble. |
| 47 | +func (f *eksAuditLogFetcherFixture) End(t *testing.T) { |
| 48 | + t.Helper() |
| 49 | + f.cancel() |
| 50 | + synctest.Wait() |
| 51 | + require.ErrorIs(t, f.fetcherErr, context.Canceled) |
| 52 | +} |
| 53 | + |
| 54 | +func (f *eksAuditLogFetcherFixture) testInitializeNewStream(t *testing.T) { |
| 55 | + t.Helper() |
| 56 | + |
| 57 | + // Wait for a NewStream action, and verify it contains what we expect |
| 58 | + msg, err := f.server.Recv() |
| 59 | + require.NoError(t, err) |
| 60 | + newStream := msg.GetNewStream() |
| 61 | + require.NotNil(t, newStream) |
| 62 | + cursor := newStream.GetInitial() |
| 63 | + require.NotNil(t, cursor) |
| 64 | + require.Equal(t, accessgraphv1alpha.KubeAuditLogCursor_KUBE_AUDIT_LOG_SOURCE_EKS, cursor.GetLogSource()) |
| 65 | + require.Equal(t, f.cluster.GetArn(), cursor.GetClusterId()) |
| 66 | + |
| 67 | + // Send back a ResumeState |
| 68 | + err = f.server.Send(newKubeAuditLogResponseResumeState(cursor)) |
| 69 | + require.NoError(t, err) |
| 70 | +} |
| 71 | + |
| 72 | +// TestEKSAuditLogFetcher_NewStream_Unknown tests that when a new log stream |
| 73 | +// is set up for a cluster, logs start being fetched from the cursor returned |
| 74 | +// by the grpc service. |
| 75 | +func TestEKSAuditLogFetcher_NewStream(t *testing.T) { |
| 76 | + synctest.Test(t, func(t *testing.T) { |
| 77 | + f := &eksAuditLogFetcherFixture{} |
| 78 | + f.Start(t) |
| 79 | + f.testInitializeNewStream(t) |
| 80 | + f.End(t) |
| 81 | + }) |
| 82 | +} |
| 83 | + |
| 84 | +func TestEKSAuditLogFetcher_Batching(t *testing.T) { |
| 85 | + synctest.Test(t, func(t *testing.T) { |
| 86 | + startTime := time.Now().UTC() |
| 87 | + logEpoch := startTime.Add(-7 * 24 * time.Hour) |
| 88 | + f := &eksAuditLogFetcherFixture{} |
| 89 | + f.Start(t) |
| 90 | + f.testInitializeNewStream(t) |
| 91 | + |
| 92 | + f.fakeLogFetcher.events <- nil |
| 93 | + // Wait for a polling loop to occur. As there are no logs left, |
| 94 | + // the time should now be the synctest epoch plus the poll interval |
| 95 | + time.Sleep(logPollInterval) |
| 96 | + synctest.Wait() |
| 97 | + require.Equal(t, startTime.Add(logPollInterval), time.Now().UTC()) |
| 98 | + |
| 99 | + // Wait for an Events action with the log listed. Verify the log and cursor. |
| 100 | + f.fakeLogFetcher.events <- []cwltypes.FilteredLogEvent{ |
| 101 | + makeEvent(logEpoch, 0, "{}"), |
| 102 | + makeEvent(logEpoch.Add(time.Second), 1, `{"log": "value"}`), |
| 103 | + } |
| 104 | + msg, err := f.server.Recv() |
| 105 | + require.NoError(t, err) |
| 106 | + events := msg.GetEvents() |
| 107 | + require.NotNil(t, events) |
| 108 | + require.Len(t, events.GetEvents(), 2) |
| 109 | + require.Len(t, events.GetEvents()[0].GetFields(), 0) |
| 110 | + require.Len(t, events.GetEvents()[1].GetFields(), 1) |
| 111 | + cursor := events.GetCursor() |
| 112 | + require.NotNil(t, cursor) |
| 113 | + require.Equal(t, accessgraphv1alpha.KubeAuditLogCursor_KUBE_AUDIT_LOG_SOURCE_EKS, cursor.GetLogSource()) |
| 114 | + require.Equal(t, f.cluster.GetArn(), cursor.GetClusterId()) |
| 115 | + require.Equal(t, "event-id-1", cursor.GetEventId()) |
| 116 | + require.Equal(t, logEpoch.Add(time.Second), cursor.GetLastEventTime().AsTime()) |
| 117 | + |
| 118 | + f.fakeLogFetcher.events <- []cwltypes.FilteredLogEvent{ |
| 119 | + makeEvent(logEpoch.Add(time.Second), 2, `{"log": "value2"}`), |
| 120 | + makeEvent(logEpoch.Add(2*time.Second), 3, `{}`), |
| 121 | + } |
| 122 | + msg, err = f.server.Recv() |
| 123 | + require.NoError(t, err) |
| 124 | + events = msg.GetEvents() |
| 125 | + require.NotNil(t, events) |
| 126 | + require.Len(t, events.GetEvents(), 2) |
| 127 | + require.Len(t, events.GetEvents()[0].GetFields(), 1) |
| 128 | + require.Len(t, events.GetEvents()[1].GetFields(), 0) |
| 129 | + cursor = events.GetCursor() |
| 130 | + require.NotNil(t, cursor) |
| 131 | + require.Equal(t, accessgraphv1alpha.KubeAuditLogCursor_KUBE_AUDIT_LOG_SOURCE_EKS, cursor.GetLogSource()) |
| 132 | + require.Equal(t, f.cluster.GetArn(), cursor.GetClusterId()) |
| 133 | + require.Equal(t, "event-id-3", cursor.GetEventId()) |
| 134 | + require.Equal(t, logEpoch.Add(2*time.Second), cursor.GetLastEventTime().AsTime()) |
| 135 | + |
| 136 | + f.End(t) |
| 137 | + }) |
| 138 | +} |
| 139 | + |
| 140 | +func TestEKSAuditLogFetcher_ContinueOnError(t *testing.T) { |
| 141 | + synctest.Test(t, func(t *testing.T) { |
| 142 | + startTime := time.Now().UTC() |
| 143 | + logEpoch := startTime.Add(-7 * 24 * time.Hour) |
| 144 | + f := &eksAuditLogFetcherFixture{} |
| 145 | + f.Start(t) |
| 146 | + f.testInitializeNewStream(t) |
| 147 | + |
| 148 | + f.fakeLogFetcher.err <- errors.New("oh noes. something went wrong") |
| 149 | + // Wait for a polling loop to occur. As there are no logs left, |
| 150 | + // the time should now be the synctest epoch plus the poll interval |
| 151 | + time.Sleep(logPollInterval) |
| 152 | + synctest.Wait() |
| 153 | + require.Equal(t, startTime.Add(logPollInterval), time.Now().UTC()) |
| 154 | + |
| 155 | + // Wait for an Events action with the log listed. Verify the log and cursor. |
| 156 | + f.fakeLogFetcher.events <- []cwltypes.FilteredLogEvent{ |
| 157 | + makeEvent(logEpoch, 0, "{}"), |
| 158 | + makeEvent(logEpoch.Add(time.Second), 1, `{"log": "value"}`), |
| 159 | + } |
| 160 | + msg, err := f.server.Recv() |
| 161 | + require.NoError(t, err) |
| 162 | + events := msg.GetEvents() |
| 163 | + require.NotNil(t, events) |
| 164 | + require.Len(t, events.GetEvents(), 2) |
| 165 | + require.Len(t, events.GetEvents()[0].GetFields(), 0) |
| 166 | + require.Len(t, events.GetEvents()[1].GetFields(), 1) |
| 167 | + cursor := events.GetCursor() |
| 168 | + require.NotNil(t, cursor) |
| 169 | + require.Equal(t, accessgraphv1alpha.KubeAuditLogCursor_KUBE_AUDIT_LOG_SOURCE_EKS, cursor.GetLogSource()) |
| 170 | + require.Equal(t, f.cluster.GetArn(), cursor.GetClusterId()) |
| 171 | + require.Equal(t, "event-id-1", cursor.GetEventId()) |
| 172 | + require.Equal(t, logEpoch.Add(time.Second), cursor.GetLastEventTime().AsTime()) |
| 173 | + |
| 174 | + f.End(t) |
| 175 | + }) |
| 176 | +} |
| 177 | + |
| 178 | +func newKubeAuditLogResponseResumeState(cursor *accessgraphv1alpha.KubeAuditLogCursor) *kalsResponse { |
| 179 | + return &kalsResponse{ |
| 180 | + State: &accessgraphv1alpha.KubeAuditLogStreamResponse_ResumeState{ |
| 181 | + ResumeState: &accessgraphv1alpha.KubeAuditLogResumeState{ |
| 182 | + Cursor: cursor, |
| 183 | + }, |
| 184 | + }, |
| 185 | + } |
| 186 | +} |
| 187 | + |
| 188 | +func makeEvent(t time.Time, id int, msg string) cwltypes.FilteredLogEvent { |
| 189 | + return cwltypes.FilteredLogEvent{ |
| 190 | + EventId: aws.String(fmt.Sprintf("event-id-%d", id)), |
| 191 | + IngestionTime: aws.Int64(t.UnixMilli()), |
| 192 | + Timestamp: aws.Int64(t.UnixMilli()), |
| 193 | + LogStreamName: aws.String("kube-apiserver-audit-12345678"), |
| 194 | + Message: aws.String(msg), |
| 195 | + } |
| 196 | +} |
| 197 | + |
| 198 | +func newFakeCloudWatchLogFetcher() *fakeCloudWatchLogFetcher { |
| 199 | + return &fakeCloudWatchLogFetcher{ |
| 200 | + events: make(chan []cwltypes.FilteredLogEvent), |
| 201 | + err: make(chan error), |
| 202 | + } |
| 203 | +} |
| 204 | + |
| 205 | +// fakeCloudWatchLogFetcher is a cloudwatch log fetcher that waits on channels |
| 206 | +// for the data to return. This allows the unit under test to rendezvous with |
| 207 | +// the tests, allowing the tests to advance the state of the fetcher as it |
| 208 | +// needs. |
| 209 | +type fakeCloudWatchLogFetcher struct { |
| 210 | + events chan []cwltypes.FilteredLogEvent |
| 211 | + err chan error |
| 212 | +} |
| 213 | + |
| 214 | +func (f *fakeCloudWatchLogFetcher) FetchEKSAuditLogs( |
| 215 | + ctx context.Context, |
| 216 | + cluster *accessgraphv1alpha.AWSEKSClusterV1, |
| 217 | + cursor *accessgraphv1alpha.KubeAuditLogCursor, |
| 218 | +) ([]cwltypes.FilteredLogEvent, error) { |
| 219 | + select { |
| 220 | + case events := <-f.events: |
| 221 | + return events, nil |
| 222 | + case err := <-f.err: |
| 223 | + return nil, err |
| 224 | + case <-ctx.Done(): |
| 225 | + return nil, ctx.Err() |
| 226 | + } |
| 227 | +} |
0 commit comments