| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package agenthost |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "os" |
| 8 | "os/signal" |
| 9 | "sync" |
| 10 | "syscall" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/netdata/netdata/go/plugins/plugin/agent" |
| 14 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 15 | ) |
| 16 | |
| 17 | // Run hosts an agent process lifecycle (signals, restart, quit, metrics-audit timer). |
| 18 | func Run(a *agent.Agent) { |
| 19 | ch := make(chan os.Signal, 1) |
| 20 | signal.Notify(ch, syscall.SIGHUP, syscall.SIGINT, syscall.SIGTERM) |
| 21 | signal.Ignore(syscall.SIGPIPE) |
| 22 | |
| 23 | var keepAliveErr <-chan error |
| 24 | if !a.IsTerminalMode() { |
| 25 | ch := make(chan error, 1) |
| 26 | keepAliveErr = ch |
| 27 | go func() { |
| 28 | if err := a.RunKeepAlive(context.Background()); err != nil { |
| 29 | select { |
| 30 | case ch <- err: |
| 31 | default: |
| 32 | } |
| 33 | } |
| 34 | }() |
| 35 | } |
| 36 | |
| 37 | var wg sync.WaitGroup |
| 38 | var exit bool |
| 39 | var finalizeReason string |
| 40 | |
| 41 | var auditTimer *time.Timer |
| 42 | var auditTimerCh <-chan time.Time |
| 43 | if mode := a.AuditDuration(); mode > 0 { |
| 44 | auditTimer = time.NewTimer(mode) |
| 45 | auditTimerCh = auditTimer.C |
| 46 | defer auditTimer.Stop() |
| 47 | } |
| 48 | |
| 49 | for { |
| 50 | collectorapi.ObsoleteCharts(true) |
| 51 | |
| 52 | ctx, cancel := context.WithCancel(context.Background()) |
| 53 | runDone := make(chan struct{}) |
| 54 | wg.Go(func() { |
| 55 | defer close(runDone) |
| 56 | a.RunContext(ctx) |
| 57 | }) |
| 58 | |
| 59 | select { |
| 60 | case sig := <-ch: |
| 61 | switch sig { |
| 62 | case syscall.SIGHUP: |
| 63 | a.Infof("received %s signal (%d). Restarting running instance", sig, sig) |
| 64 | default: |
| 65 | a.Infof("received %s signal (%d). Terminating...", sig, sig) |
| 66 | exit = true |
| 67 | finalizeReason = sig.String() |
| 68 | } |
| 69 | case <-a.QuitCh(): |
| 70 | a.Infof("received QUIT command. Terminating...") |
| 71 | exit = true |
| 72 | finalizeReason = "quit" |
| 73 | case <-auditTimerCh: |
| 74 | a.Infof("metrics-audit duration expired, finalizing metrics audit...") |
| 75 | exit = true |
| 76 | finalizeReason = "audit timer expired" |
| 77 | case <-keepAliveErr: |
| 78 | a.Info("too many keepAlive errors. Terminating...") |
| 79 | exit = true |
| 80 | finalizeReason = "keepalive error" |
| 81 | case <-runDone: |
| 82 | a.Info("agent run loop stopped. Terminating...") |
| 83 | exit = true |
| 84 | finalizeReason = "run loop stopped" |
| 85 | } |
| 86 | |
| 87 | if exit { |
| 88 | collectorapi.ObsoleteCharts(false) |
| 89 | } |
| 90 | |
| 91 | cancel() |
| 92 | |
| 93 | stopped := func() bool { |
| 94 | timeout := time.Second * 10 |
| 95 | t := time.NewTimer(timeout) |
| 96 | defer t.Stop() |
| 97 | done := make(chan struct{}) |
| 98 | |
| 99 | go func() { wg.Wait(); close(done) }() |
| 100 | |
| 101 | select { |
| 102 | case <-t.C: |
| 103 | a.Errorf("stopping all goroutines timed out after %s. Exiting...", timeout) |
| 104 | return false |
| 105 | case <-done: |
| 106 | return true |
| 107 | } |
| 108 | }() |
| 109 | |
| 110 | if !stopped { |
| 111 | if exit { |
| 112 | a.FinalizeMetricsAudit(finalizeReason + ", forced shutdown") |
| 113 | } |
| 114 | os.Exit(0) |
| 115 | } |
| 116 | |
| 117 | if exit { |
| 118 | a.FinalizeMetricsAudit(finalizeReason) |
| 119 | os.Exit(0) |
| 120 | } |
| 121 | |
| 122 | time.Sleep(time.Second) |
| 123 | } |
| 124 | } |