| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package sd |
| 4 | |
| 5 | import ( |
| 6 | "bytes" |
| 7 | "context" |
| 8 | "fmt" |
| 9 | "testing" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/netdata/netdata/go/plugins/logger" |
| 13 | "github.com/netdata/netdata/go/plugins/pkg/netdataapi" |
| 14 | "github.com/netdata/netdata/go/plugins/pkg/safewriter" |
| 15 | "github.com/netdata/netdata/go/plugins/plugin/agent/discovery/sd/pipeline" |
| 16 | "github.com/netdata/netdata/go/plugins/plugin/agent/policy" |
| 17 | "github.com/netdata/netdata/go/plugins/plugin/framework/confgroup" |
| 18 | "github.com/netdata/netdata/go/plugins/plugin/framework/dyncfg" |
| 19 | "github.com/netdata/netdata/go/plugins/plugin/framework/functions" |
| 20 | |
| 21 | "github.com/stretchr/testify/assert" |
| 22 | "github.com/stretchr/testify/require" |
| 23 | ) |
| 24 | |
| 25 | func TestServiceDiscovery_Run_WaitDecision(t *testing.T) { |
| 26 | tests := map[string]struct { |
| 27 | run func(t *testing.T, sd *ServiceDiscovery, confCh chan confFile, stop func()) |
| 28 | }{ |
| 29 | "enable command clears wait and starts pipeline": { |
| 30 | run: func(t *testing.T, sd *ServiceDiscovery, confCh chan confFile, stop func()) { |
| 31 | cfg := prepareConfigFile("/etc/netdata/sd.d/job1.conf", "job1") |
| 32 | confCh <- cfg |
| 33 | |
| 34 | require.Eventually(t, sd.handler.WaitingForDecision, time.Second, 10*time.Millisecond) |
| 35 | |
| 36 | sd.dyncfgCh <- dyncfg.NewFunction(functions.Function{ |
| 37 | UID: "enable-job1", |
| 38 | Args: []string{sd.dyncfgJobID(testDiscovererTypeNetListeners, "job1"), "enable"}, |
| 39 | }) |
| 40 | |
| 41 | require.Eventually(t, func() bool { |
| 42 | return !sd.handler.WaitingForDecision() && |
| 43 | exposedExistsByKey(sd.exposed, testDiscovererTypeNetListeners+":job1") && |
| 44 | sd.mgr.IsRunning(pipelineKeyFromSource(cfg.source)) |
| 45 | }, time.Second, 10*time.Millisecond) |
| 46 | |
| 47 | stop() |
| 48 | |
| 49 | entry, ok := sd.exposed.LookupByKey(testDiscovererTypeNetListeners + ":job1") |
| 50 | require.True(t, ok) |
| 51 | assert.Equal(t, dyncfg.StatusRunning, entry.Status) |
| 52 | }, |
| 53 | }, |
| 54 | "second config blocks while wait gate is open and proceeds after decision": { |
| 55 | run: func(t *testing.T, sd *ServiceDiscovery, confCh chan confFile, stop func()) { |
| 56 | // Without a wait-decision timeout, the run loop must stay in |
| 57 | // WaitingForDecision until an explicit enable/disable for cfg1 |
| 58 | // arrives. Sending cfg2 in the meantime must block until the |
| 59 | // gate clears. This guards against accidental gate-clearing or |
| 60 | // a regression that lets new configs interleave. |
| 61 | cfg1 := prepareConfigFile("/etc/netdata/sd.d/job1.conf", "job1") |
| 62 | cfg2 := prepareConfigFile("/etc/netdata/sd.d/job2.conf", "job2") |
| 63 | |
| 64 | confCh <- cfg1 |
| 65 | require.Eventually(t, sd.handler.WaitingForDecision, time.Second, 10*time.Millisecond) |
| 66 | |
| 67 | secondSent := make(chan struct{}) |
| 68 | go func() { |
| 69 | confCh <- cfg2 |
| 70 | close(secondSent) |
| 71 | }() |
| 72 | |
| 73 | // Give the goroutine a chance to either block (expected) or |
| 74 | // race ahead. We can't use require.Eventually for negative |
| 75 | // "still blocked" assertions, but a short window is enough to |
| 76 | // catch a regression that lets the second config flow through |
| 77 | // while the wait gate is still open. |
| 78 | select { |
| 79 | case <-secondSent: |
| 80 | t.Fatal("second config was processed while wait gate was open") |
| 81 | case <-time.After(100 * time.Millisecond): |
| 82 | } |
| 83 | require.True(t, sd.handler.WaitingForDecision(), "wait gate should still be open before decision") |
| 84 | |
| 85 | // Send the matching enable for cfg1 — this clears the wait gate. |
| 86 | sd.dyncfgCh <- dyncfg.NewFunction(functions.Function{ |
| 87 | UID: "enable-job1", |
| 88 | Args: []string{sd.dyncfgJobID(testDiscovererTypeNetListeners, "job1"), "enable"}, |
| 89 | }) |
| 90 | |
| 91 | select { |
| 92 | case <-secondSent: |
| 93 | case <-time.After(2 * time.Second): |
| 94 | t.Fatal("second config did not proceed after wait gate cleared") |
| 95 | } |
| 96 | |
| 97 | require.Eventually(t, func() bool { |
| 98 | return exposedExistsByKey(sd.exposed, testDiscovererTypeNetListeners+":job1") && |
| 99 | exposedExistsByKey(sd.exposed, testDiscovererTypeNetListeners+":job2") |
| 100 | }, time.Second, 10*time.Millisecond) |
| 101 | }, |
| 102 | }, |
| 103 | } |
| 104 | |
| 105 | for name, tc := range tests { |
| 106 | t.Run(name, func(t *testing.T) { |
| 107 | sd, confCh, cancel, done := newWaitTestServiceDiscovery(t) |
| 108 | stopped := false |
| 109 | stop := func() { |
| 110 | if stopped { |
| 111 | return |
| 112 | } |
| 113 | stopped = true |
| 114 | stopWaitTestServiceDiscovery(t, sd, cancel, done) |
| 115 | } |
| 116 | defer stop() |
| 117 | tc.run(t, sd, confCh, stop) |
| 118 | }) |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | func newWaitTestServiceDiscovery(t *testing.T) (*ServiceDiscovery, chan confFile, context.CancelFunc, <-chan struct{}) { |
| 123 | t.Helper() |
| 124 | |
| 125 | var out bytes.Buffer |
| 126 | confProv := &mockConfigProvider{ch: make(chan confFile)} |
| 127 | |
| 128 | sd := &ServiceDiscovery{ |
| 129 | Logger: logger.New(), |
| 130 | confProv: confProv, |
| 131 | pluginName: testPluginName, |
| 132 | fnReg: functions.NewManager(), |
| 133 | discoverers: testDiscovererRegistry(), |
| 134 | dyncfgApi: dyncfg.NewResponder(netdataapi.New(safewriter.New(&out))), |
| 135 | seen: dyncfg.NewSeenCache[sdConfig](), |
| 136 | exposed: dyncfg.NewExposedCache[sdConfig](), |
| 137 | dyncfgCh: make(chan dyncfg.Function, 1), |
| 138 | newPipeline: newWaitTestPipeline, |
| 139 | runModePolicy: policy.RunModePolicy{}, |
| 140 | configDefaults: nil, |
| 141 | } |
| 142 | sd.sdCb = &sdCallbacks{sd: sd} |
| 143 | sd.handler = dyncfg.NewHandler(dyncfg.HandlerOpts[sdConfig]{ |
| 144 | Logger: sd.Logger, |
| 145 | API: sd.dyncfgApi, |
| 146 | Seen: sd.seen, |
| 147 | Exposed: sd.exposed, |
| 148 | Callbacks: sd.sdCb, |
| 149 | WaitKey: func(cfg sdConfig) string { |
| 150 | return cfg.PipelineKey() |
| 151 | }, |
| 152 | |
| 153 | Path: fmt.Sprintf(dyncfgSDPath, testPluginName), |
| 154 | EnableFailCode: 422, |
| 155 | JobCommands: []dyncfg.Command{ |
| 156 | dyncfg.CommandSchema, |
| 157 | dyncfg.CommandGet, |
| 158 | dyncfg.CommandEnable, |
| 159 | dyncfg.CommandDisable, |
| 160 | dyncfg.CommandUpdate, |
| 161 | dyncfg.CommandTest, |
| 162 | dyncfg.CommandUserconfig, |
| 163 | }, |
| 164 | }) |
| 165 | |
| 166 | send := func(context.Context, []*confgroup.Group) {} |
| 167 | sd.mgr = NewPipelineManager(sd.Logger, sd.newPipeline, send) |
| 168 | |
| 169 | ctx, cancel := context.WithCancel(context.Background()) |
| 170 | sd.ctx = ctx |
| 171 | |
| 172 | done := make(chan struct{}) |
| 173 | go func() { |
| 174 | defer close(done) |
| 175 | sd.run(ctx) |
| 176 | }() |
| 177 | |
| 178 | return sd, confProv.ch, cancel, done |
| 179 | } |
| 180 | |
| 181 | func stopWaitTestServiceDiscovery(t *testing.T, sd *ServiceDiscovery, cancel context.CancelFunc, done <-chan struct{}) { |
| 182 | t.Helper() |
| 183 | |
| 184 | cancel() |
| 185 | |
| 186 | select { |
| 187 | case <-done: |
| 188 | case <-time.After(5 * time.Second): |
| 189 | t.Fatal("service discovery run loop did not stop") |
| 190 | } |
| 191 | |
| 192 | sd.mgr.StopAll() |
| 193 | } |
| 194 | |
| 195 | func newWaitTestPipeline(cfg pipeline.Config) (sdPipeline, error) { |
| 196 | return newTestPipeline(cfg.Name), nil |
| 197 | } |
| 198 | |
| 199 | func exposedExistsByKey(cache *dyncfg.ExposedCache[sdConfig], key string) bool { |
| 200 | _, ok := cache.LookupByKey(key) |
| 201 | return ok |
| 202 | } |