| Tim Windelschmidt | 6d33a43 | 2025-02-04 14:34:25 +0100 | [diff] [blame] | 1 | // Copyright The Monogon Project Authors. |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| Serge Bazanski | 6f59951 | 2023-04-26 19:08:19 +0200 | [diff] [blame] | 4 | // Package scruffy implements Scruffy, The Janitor. |
| 5 | // |
| 6 | // Scruffy is a BMaaS component which runs a bunch of important, housekeeping-ish |
| 7 | // processes that aren't tied to any particular provider and are mostly |
| 8 | // batch-oriented. |
| 9 | // |
| 10 | // Currently Scruffy just collects metrics from the BMDB. |
| 11 | package scruffy |
| 12 | |
| 13 | import ( |
| 14 | "context" |
| Tim Windelschmidt | d5f851b | 2024-04-23 14:59:37 +0200 | [diff] [blame] | 15 | "errors" |
| Serge Bazanski | 6f59951 | 2023-04-26 19:08:19 +0200 | [diff] [blame] | 16 | "flag" |
| 17 | "fmt" |
| 18 | "os" |
| 19 | "time" |
| 20 | |
| 21 | "github.com/cenkalti/backoff/v4" |
| 22 | "k8s.io/klog/v2" |
| 23 | |
| 24 | "source.monogon.dev/cloud/bmaas/bmdb" |
| 25 | "source.monogon.dev/cloud/bmaas/bmdb/metrics" |
| 26 | "source.monogon.dev/cloud/bmaas/bmdb/webug" |
| 27 | "source.monogon.dev/cloud/lib/component" |
| 28 | ) |
| 29 | |
| 30 | type Config struct { |
| 31 | Component component.ComponentConfig |
| 32 | BMDB bmdb.BMDB |
| 33 | Webug webug.Config |
| 34 | |
| 35 | StatsRunnerRate time.Duration |
| 36 | } |
| 37 | |
| 38 | // TODO(q3k): factor this out to BMDB library? |
| 39 | func runtimeInfo() string { |
| 40 | hostname, _ := os.Hostname() |
| 41 | if hostname == "" { |
| 42 | hostname = "UNKNOWN" |
| 43 | } |
| 44 | return fmt.Sprintf("host %s", hostname) |
| 45 | } |
| 46 | |
| 47 | func (c *Config) RegisterFlags() { |
| 48 | c.Component.RegisterFlags("scruffy") |
| 49 | c.BMDB.ComponentName = "scruffy" |
| 50 | c.BMDB.RuntimeInfo = runtimeInfo() |
| 51 | c.BMDB.Database.RegisterFlags("bmdb") |
| 52 | c.Webug.RegisterFlags() |
| 53 | |
| 54 | flag.DurationVar(&c.StatsRunnerRate, "scruffy_stats_collection_rate", time.Minute, "How often the stats collection loop will run against BMDB") |
| 55 | } |
| 56 | |
| 57 | type Server struct { |
| 58 | Config Config |
| 59 | |
| 60 | bmdb *bmdb.Connection |
| 61 | sessionC chan *bmdb.Session |
| 62 | } |
| 63 | |
| 64 | func (s *Server) Start(ctx context.Context) { |
| 65 | reg := s.Config.Component.PrometheusRegistry() |
| 66 | s.Config.BMDB.EnableMetrics(reg) |
| 67 | s.Config.Component.StartPrometheus(ctx) |
| 68 | |
| 69 | conn, err := s.Config.BMDB.Open(true) |
| 70 | if err != nil { |
| 71 | klog.Exitf("Failed to connect to BMDB: %v", err) |
| 72 | } |
| 73 | s.bmdb = conn |
| 74 | s.sessionC = make(chan *bmdb.Session) |
| 75 | go s.sessionWorker(ctx) |
| 76 | |
| 77 | bsr := newBMDBStatsRunner(s, reg) |
| 78 | go bsr.run(ctx) |
| 79 | |
| 80 | hwr := newHWStatsRunner(s, reg) |
| 81 | go hwr.run(ctx) |
| 82 | |
| 83 | go func() { |
| Tim Windelschmidt | d5f851b | 2024-04-23 14:59:37 +0200 | [diff] [blame] | 84 | if err := s.Config.Webug.Start(ctx, conn); err != nil && !errors.Is(err, ctx.Err()) { |
| Serge Bazanski | 6f59951 | 2023-04-26 19:08:19 +0200 | [diff] [blame] | 85 | klog.Exitf("Failed to start webug: %v", err) |
| 86 | } |
| 87 | }() |
| 88 | } |
| 89 | |
| 90 | // sessionWorker emits a valid BMDB session to sessionC as long as ctx is active. |
| 91 | // |
| 92 | // TODO(q3k): factor out into bmdb client lib |
| 93 | func (s *Server) sessionWorker(ctx context.Context) { |
| 94 | var session *bmdb.Session |
| 95 | for { |
| 96 | if session == nil || session.Expired() { |
| 97 | klog.Infof("Starting new session...") |
| 98 | bo := backoff.NewExponentialBackOff() |
| 99 | err := backoff.Retry(func() error { |
| 100 | var err error |
| 101 | session, err = s.bmdb.StartSession(ctx, bmdb.SessionOption{Processor: metrics.ProcessorScruffyStats}) |
| 102 | if err != nil { |
| 103 | klog.Errorf("Failed to start session: %v", err) |
| 104 | return err |
| 105 | } else { |
| 106 | return nil |
| 107 | } |
| 108 | }, backoff.WithContext(bo, ctx)) |
| 109 | if err != nil { |
| 110 | // If something's really wrong just crash. |
| 111 | klog.Exitf("Gave up on starting session: %v", err) |
| 112 | } |
| 113 | klog.Infof("New session: %s", session.UUID) |
| 114 | } |
| 115 | |
| 116 | select { |
| 117 | case <-ctx.Done(): |
| 118 | return |
| 119 | case s.sessionC <- session: |
| 120 | } |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | func (s *Server) session(ctx context.Context) (*bmdb.Session, error) { |
| 125 | select { |
| 126 | case sess := <-s.sessionC: |
| 127 | return sess, nil |
| 128 | case <-ctx.Done(): |
| 129 | return nil, ctx.Err() |
| 130 | } |
| 131 | } |