| 1 | package schedule |
| 2 | |
| 3 | import ( |
| 4 | "crypto/sha256" |
| 5 | "crypto/subtle" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "net/http" |
| 9 | "os" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/registry" |
| 13 | ) |
| 14 | |
| 15 | // TimerStore is narrowed to the five methods Mount actually calls, not the |
| 16 | // whole *registry.Registry: *registry.Registry satisfies it directly |
| 17 | // (internal/host/agent.go), and internal/activator — the other side of the |
| 18 | // package boundary this crosses (Backend.Timers) — stubs it with an |
| 19 | // in-memory fake so its wiring tests never touch a real database. |
| 20 | type TimerStore interface { |
| 21 | UpsertTimer(registry.Timer) error |
| 22 | DeleteTimer(host, name string) error |
| 23 | GetTimer(host, name string) (*registry.Timer, error) |
| 24 | TimersFor(host string) ([]registry.Timer, error) |
| 25 | CountTimers(host, kind string) (int, error) |
| 26 | } |
| 27 | |
| 28 | // RetiredWriteMessage is the 409 body a retiring generation's schedule |
| 29 | // writes get once the switch has happened (zero-downtime spec §7.10). App |
| 30 | // developers read it in their logs; it is copy-reviewed. |
| 31 | const RetiredWriteMessage = "this process has been replaced; the new process owns the schedule" |
| 32 | |
| 33 | // WriteFence serializes one generation's schedule writes against the |
| 34 | // switch that retires it (zero-downtime spec §7.10, final review F8). It |
| 35 | // runs write, the final retirement decision and the mutation together, |
| 36 | // unless the generation has been retired, and reports whether it ran. |
| 37 | // While write runs, the switch that would retire this generation waits; |
| 38 | // once it has retired it, every later call answers false, including one |
| 39 | // for a request that began before the switch. A request's body is parsed |
| 40 | // and validated before the fence is asked, so a slow client never holds |
| 41 | // the switch. |
| 42 | type WriteFence func(write func()) bool |
| 43 | |
| 44 | // Mount is MountGated with no gate. |
| 45 | func Mount(mux *http.ServeMux, host string, store TimerStore, token func() (string, error), now func() time.Time) { |
| 46 | MountGated(mux, host, store, token, now, nil) |
| 47 | } |
| 48 | |
| 49 | // MountGated adds the once-timer verbs to an instance's control socket (spec |
| 50 | // §7.2). host is fixed here and never read from a request — the |
| 51 | // confused-deputy guarantee edge.ListenControl gives — and every verb needs |
| 52 | // the host's own CARLOS_ADMIN_TOKEN as a bearer. Be precise about what that |
| 53 | // bearer buys (internal/edge/purge.go has the long form): it closes the |
| 54 | // channel against OTHER UNIX users on the box and against any |
| 55 | // unauthenticated dialer — reaching the socket is no longer enough. It does |
| 56 | // NOT close the same-uid sibling gap: every exec child shares the agent's |
| 57 | // uid, the token file is 0600 owner-only, so a sibling can read another |
| 58 | // host's token and present it here. That residual is bounded — a sibling |
| 59 | // with host B's token could already dial B's own socket with it; these |
| 60 | // verbs add reach (a platform wake plus a POST at a chosen time and path), |
| 61 | // not privilege — and per-instance uids are the real fix. |
| 62 | // |
| 63 | // fence gates the two writing verbs: every store access a PUT or DELETE |
| 64 | // makes runs inside it, and once it refuses, the verb answers 409 and |
| 65 | // changes nothing. A generation the activator has switched away from must |
| 66 | // not rewrite timers the successor now owns, which is what its |
| 67 | // reassert-at-shutdown would do (zero-downtime spec §7.10). Reads stay |
| 68 | // open, the bearer is checked first, and the request is validated before |
| 69 | // the fence is asked. A nil fence is Mount exactly. |
| 70 | func MountGated(mux *http.ServeMux, host string, store TimerStore, token func() (string, error), now func() time.Time, fence WriteFence) { |
| 71 | // fenced runs write inside the fence, and reports false (the caller |
| 72 | // answers 409) when the generation is retired. |
| 73 | fenced := func(write func()) bool { |
| 74 | if fence == nil { |
| 75 | write() |
| 76 | return true |
| 77 | } |
| 78 | return fence(write) |
| 79 | } |
| 80 | authed := func(fn http.HandlerFunc) http.HandlerFunc { |
| 81 | return func(w http.ResponseWriter, r *http.Request) { |
| 82 | want, err := token() |
| 83 | if err != nil || want == "" { |
| 84 | http.Error(w, "unauthorized", http.StatusUnauthorized) |
| 85 | return |
| 86 | } |
| 87 | got := sha256.Sum256([]byte(r.Header.Get("Authorization"))) |
| 88 | exp := sha256.Sum256([]byte("Bearer " + want)) |
| 89 | if subtle.ConstantTimeCompare(got[:], exp[:]) != 1 { |
| 90 | http.Error(w, "unauthorized", http.StatusUnauthorized) |
| 91 | return |
| 92 | } |
| 93 | fn(w, r) |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | mux.HandleFunc("GET /schedule-capabilities", authed(func(w http.ResponseWriter, r *http.Request) { |
| 98 | w.Header().Set("Content-Type", "application/json") |
| 99 | json.NewEncoder(w).Encode(struct { |
| 100 | RegistrationFence bool `json:"registration_fence"` |
| 101 | }{RegistrationFence: true}) |
| 102 | })) |
| 103 | |
| 104 | mux.HandleFunc("PUT /schedule/{name}", authed(func(w http.ResponseWriter, r *http.Request) { |
| 105 | name := r.PathValue("name") |
| 106 | if !ValidName(name) { |
| 107 | http.Error(w, "name: want [a-z0-9][a-z0-9-]{0,31}", http.StatusBadRequest) |
| 108 | return |
| 109 | } |
| 110 | var body struct { |
| 111 | At string `json:"at"` |
| 112 | Path string `json:"path"` |
| 113 | } |
| 114 | dec := json.NewDecoder(http.MaxBytesReader(w, r.Body, 4096)) |
| 115 | if err := dec.Decode(&body); err != nil { |
| 116 | http.Error(w, "bad request: "+err.Error(), http.StatusBadRequest) |
| 117 | return |
| 118 | } |
| 119 | // Decode stops at the first JSON value and silently ignores |
| 120 | // whatever follows it; a client that appended a second object (or |
| 121 | // stray bytes) meant something by that, so refuse rather than act |
| 122 | // on the first value alone. |
| 123 | if dec.More() { |
| 124 | http.Error(w, "bad request: trailing data after the JSON body", http.StatusBadRequest) |
| 125 | return |
| 126 | } |
| 127 | atT, err := time.Parse(time.RFC3339, body.At) |
| 128 | if err != nil { |
| 129 | http.Error(w, "at: want RFC3339, e.g. 2026-09-01T08:00:00Z", http.StatusBadRequest) |
| 130 | return |
| 131 | } |
| 132 | if atT.After(now().Add(MaxOnceAhead)) { |
| 133 | http.Error(w, "at: more than 400 days ahead", http.StatusBadRequest) |
| 134 | return |
| 135 | } |
| 136 | if err := ValidPath(body.Path); err != nil { |
| 137 | http.Error(w, err.Error(), http.StatusBadRequest) |
| 138 | return |
| 139 | } |
| 140 | // The store's checks and the write run inside the fence, so the |
| 141 | // retirement decision and the mutation are one step against the |
| 142 | // switch. The response is written after it. |
| 143 | var ( |
| 144 | fail string |
| 145 | code int |
| 146 | row *registry.Timer |
| 147 | ) |
| 148 | ran := fenced(func() { |
| 149 | existing, err := store.GetTimer(host, name) |
| 150 | switch { |
| 151 | case err == nil && existing.Kind != KindOnce: |
| 152 | fail, code = "name is a declared schedule", http.StatusConflict |
| 153 | return |
| 154 | case err != nil && !errors.Is(err, os.ErrNotExist): |
| 155 | fail, code = "registry error", http.StatusInternalServerError |
| 156 | return |
| 157 | case err != nil: |
| 158 | n, cerr := store.CountTimers(host, KindOnce) |
| 159 | if cerr != nil { |
| 160 | fail, code = "registry error", http.StatusInternalServerError |
| 161 | return |
| 162 | } |
| 163 | if n >= MaxOncePerHost { |
| 164 | fail, code = "too many one-shot timers (limit 1000) — delete finished ones or use a schedule", http.StatusTooManyRequests |
| 165 | return |
| 166 | } |
| 167 | } |
| 168 | sp := Spec{Kind: KindOnce, At: atT.UTC()} |
| 169 | next := atT.Unix() |
| 170 | if next <= 0 { |
| 171 | next = max(int64(1), now().Unix()) |
| 172 | } |
| 173 | t := registry.Timer{Host: host, Name: name, Kind: KindOnce, Spec: sp.String(), Path: body.Path, NextAt: next, DueAt: atT.Unix(), CreatedAt: now().Unix()} |
| 174 | if err := store.UpsertTimer(t); err != nil { |
| 175 | fail, code = "registry error", http.StatusInternalServerError |
| 176 | return |
| 177 | } |
| 178 | if row, err = store.GetTimer(host, name); err != nil { |
| 179 | fail, code = "registry error", http.StatusInternalServerError |
| 180 | } |
| 181 | }) |
| 182 | switch { |
| 183 | case !ran: |
| 184 | http.Error(w, RetiredWriteMessage, http.StatusConflict) |
| 185 | case fail != "": |
| 186 | http.Error(w, fail, code) |
| 187 | default: |
| 188 | writeTimer(w, *row, now()) |
| 189 | } |
| 190 | })) |
| 191 | |
| 192 | mux.HandleFunc("DELETE /schedule/{name}", authed(func(w http.ResponseWriter, r *http.Request) { |
| 193 | name := r.PathValue("name") |
| 194 | if !ValidName(name) { |
| 195 | http.Error(w, "bad name", http.StatusBadRequest) |
| 196 | return |
| 197 | } |
| 198 | var ( |
| 199 | fail string |
| 200 | code int |
| 201 | ) |
| 202 | ran := fenced(func() { |
| 203 | switch row, err := store.GetTimer(host, name); { |
| 204 | case err == nil && row.Kind != KindOnce: |
| 205 | fail, code = "name is a declared schedule — remove it with carlos schedule rm", http.StatusConflict |
| 206 | return |
| 207 | case err != nil && !errors.Is(err, os.ErrNotExist): |
| 208 | fail, code = "registry error", http.StatusInternalServerError |
| 209 | return |
| 210 | } |
| 211 | if err := store.DeleteTimer(host, name); err != nil { |
| 212 | fail, code = "registry error", http.StatusInternalServerError |
| 213 | } |
| 214 | }) |
| 215 | switch { |
| 216 | case !ran: |
| 217 | http.Error(w, RetiredWriteMessage, http.StatusConflict) |
| 218 | case fail != "": |
| 219 | http.Error(w, fail, code) |
| 220 | default: |
| 221 | w.WriteHeader(http.StatusNoContent) |
| 222 | } |
| 223 | })) |
| 224 | |
| 225 | mux.HandleFunc("GET /schedule", authed(func(w http.ResponseWriter, r *http.Request) { |
| 226 | rows, err := store.TimersFor(host) |
| 227 | if err != nil { |
| 228 | http.Error(w, "registry error", http.StatusInternalServerError) |
| 229 | return |
| 230 | } |
| 231 | out := struct { |
| 232 | Timers []TimerStatus `json:"timers"` |
| 233 | }{Timers: []TimerStatus{}} |
| 234 | for _, row := range rows { |
| 235 | // The control socket has no view of the activator's idle window |
| 236 | // (it is reached from inside the exec child, not the box's |
| 237 | // hibernation wiring), so it makes no holds_awake claim — same |
| 238 | // as any box that predates the field. |
| 239 | out.Timers = append(out.Timers, toStatus(row, 0, now())) |
| 240 | } |
| 241 | w.Header().Set("Content-Type", "application/json") |
| 242 | json.NewEncoder(w).Encode(out) |
| 243 | })) |
| 244 | } |
| 245 | |
| 246 | // writeTimer writes one registry row as its wire TimerStatus (toStatus is |
| 247 | // converge.go's — one row-to-status mapping for the whole package). idle is |
| 248 | // 0 here too, for the reason the /schedule GET handler above documents: the |
| 249 | // control socket cannot see the activator's window. |
| 250 | func writeTimer(w http.ResponseWriter, r registry.Timer, now time.Time) { |
| 251 | w.Header().Set("Content-Type", "application/json") |
| 252 | json.NewEncoder(w).Encode(toStatus(r, 0, now)) |
| 253 | } |
| 254 | |
amadan