|
| 1 | +// +build requires_docker |
| 2 | + |
| 3 | +package integration |
| 4 | + |
| 5 | +import ( |
| 6 | + "strconv" |
| 7 | + "sync" |
| 8 | + "testing" |
| 9 | + "time" |
| 10 | + |
| 11 | + "github.com/prometheus/common/model" |
| 12 | + "github.com/prometheus/prometheus/prompb" |
| 13 | + "github.com/stretchr/testify/assert" |
| 14 | + "github.com/stretchr/testify/require" |
| 15 | + |
| 16 | + "github.com/cortexproject/cortex/integration/e2e" |
| 17 | + e2ecache "github.com/cortexproject/cortex/integration/e2e/cache" |
| 18 | + e2edb "github.com/cortexproject/cortex/integration/e2e/db" |
| 19 | + "github.com/cortexproject/cortex/integration/e2ecortex" |
| 20 | +) |
| 21 | + |
| 22 | +func TestQuerierSharding(t *testing.T) { |
| 23 | + runQuerierShardingTest(t, true) |
| 24 | +} |
| 25 | + |
| 26 | +func TestQuerierNoSharding(t *testing.T) { |
| 27 | + runQuerierShardingTest(t, false) |
| 28 | +} |
| 29 | + |
| 30 | +func runQuerierShardingTest(t *testing.T, sharding bool) { |
| 31 | + // Going to high starts hitting filedescriptor limit, since we run all queriers concurrently. |
| 32 | + const numQueries = 100 |
| 33 | + |
| 34 | + s, err := e2e.NewScenario(networkName) |
| 35 | + require.NoError(t, err) |
| 36 | + defer s.Close() |
| 37 | + |
| 38 | + memcached := e2ecache.NewMemcached() |
| 39 | + consul := e2edb.NewConsul() |
| 40 | + require.NoError(t, s.StartAndWaitReady(consul, memcached)) |
| 41 | + |
| 42 | + minio := e2edb.NewMinio(9000, BlocksStorageFlags["-blocks-storage.s3.bucket-name"]) |
| 43 | + require.NoError(t, s.StartAndWaitReady(minio)) |
| 44 | + |
| 45 | + flags := BlocksStorageFlags |
| 46 | + |
| 47 | + flags = mergeFlags(flags, map[string]string{ |
| 48 | + "-querier.cache-results": "true", |
| 49 | + "-querier.split-queries-by-interval": "24h", |
| 50 | + "-querier.query-ingesters-within": "12h", // Required by the test on query /series out of ingesters time range |
| 51 | + "-frontend.memcached.addresses": "dns+" + memcached.NetworkEndpoint(e2ecache.MemcachedPort), |
| 52 | + "-querier.max-outstanding-requests-per-tenant": strconv.Itoa(numQueries), // To avoid getting errors. |
| 53 | + }) |
| 54 | + |
| 55 | + if sharding { |
| 56 | + // Use only single querier for each user. |
| 57 | + flags["-frontend.max-queriers-per-user"] = "1" |
| 58 | + } |
| 59 | + |
| 60 | + // Start Cortex components. |
| 61 | + queryFrontend := e2ecortex.NewQueryFrontendWithConfigFile("query-frontend", "", flags, "") |
| 62 | + ingester := e2ecortex.NewIngesterWithConfigFile("ingester", consul.NetworkHTTPEndpoint(), "", flags, "") |
| 63 | + distributor := e2ecortex.NewDistributorWithConfigFile("distributor", consul.NetworkHTTPEndpoint(), "", flags, "") |
| 64 | + |
| 65 | + require.NoError(t, s.Start(queryFrontend)) |
| 66 | + |
| 67 | + querier1 := e2ecortex.NewQuerierWithConfigFile("querier-1", consul.NetworkHTTPEndpoint(), "", mergeFlags(flags, map[string]string{ |
| 68 | + "-querier.frontend-address": queryFrontend.NetworkGRPCEndpoint(), |
| 69 | + }), "") |
| 70 | + querier2 := e2ecortex.NewQuerierWithConfigFile("querier-2", consul.NetworkHTTPEndpoint(), "", mergeFlags(flags, map[string]string{ |
| 71 | + "-querier.frontend-address": queryFrontend.NetworkGRPCEndpoint(), |
| 72 | + }), "") |
| 73 | + |
| 74 | + require.NoError(t, s.StartAndWaitReady(querier1, querier2, ingester, distributor)) |
| 75 | + require.NoError(t, s.WaitReady(queryFrontend)) |
| 76 | + |
| 77 | + // Wait until distributor and queriers have updated the ring. |
| 78 | + require.NoError(t, distributor.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 79 | + require.NoError(t, querier1.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 80 | + require.NoError(t, querier2.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total")) |
| 81 | + |
| 82 | + // Push a series for each user to Cortex. |
| 83 | + now := time.Now() |
| 84 | + |
| 85 | + distClient, err := e2ecortex.NewClient(distributor.HTTPEndpoint(), "", "", "", userID) |
| 86 | + require.NoError(t, err) |
| 87 | + |
| 88 | + var series []prompb.TimeSeries |
| 89 | + series, expectedVector := generateSeries("series_1", now) |
| 90 | + |
| 91 | + res, err := distClient.Push(series) |
| 92 | + require.NoError(t, err) |
| 93 | + require.Equal(t, 200, res.StatusCode) |
| 94 | + |
| 95 | + // Send both queriers a single query, so that they both initialize their cortex_querier_request_duration_seconds metrics. |
| 96 | + for _, q := range []*e2ecortex.CortexService{querier1, querier2} { |
| 97 | + c, err := e2ecortex.NewClient("", q.HTTPEndpoint(), "", "", userID) |
| 98 | + require.NoError(t, err) |
| 99 | + |
| 100 | + _, err = c.Query("series_1", now) |
| 101 | + require.NoError(t, err) |
| 102 | + } |
| 103 | + |
| 104 | + wg := sync.WaitGroup{} |
| 105 | + |
| 106 | + // Run all queries concurrently to get better distribution of requests between queriers. |
| 107 | + for i := 0; i < numQueries; i++ { |
| 108 | + wg.Add(1) |
| 109 | + |
| 110 | + go func() { |
| 111 | + defer wg.Done() |
| 112 | + c, err := e2ecortex.NewClient("", queryFrontend.HTTPEndpoint(), "", "", userID) |
| 113 | + require.NoError(t, err) |
| 114 | + |
| 115 | + result, err := c.Query("series_1", now) |
| 116 | + require.NoError(t, err) |
| 117 | + require.Equal(t, model.ValVector, result.Type()) |
| 118 | + assert.Equal(t, expectedVector, result.(model.Vector)) |
| 119 | + }() |
| 120 | + } |
| 121 | + |
| 122 | + wg.Wait() |
| 123 | + |
| 124 | + require.NoError(t, queryFrontend.WaitSumMetrics(e2e.Equals(numQueries), "cortex_query_frontend_queries_total")) |
| 125 | + |
| 126 | + // Verify that only single querier handled all the queries when sharding is enabled, otherwise queries have been fairly distributed across queriers. |
| 127 | + q1Values, err := querier1.SumMetrics([]string{"cortex_querier_request_duration_seconds"}, e2e.WithMetricCount) |
| 128 | + require.NoError(t, err) |
| 129 | + require.Len(t, q1Values, 1) |
| 130 | + |
| 131 | + q2Values, err := querier2.SumMetrics([]string{"cortex_querier_request_duration_seconds"}, e2e.WithMetricCount) |
| 132 | + require.NoError(t, err) |
| 133 | + require.Len(t, q2Values, 1) |
| 134 | + |
| 135 | + total := q1Values[0] + q2Values[0] |
| 136 | + diff := q1Values[0] - q2Values[0] |
| 137 | + if diff < 0 { |
| 138 | + diff = -diff |
| 139 | + } |
| 140 | + |
| 141 | + require.Equal(t, float64(numQueries), total-2) // Remove 2 requests used for metrics initialization. |
| 142 | + |
| 143 | + if sharding { |
| 144 | + require.Equal(t, float64(numQueries), diff) |
| 145 | + } else { |
| 146 | + require.InDelta(t, 0, diff, numQueries*0.20) // Both queriers should have roughly equal number of requests, with possible delta. |
| 147 | + } |
| 148 | + |
| 149 | + // Ensure no service-specific metrics prefix is used by the wrong service. |
| 150 | + assertServiceMetricsPrefixes(t, Distributor, distributor) |
| 151 | + assertServiceMetricsPrefixes(t, Ingester, ingester) |
| 152 | + assertServiceMetricsPrefixes(t, Querier, querier1) |
| 153 | + assertServiceMetricsPrefixes(t, Querier, querier2) |
| 154 | + assertServiceMetricsPrefixes(t, QueryFrontend, queryFrontend) |
| 155 | +} |
0 commit comments