| 1 | package edgesfile |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "log" |
| 6 | "slices" |
| 7 | "time" |
| 8 | |
| 9 | "github.com/carlosframework/platform/internal/release" |
| 10 | "github.com/carlosframework/platform/internal/status" |
| 11 | ) |
| 12 | |
| 13 | // EdgeIPs is the deployment's live edge address set: every EIP published by a |
| 14 | // box whose STATUS SNAPSHOT is still fresh, unioned with env (the operator's |
| 15 | // PLATFORM_EDGE_IPS). |
| 16 | // |
| 17 | // The union is the whole failure story. Nothing here deletes anything, so the |
| 18 | // fold can only WIDEN what counts as "pointing at us" — which means a short |
| 19 | // fold, an unreadable store, or a deployment with no snapshots at all |
| 20 | // degrades to exactly the pre-fold behaviour (env only) and never below it. |
| 21 | // Contrast internal/steering, which fails CLOSED on a short fold because a |
| 22 | // short fold THERE deletes DNS records. |
| 23 | // |
| 24 | // Liveness is the status snapshot, never the edges file's own Generation: |
| 25 | // Generation moves only when the file is WRITTEN, and the writer's change |
| 26 | // gate exists precisely so an unchanged file is never republished — so a |
| 27 | // healthy box's Generation is arbitrarily old and would fail any age bound. |
| 28 | // statusTick's WriteBox is unconditional every tick, so BoxStale is a real |
| 29 | // liveness signal. Without one, a retired box's lingering edges file would |
| 30 | // keep a DEAD customer domain reading "pointing" forever. |
| 31 | // |
| 32 | // LIMITATION, and the reason for the log line below: that join reads both |
| 33 | // halves out of the ONE store it is handed, but the two halves are written |
| 34 | // through different ones. Edges files go through openDeployStore |
| 35 | // (CARLOS_DEPLOYMENT_BUCKET, unprefixed); box snapshots go through |
| 36 | // openStatusStore, which honours CARLOS_STATUS_BUCKET and rebases everything |
| 37 | // under CARLOS_STATUS_PREFIX. They are the same store by DEFAULT, not by |
| 38 | // guarantee — README.md's fleet section suggests a separate status bucket |
| 39 | // outright — and on a deployment that splits them nothing here can see the |
| 40 | // other half, so every edges file is dropped as unproven and the answer is |
| 41 | // the env list. That degradation is SAFE (env-only is exactly the |
| 42 | // pre-fold behaviour) but it is INVISIBLE, which is the failure class this |
| 43 | // whole rule exists to avoid. So when files were found and none of them |
| 44 | // joined, say so — configuration is the likely cause and no test can catch |
| 45 | // it. |
| 46 | func EdgeIPs(ctx context.Context, s release.Store, env []string, now time.Time) []string { |
| 47 | out := append([]string(nil), env...) |
| 48 | add := func(ip string) { |
| 49 | if ip != "" && !slices.Contains(out, ip) { |
| 50 | out = append(out, ip) |
| 51 | } |
| 52 | } |
| 53 | if s == nil { |
| 54 | return out |
| 55 | } |
| 56 | live := map[string]bool{} |
| 57 | boxes, err := status.LoadBoxes(ctx, s) |
| 58 | if err != nil { |
| 59 | log.Printf("edgesfile: could not load box snapshots (%v) — edge set falls back to the configured list", err) |
| 60 | return out |
| 61 | } |
| 62 | for _, b := range boxes { |
| 63 | if !status.BoxStale(b, now, status.SnapshotPeriod) { |
| 64 | live[b.Label] = true |
| 65 | } |
| 66 | } |
| 67 | keys, err := s.List(ctx, Prefix) |
| 68 | if err != nil { |
| 69 | log.Printf("edgesfile: could not list %s (%v) — edge set falls back to the configured list", Prefix, err) |
| 70 | return out |
| 71 | } |
| 72 | joined := 0 |
| 73 | for _, key := range keys { |
| 74 | body, _, err := s.Get(ctx, key) |
| 75 | if err != nil { |
| 76 | log.Printf("edgesfile: %s: %v", key, err) |
| 77 | continue |
| 78 | } |
| 79 | f, err := Parse(body) |
| 80 | if err != nil { |
| 81 | log.Printf("edgesfile: %s: %v", key, err) |
| 82 | continue |
| 83 | } |
| 84 | if !live[f.Box] { |
| 85 | continue // no fresh snapshot: not evidence of a live edge |
| 86 | } |
| 87 | joined++ |
| 88 | add(f.EIP) |
| 89 | } |
| 90 | // Files exist and not one of them proved live. A single retired box is |
| 91 | // the ordinary reading of a partial miss, but a TOTAL miss usually is |
| 92 | // not: either every box in the deployment went dark at once, or the two |
| 93 | // halves of the join are looking at different buckets (see the doc |
| 94 | // comment), or the box names on the two sides have drifted apart. All |
| 95 | // three are silent otherwise — the answer is simply the env list, with |
| 96 | // no error to return and nothing red anywhere. This line is the |
| 97 | // tripwire. Nothing is logged when there are no files at all: a |
| 98 | // deployment that has not published any is an ordinary state, not a |
| 99 | // fault. |
| 100 | if len(keys) > 0 && joined == 0 { |
| 101 | log.Printf("edgesfile: %d edges file(s) under %s but no fresh box snapshot matched any of them — edge set is the configured list only (are edges and status in different buckets? see CARLOS_STATUS_BUCKET/CARLOS_STATUS_PREFIX)", len(keys), Prefix) |
| 102 | } |
| 103 | return out |
| 104 | } |
| 105 | |
| 106 | // DeploymentRoutes is every hostname a LIVE box publishes as a real route — |
| 107 | // a home entry WITHOUT the alias mark. It is the evidence a hostname claim is |
| 108 | // checked against before it may materialize as an alias. |
| 109 | // |
| 110 | // Alias entries are excluded deliberately and the exclusion is the whole |
| 111 | // point: an alias IS a materialized claim, so counting one would make every |
| 112 | // claim reserve its own hostname against itself and no re-attach or |
| 113 | // re-derivation would ever succeed again. Only a NON-alias home entry — an |
| 114 | // operator-minted or reconciler-minted route row — is evidence that some |
| 115 | // other tenant's traffic already belongs to that name. |
| 116 | // |
| 117 | // Live is the same box-snapshot rule EdgeIPs uses, and for the same reason: a |
| 118 | // retired box's lingering file must not go on reserving hostnames forever. |
| 119 | // |
| 120 | // Best-effort by contract: a store that cannot be read returns nil, and a nil |
| 121 | // answer means "nothing known", which callers must treat as permissive. The |
| 122 | // per-box byHost check and the fold's route preference are the other two |
| 123 | // layers; this one is the only one that can act across boxes, but it must not |
| 124 | // be able to break ordinary attach when the bucket is unavailable. |
| 125 | func DeploymentRoutes(ctx context.Context, s release.Store, now time.Time) map[string]bool { |
| 126 | if s == nil { |
| 127 | return nil |
| 128 | } |
| 129 | live := map[string]bool{} |
| 130 | boxes, err := status.LoadBoxes(ctx, s) |
| 131 | if err != nil { |
| 132 | log.Printf("edgesfile: could not load box snapshots (%v) — deployment route set unknown this pass", err) |
| 133 | return nil |
| 134 | } |
| 135 | for _, b := range boxes { |
| 136 | if !status.BoxStale(b, now, status.SnapshotPeriod) { |
| 137 | live[b.Label] = true |
| 138 | } |
| 139 | } |
| 140 | keys, err := s.List(ctx, Prefix) |
| 141 | if err != nil { |
| 142 | log.Printf("edgesfile: could not list %s (%v) — deployment route set unknown this pass", Prefix, err) |
| 143 | return nil |
| 144 | } |
| 145 | out := map[string]bool{} |
| 146 | joined := 0 |
| 147 | for _, key := range keys { |
| 148 | body, _, err := s.Get(ctx, key) |
| 149 | if err != nil { |
| 150 | log.Printf("edgesfile: %s: %v", key, err) |
| 151 | continue |
| 152 | } |
| 153 | f, err := Parse(body) |
| 154 | if err != nil { |
| 155 | log.Printf("edgesfile: %s: %v", key, err) |
| 156 | continue |
| 157 | } |
| 158 | if !live[f.Box] { |
| 159 | continue // no fresh snapshot: not evidence of a live route |
| 160 | } |
| 161 | joined++ |
| 162 | for host, h := range f.Hosts { |
| 163 | if h.Role == RoleHome && !h.Alias { |
| 164 | out[host] = true |
| 165 | } |
| 166 | } |
| 167 | } |
| 168 | // EdgeIPs' tripwire, and it matters MORE here. There a total miss |
| 169 | // degrades to the operator's configured edge list, which is merely the |
| 170 | // pre-fold behaviour; here it degrades to "no hostname is anyone's", |
| 171 | // which silently disarms a SECURITY guard — the answer is an empty map, |
| 172 | // no error, nothing red, and every claim materializes exactly as it did |
| 173 | // before this rule existed. The likely causes are the same three |
| 174 | // (deployment vs status bucket split, every box dark at once, box-name |
| 175 | // drift), and they are all invisible otherwise. Silent on zero files: a |
| 176 | // deployment that has published none is an ordinary state, not a fault. |
| 177 | if len(keys) > 0 && joined == 0 { |
| 178 | log.Printf("edgesfile: %d edges file(s) under %s but no fresh box snapshot matched any of them — the deployment route set is EMPTY this pass, so hostname claims are unchecked against other boxes' routes (are edges and status in different buckets? see CARLOS_STATUS_BUCKET/CARLOS_STATUS_PREFIX)", len(keys), Prefix) |
| 179 | } |
| 180 | return out |
| 181 | } |
| 182 | |
amadan