|
| 1 | +// +build requires_docker |
| 2 | + |
| 3 | +package integration |
| 4 | + |
| 5 | +import ( |
| 6 | + "fmt" |
| 7 | + "strings" |
| 8 | + "testing" |
| 9 | + "time" |
| 10 | + |
| 11 | + "github.com/prometheus/common/model" |
| 12 | + "github.com/prometheus/prometheus/pkg/labels" |
| 13 | + "github.com/prometheus/prometheus/prompb" |
| 14 | + "github.com/stretchr/testify/assert" |
| 15 | + "github.com/stretchr/testify/require" |
| 16 | + |
| 17 | + "github.com/cortexproject/cortex/integration/e2e" |
| 18 | + e2ecache "github.com/cortexproject/cortex/integration/e2e/cache" |
| 19 | + e2edb "github.com/cortexproject/cortex/integration/e2e/db" |
| 20 | + "github.com/cortexproject/cortex/integration/e2ecortex" |
| 21 | +) |
| 22 | + |
| 23 | +type querierTenantFederationConfig struct { |
| 24 | + querySchedulerEnabled bool |
| 25 | + shuffleShardingEnabled bool |
| 26 | +} |
| 27 | + |
| 28 | +func TestQuerierTenantFederation(t *testing.T) { |
| 29 | + runQuerierTenantFederationTest(t, querierTenantFederationConfig{}) |
| 30 | +} |
| 31 | + |
| 32 | +func TestQuerierTenantFederationWithQueryScheduler(t *testing.T) { |
| 33 | + runQuerierTenantFederationTest(t, querierTenantFederationConfig{ |
| 34 | + querySchedulerEnabled: true, |
| 35 | + }) |
| 36 | +} |
| 37 | + |
| 38 | +func TestQuerierTenantFederationWithShuffleSharding(t *testing.T) { |
| 39 | + runQuerierTenantFederationTest(t, querierTenantFederationConfig{ |
| 40 | + shuffleShardingEnabled: true, |
| 41 | + }) |
| 42 | +} |
| 43 | + |
| 44 | +func TestQuerierTenantFederationWithQuerySchedulerAndShuffleSharding(t *testing.T) { |
| 45 | + runQuerierTenantFederationTest(t, querierTenantFederationConfig{ |
| 46 | + querySchedulerEnabled: true, |
| 47 | + shuffleShardingEnabled: true, |
| 48 | + }) |
| 49 | +} |
| 50 | + |
| 51 | +func runQuerierTenantFederationTest(t *testing.T, cfg querierTenantFederationConfig) { |
| 52 | + const numUsers = 10 |
| 53 | + |
| 54 | + s, err := e2e.NewScenario(networkName) |
| 55 | + require.NoError(t, err) |
| 56 | + defer s.Close() |
| 57 | + |
| 58 | + memcached := e2ecache.NewMemcached() |
| 59 | + consul := e2edb.NewConsul() |
| 60 | + require.NoError(t, s.StartAndWaitReady(consul, memcached)) |
| 61 | + |
| 62 | + flags := mergeFlags(BlocksStorageFlags(), map[string]string{ |
| 63 | + "-querier.cache-results": "true", |
| 64 | + "-querier.split-queries-by-interval": "24h", |
| 65 | + "-querier.query-ingesters-within": "12h", // Required by the test on query /series out of ingesters time range |
| 66 | + "-frontend.memcached.addresses": "dns+" + memcached.NetworkEndpoint(e2ecache.MemcachedPort), |
| 67 | + "-tenant-federation.enabled": "true", |
| 68 | + }) |
| 69 | + |
| 70 | + // Start the query-scheduler if enabled. |
| 71 | + var queryScheduler *e2ecortex.CortexService |
| 72 | + if cfg.querySchedulerEnabled { |
| 73 | + queryScheduler = e2ecortex.NewQueryScheduler("query-scheduler", flags, "") |
| 74 | + require.NoError(t, s.StartAndWaitReady(queryScheduler)) |
| 75 | + flags["-frontend.scheduler-address"] = queryScheduler.NetworkGRPCEndpoint() |
| 76 | + flags["-querier.scheduler-address"] = queryScheduler.NetworkGRPCEndpoint() |
| 77 | + } |
| 78 | + |
| 79 | + if cfg.shuffleShardingEnabled { |
| 80 | + // Use only single querier for each user. |
| 81 | + flags["-frontend.max-queriers-per-tenant"] = "1" |
| 82 | + } |
| 83 | + |
| 84 | + minio := e2edb.NewMinio(9000, flags["-blocks-storage.s3.bucket-name"]) |
| 85 | + require.NoError(t, s.StartAndWaitReady(minio)) |
| 86 | + |
| 87 | + // Start the query-frontend. |
| 88 | + queryFrontend := e2ecortex.NewQueryFrontend("query-frontend", flags, "") |
| 89 | + require.NoError(t, s.Start(queryFrontend)) |
| 90 | + |
| 91 | + if !cfg.querySchedulerEnabled { |
| 92 | + flags["-querier.frontend-address"] = queryFrontend.NetworkGRPCEndpoint() |
| 93 | + } |
| 94 | + |
| 95 | + // Start all other services. |
| 96 | + ingester := e2ecortex.NewIngester("ingester", consul.NetworkHTTPEndpoint(), flags, "") |
| 97 | + distributor := e2ecortex.NewDistributor("distributor", consul.NetworkHTTPEndpoint(), flags, "") |
| 98 | + querier := e2ecortex.NewQuerier("querier", consul.NetworkHTTPEndpoint(), flags, "") |
| 99 | + |
| 100 | + var querier2 *e2ecortex.CortexService |
| 101 | + if cfg.shuffleShardingEnabled { |
| 102 | + querier2 = e2ecortex.NewQuerier("querier-2", consul.NetworkHTTPEndpoint(), flags, "") |
| 103 | + } |
| 104 | + |
| 105 | + require.NoError(t, s.StartAndWaitReady(querier, ingester, distributor)) |
| 106 | + require.NoError(t, s.WaitReady(queryFrontend)) |
| 107 | + if cfg.shuffleShardingEnabled { |
| 108 | + require.NoError(t, s.StartAndWaitReady(querier2)) |
| 109 | + } |
| 110 | + |
| 111 | + // Wait until distributor and queriers have updated the ring. |
| 112 | + require.NoError(t, distributor.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 113 | + require.NoError(t, querier.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 114 | + if cfg.shuffleShardingEnabled { |
| 115 | + require.NoError(t, querier2.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 116 | + } |
| 117 | + |
| 118 | + // Push a series for each user to Cortex. |
| 119 | + now := time.Now() |
| 120 | + expectedVectors := make([]model.Vector, numUsers) |
| 121 | + tenantIDs := make([]string, numUsers) |
| 122 | + |
| 123 | + for u := 0; u < numUsers; u++ { |
| 124 | + tenantIDs[u] = fmt.Sprintf("user-%d", u) |
| 125 | + c, err := e2ecortex.NewClient(distributor.HTTPEndpoint(), "", "", "", tenantIDs[u]) |
| 126 | + require.NoError(t, err) |
| 127 | + |
| 128 | + var series []prompb.TimeSeries |
| 129 | + series, expectedVectors[u] = generateSeries("series_1", now) |
| 130 | + |
| 131 | + res, err := c.Push(series) |
| 132 | + require.NoError(t, err) |
| 133 | + require.Equal(t, 200, res.StatusCode) |
| 134 | + } |
| 135 | + |
| 136 | + // query all tenants |
| 137 | + c, err := e2ecortex.NewClient(distributor.HTTPEndpoint(), queryFrontend.HTTPEndpoint(), "", "", strings.Join(tenantIDs, "|")) |
| 138 | + require.NoError(t, err) |
| 139 | + |
| 140 | + result, err := c.Query("series_1", now) |
| 141 | + require.NoError(t, err) |
| 142 | + |
| 143 | + assert.Equal(t, mergeResults(tenantIDs, expectedVectors), result.(model.Vector)) |
| 144 | + |
| 145 | + // ensure a push to multiple tenants is failing |
| 146 | + series, _ := generateSeries("series_1", now) |
| 147 | + res, err := c.Push(series) |
| 148 | + require.NoError(t, err) |
| 149 | + require.Equal(t, 500, res.StatusCode) |
| 150 | + |
| 151 | + // check metric label values for total queries in the query frontend |
| 152 | + require.NoError(t, queryFrontend.WaitSumMetricsWithOptions(e2e.Equals(1), []string{"cortex_query_frontend_queries_total"}, e2e.WithLabelMatchers( |
| 153 | + labels.MustNewMatcher(labels.MatchEqual, "user", strings.Join(tenantIDs, "|")), |
| 154 | + labels.MustNewMatcher(labels.MatchEqual, "op", "query")))) |
| 155 | + |
| 156 | + // check metric label values for query queue length in either query frontend or query scheduler |
| 157 | + queueComponent := queryFrontend |
| 158 | + queueMetricName := "cortex_query_frontend_queue_length" |
| 159 | + if cfg.querySchedulerEnabled { |
| 160 | + queueComponent = queryScheduler |
| 161 | + queueMetricName = "cortex_query_scheduler_queue_length" |
| 162 | + } |
| 163 | + require.NoError(t, queueComponent.WaitSumMetricsWithOptions(e2e.Equals(0), []string{queueMetricName}, e2e.WithLabelMatchers( |
| 164 | + labels.MustNewMatcher(labels.MatchEqual, "user", strings.Join(tenantIDs, "|"))))) |
| 165 | + |
| 166 | + // TODO: check cache invalidation on tombstone cache gen increase |
| 167 | + // TODO: check fairness in queryfrontend |
| 168 | +} |
| 169 | + |
| 170 | +func mergeResults(tenantIDs []string, resultsPerTenant []model.Vector) model.Vector { |
| 171 | + var v model.Vector |
| 172 | + for pos, tenantID := range tenantIDs { |
| 173 | + for _, r := range resultsPerTenant[pos] { |
| 174 | + var s model.Sample = *r |
| 175 | + s.Metric = r.Metric.Clone() |
| 176 | + s.Metric[model.LabelName("__tenant_id__")] = model.LabelValue(tenantID) |
| 177 | + v = append(v, &s) |
| 178 | + } |
| 179 | + } |
| 180 | + return v |
| 181 | +} |
0 commit comments