| 1 | // Package lbhealth is a box's own liveness, as the load balancer's health |
| 2 | // check reads it (region HA spec 2 §6). Every fact here is this box's own: |
| 3 | // a flag or a timestamp that one of its own loops sets. Nothing here reads |
| 4 | // a disk, a bucket or the network, so an outage of something the boxes |
| 5 | // share (the deployment bucket) never turns a working box unhealthy and |
| 6 | // never gets it terminated. |
| 7 | // |
| 8 | // The edge answers GET or HEAD /.carlos/lb-health on port 80 from State. |
| 9 | // The loops that feed it are the agent's: the claim agent sets hosting, |
| 10 | // the home map latches forwarding and stamps its refresh attempts, the |
| 11 | // certificate mirror latches its first load, and the edge's registry watch |
| 12 | // and the activator's sweeper each beat. |
| 13 | package lbhealth |
| 14 | |
| 15 | import ( |
| 16 | "io" |
| 17 | "net/http" |
| 18 | "strings" |
| 19 | "sync" |
| 20 | "sync/atomic" |
| 21 | "time" |
| 22 | ) |
| 23 | |
| 24 | // Path is the health check's path, matched exactly. |
| 25 | const Path = "/.carlos/lb-health" |
| 26 | |
| 27 | // HeartbeatWindow is how recent the edge's and the activator's beats must |
| 28 | // be. Both loops beat every 30 seconds or faster. |
| 29 | const HeartbeatWindow = 60 * time.Second |
| 30 | |
| 31 | // RefreshWindow is how recent the home map's last completed refresh |
| 32 | // attempt must be: its 30-second interval plus its two-minute request |
| 33 | // timeout, with room to spare. A failed attempt counts, so a bucket outage |
| 34 | // where every attempt fails fast keeps this satisfied; a worker stuck |
| 35 | // inside one attempt does not. |
| 36 | const RefreshWindow = 5 * time.Minute |
| 37 | |
| 38 | // The condition names a 503 lists, in this order. |
| 39 | const ( |
| 40 | CondHosting = "hosting" |
| 41 | CondForwarding = "forwarding" |
| 42 | CondCerts = "certificates" |
| 43 | CondEdge = "edge" |
| 44 | CondActivator = "activator" |
| 45 | CondRefresh = "refresh" |
| 46 | ) |
| 47 | |
| 48 | // State is one box's liveness facts. The zero value is a box that has |
| 49 | // proven nothing, and answers 503. A nil *State ignores every update. |
| 50 | type State struct { |
| 51 | hosting atomic.Bool |
| 52 | mapLoaded atomic.Bool |
| 53 | keyLoaded atomic.Bool |
| 54 | certs atomic.Bool |
| 55 | edgeBeat atomic.Int64 // unix nanoseconds; 0 is never |
| 56 | refreshed atomic.Int64 // unix nanoseconds; 0 is never |
| 57 | |
| 58 | mu sync.RWMutex |
| 59 | activator func() time.Time |
| 60 | } |
| 61 | |
| 62 | // Default is the process's state. The agent runs the edge and the |
| 63 | // activator in one process, so the claim wiring, the edge and the |
| 64 | // activator share it. |
| 65 | var Default = &State{} |
| 66 | |
| 67 | // SetHosting records whether the agent is in hosting mode: enrolled, with |
| 68 | // its guard armed (§5). Region HA spec 2's plan 2b adds joining mode, whose |
| 69 | // mode selection is what sets this false; until then the claim agent sets |
| 70 | // it true once its claim wiring is running (ClaimWiring.startRegional). |
| 71 | func (s *State) SetHosting(on bool) { |
| 72 | if s != nil { |
| 73 | s.hosting.Store(on) |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | // Hosting reports what SetHosting last recorded. |
| 78 | func (s *State) Hosting() bool { return s != nil && s.hosting.Load() } |
| 79 | |
| 80 | // MapLoaded latches the first successful load of the forwarding map. |
| 81 | func (s *State) MapLoaded() { |
| 82 | if s != nil { |
| 83 | s.mapLoaded.Store(true) |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | // KeyLoaded latches the first successful load of a mesh key. |
| 88 | func (s *State) KeyLoaded() { |
| 89 | if s != nil { |
| 90 | s.keyLoaded.Store(true) |
| 91 | } |
| 92 | } |
| 93 | |
| 94 | // CertsLoaded latches the first successful load of the certificate mirror |
| 95 | // into the local cache the TLS handshake reads. |
| 96 | func (s *State) CertsLoaded() { |
| 97 | if s != nil { |
| 98 | s.certs.Store(true) |
| 99 | } |
| 100 | } |
| 101 | |
| 102 | // EdgeBeat records that the edge's registry watch completed a poll at t. |
| 103 | func (s *State) EdgeBeat(t time.Time) { |
| 104 | if s != nil { |
| 105 | s.edgeBeat.Store(t.UnixNano()) |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | // RefreshAttempted records that the home map's refresh finished an |
| 110 | // attempt at t, whether it succeeded or failed. |
| 111 | func (s *State) RefreshAttempted(t time.Time) { |
| 112 | if s != nil { |
| 113 | s.refreshed.Store(t.UnixNano()) |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | // SetActivator gives the state the activator's own beat: the time its |
| 118 | // sweeper last finished a sweep. nil means there is no activator, which |
| 119 | // fails the activator condition. |
| 120 | func (s *State) SetActivator(last func() time.Time) { |
| 121 | if s == nil { |
| 122 | return |
| 123 | } |
| 124 | s.mu.Lock() |
| 125 | s.activator = last |
| 126 | s.mu.Unlock() |
| 127 | } |
| 128 | |
| 129 | func (s *State) activatorBeat() time.Time { |
| 130 | s.mu.RLock() |
| 131 | last := s.activator |
| 132 | s.mu.RUnlock() |
| 133 | if last == nil { |
| 134 | return time.Time{} |
| 135 | } |
| 136 | return last() |
| 137 | } |
| 138 | |
| 139 | // within reports whether at is set and no older than window at now. |
| 140 | func within(at, now time.Time, window time.Duration) bool { |
| 141 | return !at.IsZero() && now.Sub(at) <= window |
| 142 | } |
| 143 | |
| 144 | func stamp(n int64) time.Time { |
| 145 | if n == 0 { |
| 146 | return time.Time{} |
| 147 | } |
| 148 | return time.Unix(0, n) |
| 149 | } |
| 150 | |
| 151 | // Failing lists the conditions that do not hold at now, in the order of |
| 152 | // the Cond names. An empty list is a healthy box. A nil *State fails |
| 153 | // every condition. |
| 154 | func (s *State) Failing(now time.Time) []string { |
| 155 | if s == nil { |
| 156 | return []string{CondHosting, CondForwarding, CondCerts, CondEdge, CondActivator, CondRefresh} |
| 157 | } |
| 158 | var out []string |
| 159 | if !s.hosting.Load() { |
| 160 | out = append(out, CondHosting) |
| 161 | } |
| 162 | if !s.mapLoaded.Load() || !s.keyLoaded.Load() { |
| 163 | out = append(out, CondForwarding) |
| 164 | } |
| 165 | if !s.certs.Load() { |
| 166 | out = append(out, CondCerts) |
| 167 | } |
| 168 | if !within(stamp(s.edgeBeat.Load()), now, HeartbeatWindow) { |
| 169 | out = append(out, CondEdge) |
| 170 | } |
| 171 | if !within(s.activatorBeat(), now, HeartbeatWindow) { |
| 172 | out = append(out, CondActivator) |
| 173 | } |
| 174 | if !within(stamp(s.refreshed.Load()), now, RefreshWindow) { |
| 175 | out = append(out, CondRefresh) |
| 176 | } |
| 177 | return out |
| 178 | } |
| 179 | |
| 180 | // Serve answers the health check when r is one, and reports whether it |
| 181 | // did. It matches as narrowly as the steering probe: the exact path, GET |
| 182 | // or HEAD, any host. It answers 200 "ok" when every condition holds and |
| 183 | // 503 naming the failing conditions otherwise, never cached. It only reads |
| 184 | // s. |
| 185 | func Serve(w http.ResponseWriter, r *http.Request, s *State, now time.Time) bool { |
| 186 | if r.URL.Path != Path || (r.Method != http.MethodGet && r.Method != http.MethodHead) { |
| 187 | return false |
| 188 | } |
| 189 | body, code := "ok\n", http.StatusOK |
| 190 | if failing := s.Failing(now); len(failing) > 0 { |
| 191 | body, code = "unavailable: "+strings.Join(failing, ",")+"\n", http.StatusServiceUnavailable |
| 192 | } |
| 193 | w.Header().Set("Content-Type", "text/plain; charset=utf-8") |
| 194 | w.Header().Set("Cache-Control", "no-store") |
| 195 | w.WriteHeader(code) |
| 196 | if r.Method == http.MethodGet { |
| 197 | _, _ = io.WriteString(w, body) |
| 198 | } |
| 199 | return true |
| 200 | } |
| 201 | |
amadan