| 1 | // Cert warm (#299 stage 3): the platform orders and RENEWS certificates for |
| 2 | // the custom domains it serves, instead of waiting for a visitor's handshake |
| 3 | // to trigger the first issuance. |
| 4 | // |
| 5 | // Why this exists at all. Per-host certs are minted lazily by autocert on the |
| 6 | // first TLS handshake for a name. That is fine for a name people visit from |
| 7 | // the box that serves it, and wrong for a custom domain on a steering |
| 8 | // deployment, for two compounding reasons: |
| 9 | // |
| 10 | // - The FIRST visitor pays a multi-second ACME order, and if the customer's |
| 11 | // DNS only just landed, they are also the one who discovers it failed. |
| 12 | // - RENEWAL is worse. autocert renews through per-host `domainRenewal` |
| 13 | // timers that live in process memory. They die on every fleet roll and |
| 14 | // re-arm only when a handshake reaches m.cert again. Under latency |
| 15 | // steering a custom domain's visitors may never land on the issuer, so |
| 16 | // after a roll NOTHING re-arms the timer and the cert quietly expires at |
| 17 | // day 90 — on a host that was serving perfectly the whole time. |
| 18 | // |
| 19 | // So the warm calls GetCertificate for every host in its set on EVERY pass, |
| 20 | // unconditionally. That is not wasteful and it is not a retry loop: for a |
| 21 | // healthy cert it is one cache read plus the startRenew that re-arms the |
| 22 | // timer, and the re-arm is the entire point. A skip-if-fresh check would |
| 23 | // defeat it — autocert's m.cert returns any unexpired cached cert, so a cert |
| 24 | // 29 days from expiry is a cache hit and no renewal is ever scheduled. |
| 25 | package edge |
| 26 | |
| 27 | import ( |
| 28 | "context" |
| 29 | "crypto/tls" |
| 30 | "errors" |
| 31 | "log" |
| 32 | "net" |
| 33 | "sort" |
| 34 | "strings" |
| 35 | "sync" |
| 36 | "time" |
| 37 | |
| 38 | "golang.org/x/crypto/acme" |
| 39 | |
| 40 | "github.com/carlosframework/platform/internal/certwarm" |
| 41 | "github.com/carlosframework/platform/internal/release" |
| 42 | ) |
| 43 | |
| 44 | const ( |
| 45 | // warmInterval is the quiet-path cadence. Renewal is the job with the |
| 46 | // real deadline (autocert renews at ~30 days remaining), so hourly is |
| 47 | // far more often than strictly needed — the cost of a pass over a |
| 48 | // healthy set is a cache read per host, and being early is free. |
| 49 | warmInterval = time.Hour |
| 50 | // warmBackoff is the per-host floor after a failed order. Matches the |
| 51 | // wildcard manager's hourly retry. |
| 52 | warmBackoff = time.Hour |
| 53 | // warmRateLimitBackoff is the floor after the CA says rateLimited. |
| 54 | // Deliberately much harder than warmBackoff: hourly retries INTO a rate |
| 55 | // limit are what turn a transient limit into a sustained one, and every |
| 56 | // host in the deployment shares the one ACME account's budget. |
| 57 | warmRateLimitBackoff = 12 * time.Hour |
| 58 | // warmOrdersPerPass bounds NEW orders per pass. Renewal re-arms are not |
| 59 | // orders and are never capped — they are the reason the pass exists. |
| 60 | // When the cap bites, pass() returns a short delay so a backlog drains |
| 61 | // in minutes rather than trickling out over days (the wildcard pass() |
| 62 | // precedent). |
| 63 | warmOrdersPerPass = 10 |
| 64 | // warmCappedDelay is how soon to come back with capped work remaining. |
| 65 | warmCappedDelay = time.Minute |
| 66 | ) |
| 67 | |
| 68 | // certWarm proactively orders and renews per-host certificates. |
| 69 | // |
| 70 | // It is a sibling of wildcardManager, with three deliberate departures from |
| 71 | // it, each of which is a bug if copied across: |
| 72 | // |
| 73 | // - No holding-context claim. The wildcard manager runs its own ACME |
| 74 | // client, so it can hand the holding's context to the order. The warm's |
| 75 | // trigger is autocert's GetCertificate, which takes NO context and builds |
| 76 | // its own background one with a 5-minute timeout. So the gate is |
| 77 | // re-checked BETWEEN orders instead, and a ≤5-minute overhang past a lost |
| 78 | // holding is accepted — inside the double-holder envelope issuerHostPolicy |
| 79 | // already documents. If the holding ends mid-order, the finishing |
| 80 | // cachePut re-checks the gate and the cert lands local-only; the new |
| 81 | // issuer re-orders it. Bounded waste, not a hazard. |
| 82 | // - No cache-load-before-gate step. The wildcard manager loads cached certs |
| 83 | // before standing down because it SERVES from its own map. The warm |
| 84 | // serves nothing — per-host serving is the mirror plus autocert — so a |
| 85 | // non-issuer scanning caches would be pure waste. It stands down at the |
| 86 | // gate. |
| 87 | // - The unconditional GetCertificate described in the package comment. |
| 88 | type certWarm struct { |
| 89 | // hosts is the warm set: exact hostnames this deployment serves that are |
| 90 | // not under the apps domain and not covered by a wildcard. See |
| 91 | // Edge.warmSet. |
| 92 | hosts func() []string |
| 93 | // getCert is autocert's per-host path (NOT the wildcard-wrapped one — a |
| 94 | // wrapped call would be answered by the wildcard and order nothing). |
| 95 | getCert func(*tls.ClientHelloInfo) (*tls.Certificate, error) |
| 96 | // pointsHere reports whether host's DNS resolves to this deployment. |
| 97 | // Ordering for a half-pointed name burns the shared account's |
| 98 | // failed-validation budget; see dnsGate. |
| 99 | pointsHere func(host string) bool |
| 100 | // issuer gates ORDERING, exactly as it gates the wildcard manager's. |
| 101 | issuer IssuerGate |
| 102 | // store is the deployment bucket, for the durable outcome records |
| 103 | // (#299 §3e, internal/certwarm). Nil disables recording entirely and |
| 104 | // changes nothing else: warming, backoff and logging are unaffected, so |
| 105 | // an agent-less `carlos edge` and every test that does not care keep |
| 106 | // working exactly as before. |
| 107 | store release.Store |
| 108 | |
| 109 | mu sync.Mutex |
| 110 | // nextAttempt is the per-host backoff floor. |
| 111 | nextAttempt map[string]time.Time |
| 112 | // warmed is the set of hosts a cert has been obtained for at least once, |
| 113 | // so a first order can be told from a renewal re-arm for the order cap |
| 114 | // and for logging. |
| 115 | warmed map[string]bool |
| 116 | |
| 117 | lastStandDownLog time.Time |
| 118 | // lastSweep and lastExpirySweep pace the cert sweep (certsweep.go). |
| 119 | // Deliberately NOT guarded by mu, unlike everything else on this struct: |
| 120 | // they are touched only from sweep(), which is only ever reached from |
| 121 | // pass(), which only ever runs on run()'s single goroutine. A mutex here |
| 122 | // would suggest a concurrency that does not exist. |
| 123 | lastSweep time.Time |
| 124 | lastExpirySweep time.Time |
| 125 | } |
| 126 | |
| 127 | func newCertWarm(hosts func() []string, getCert func(*tls.ClientHelloInfo) (*tls.Certificate, error), pointsHere func(string) bool, issuer IssuerGate, store release.Store) *certWarm { |
| 128 | return &certWarm{ |
| 129 | hosts: hosts, |
| 130 | getCert: getCert, |
| 131 | pointsHere: pointsHere, |
| 132 | issuer: issuer, |
| 133 | store: store, |
| 134 | nextAttempt: map[string]time.Time{}, |
| 135 | warmed: map[string]bool{}, |
| 136 | } |
| 137 | } |
| 138 | |
| 139 | // issuerGate is this warm's read of the deployment's gate, ungated default |
| 140 | // applied (see IssuerGate, certsync.go). |
| 141 | func (c *certWarm) issuerGate() (context.Context, bool) { |
| 142 | if c.issuer == nil { |
| 143 | return context.Background(), true |
| 144 | } |
| 145 | return c.issuer() |
| 146 | } |
| 147 | |
| 148 | func (c *certWarm) run() { |
| 149 | for { |
| 150 | time.Sleep(c.pass()) |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | // pass warms every host in the set once and returns how long to wait. |
| 155 | func (c *certWarm) pass() time.Duration { |
| 156 | hosts := c.hosts() |
| 157 | // Stand down before ANY work: unlike the wildcard manager there is no |
| 158 | // cache to load first, so a non-issuer has nothing useful to do here. |
| 159 | // The gate comes BEFORE the empty-set check, not after, because the |
| 160 | // sweep at the bottom of this pass has work to do in a deployment with |
| 161 | // no custom domains at all — spent challenge tokens and long-expired |
| 162 | // certs are not the warm set's business. |
| 163 | ctx, ok := c.issuerGate() |
| 164 | if !ok { |
| 165 | c.logStandDown(len(hosts)) |
| 166 | return warmInterval |
| 167 | } |
| 168 | if len(hosts) == 0 { |
| 169 | c.sweep(ctx, time.Now()) |
| 170 | return warmInterval |
| 171 | } |
| 172 | // The durable outcome records (#299 §3e), read ONCE per pass. Two things |
| 173 | // come out of this read and neither is available in memory: the attempt |
| 174 | // count a previous issuer accumulated (so a failover does not restart the |
| 175 | // clock on a domain that has been broken for days), and the set of hosts |
| 176 | // that currently HAVE a record, which is what makes both the |
| 177 | // delete-on-success and the prune below decidable without a Get per host. |
| 178 | records := c.loadRecords(ctx) |
| 179 | inSet := make(map[string]bool, len(hosts)) |
| 180 | for _, h := range hosts { |
| 181 | inSet[h] = true |
| 182 | } |
| 183 | orders, capped := 0, false |
| 184 | for _, h := range hosts { |
| 185 | now := time.Now() |
| 186 | c.mu.Lock() |
| 187 | until, backing := c.nextAttempt[h] |
| 188 | seen := c.warmed[h] |
| 189 | c.mu.Unlock() |
| 190 | if backing && now.Before(until) { |
| 191 | continue |
| 192 | } |
| 193 | // The gate is re-checked BETWEEN orders, not once per pass: a pass |
| 194 | // over a large set can outlive a holding, and every order after the |
| 195 | // holding ends is an order this box has no authority to place. |
| 196 | gctx, ok := c.issuerGate() |
| 197 | if !ok { |
| 198 | c.logStandDown(len(hosts) - orders) |
| 199 | return warmInterval |
| 200 | } |
| 201 | // A host with no cert yet is a NEW order and is both capped and |
| 202 | // DNS-gated. A host already warmed is a renewal re-arm: uncapped and |
| 203 | // ungated, because it is a cache read that costs the CA nothing and |
| 204 | // re-arms the timer that renewal depends on. Gating the re-arm on DNS |
| 205 | // would strand exactly the domain whose resolver is briefly sick. |
| 206 | if !seen { |
| 207 | if orders >= warmOrdersPerPass { |
| 208 | capped = true |
| 209 | continue |
| 210 | } |
| 211 | if !c.pointsHere(h) { |
| 212 | continue |
| 213 | } |
| 214 | orders++ |
| 215 | } |
| 216 | if err := c.warmOne(h); err != nil { |
| 217 | c.fail(gctx, h, err, records[h]) |
| 218 | continue |
| 219 | } |
| 220 | c.mu.Lock() |
| 221 | delete(c.nextAttempt, h) |
| 222 | first := !c.warmed[h] |
| 223 | c.warmed[h] = true |
| 224 | c.mu.Unlock() |
| 225 | if first { |
| 226 | log.Printf("edge: cert warmed for %s", h) |
| 227 | } |
| 228 | // A record exists only while a host is FAILING, so success removes |
| 229 | // it. That is what makes "no record" the healthy case and keeps the |
| 230 | // console's fold one object per broken host rather than one per |
| 231 | // domain (internal/certwarm's package comment). |
| 232 | if _, had := records[h]; had { |
| 233 | c.clearRecord(gctx, h, "recovered") |
| 234 | delete(records, h) |
| 235 | } |
| 236 | } |
| 237 | c.prune(ctx, records, inSet) |
| 238 | // Housekeeping last: it is the least urgent thing this pass does, and |
| 239 | // running it after the orders means a slow bucket delays litter removal |
| 240 | // rather than certificate issuance (certsweep.go). |
| 241 | c.sweep(ctx, time.Now()) |
| 242 | if capped { |
| 243 | return warmCappedDelay |
| 244 | } |
| 245 | return warmInterval |
| 246 | } |
| 247 | |
| 248 | // warmOne asks autocert for host's certificate, which mints it when absent |
| 249 | // and re-arms its renewal timer when present. |
| 250 | func (c *certWarm) warmOne(host string) error { |
| 251 | _, err := c.getCert(warmHello(host)) |
| 252 | return err |
| 253 | } |
| 254 | |
| 255 | // fail arms the per-host backoff, harder when the CA is rate-limiting, and |
| 256 | // records the outcome durably. |
| 257 | // |
| 258 | // prior is this host's record as the pass read it (the zero Record when |
| 259 | // there is none). The attempt count and the "failing since" stamp are |
| 260 | // carried FORWARD from it rather than recomputed from memory — that is the |
| 261 | // whole reason the record is in the bucket rather than on the heap. |
| 262 | func (c *certWarm) fail(ctx context.Context, host string, err error, prior certwarm.Record) { |
| 263 | backoff := warmBackoff |
| 264 | class := certwarm.ClassOther |
| 265 | var ae *acme.Error |
| 266 | if errors.As(err, &ae) && ae.ProblemType == "urn:ietf:params:acme:error:rateLimited" { |
| 267 | backoff = warmRateLimitBackoff |
| 268 | class = certwarm.ClassRateLimited |
| 269 | log.Printf("edge: cert warm for %s rate-limited by the CA — backing off %s (the whole deployment shares one ACME account)", host, backoff) |
| 270 | } else { |
| 271 | log.Printf("edge: cert warm for %s: %v — retrying in %s", host, err, backoff) |
| 272 | } |
| 273 | now := time.Now() |
| 274 | next := now.Add(backoff) |
| 275 | c.mu.Lock() |
| 276 | c.nextAttempt[host] = next |
| 277 | c.mu.Unlock() |
| 278 | |
| 279 | first := prior.FirstFailedAt |
| 280 | if first == 0 { |
| 281 | first = now.Unix() |
| 282 | } |
| 283 | c.writeRecord(ctx, certwarm.Record{ |
| 284 | Host: host, |
| 285 | Attempts: prior.Attempts + 1, |
| 286 | FirstFailedAt: first, |
| 287 | LastFailedAt: now.Unix(), |
| 288 | NextAttemptAt: next.Unix(), |
| 289 | Class: class, |
| 290 | Error: err.Error(), |
| 291 | }, now) |
| 292 | } |
| 293 | |
| 294 | func (c *certWarm) logStandDown(hosts int) { |
| 295 | c.mu.Lock() |
| 296 | defer c.mu.Unlock() |
| 297 | if time.Since(c.lastStandDownLog) < time.Hour { |
| 298 | return |
| 299 | } |
| 300 | c.lastStandDownLog = time.Now() |
| 301 | log.Printf("edge: cert warm standing down for %d host(s) — another box holds the issuer lease", hosts) |
| 302 | } |
| 303 | |
| 304 | // warmHello builds the synthetic ClientHelloInfo the warm hands autocert. |
| 305 | // |
| 306 | // The CipherSuites entry is LOAD-BEARING and must not be dropped as |
| 307 | // decoration. autocert picks the key type from the hello via supportsECDSA |
| 308 | // (ported here as helloSupportsECDSA): with a hello carrying no signature |
| 309 | // schemes, no curves and no cipher suites, all three checks fall through to |
| 310 | // false, autocert chooses certKey{isRSA: true}, and the warm mints an RSA |
| 311 | // cert under "<host>+rsa" — while every real browser asks for "<host>" and |
| 312 | // gets nothing. The order would be spent, the name still uncovered, and the |
| 313 | // first visitor would trigger a SECOND issuance. |
| 314 | // |
| 315 | // One ECDSA suite is enough: helloSupportsECDSA's first two checks skip on |
| 316 | // nil, and its cipher-suite loop returns true on this suite. |
| 317 | func warmHello(host string) *tls.ClientHelloInfo { |
| 318 | return &tls.ClientHelloInfo{ |
| 319 | ServerName: host, |
| 320 | CipherSuites: []uint16{tls.TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256}, |
| 321 | } |
| 322 | } |
| 323 | |
| 324 | // warmSet is the set of hostnames the cert warm orders and renews for: |
| 325 | // EXACT names this deployment serves that are not under the apps domain and |
| 326 | // not covered by a wildcard. |
| 327 | // |
| 328 | // The predicate is deliberately simpler than "aliases" or "custom domains". |
| 329 | // A custom domain is a foreign domain by definition — the attach door refuses |
| 330 | // apps-domain hosts outright — so "not under the apps domain" covers attached |
| 331 | // aliases and first-route claims alike, with no need to reason about which |
| 332 | // registry table a name arrived through. That matters: first-route claims |
| 333 | // (`carlos deploy -kind static -host <site>`) mint a real ROUTE, not an |
| 334 | // alias, so an alias-only rule would silently never warm the shipped headline |
| 335 | // flow. |
| 336 | // |
| 337 | // What it excludes, and why each one: |
| 338 | // |
| 339 | // - "*." rows — covered by a wildcard PEM, and autocert refuses the |
| 340 | // wildcard label anyway. |
| 341 | // - anything under the apps domain — canary hosts and platform hosts. This |
| 342 | // is the exclusion that keeps a first run from being a burst: canaries are |
| 343 | // three labels under the apps domain, no wildcard parent covers them, and |
| 344 | // a naive "everything we serve" set would order a cert AND a CT-log entry |
| 345 | // for every one of them. |
| 346 | // - anything one label under a CONFIGURED wildcard parent, tested BY SHAPE |
| 347 | // against currentParents() rather than by asking certFor. certFor returns |
| 348 | // nil for a parent whose cert has not been obtained yet, so a shape-blind |
| 349 | // test would warm every app of a new account per-host during exactly the |
| 350 | // window before their wildcard lands. |
| 351 | func (e *Edge) warmSet(appsDomain string, parents func() []string) []string { |
| 352 | appsDomain = strings.ToLower(strings.TrimSpace(appsDomain)) |
| 353 | covered := wildcardCoverer(parents) |
| 354 | seen := map[string]bool{} |
| 355 | var out []string |
| 356 | consider := func(h string) { |
| 357 | h = strings.ToLower(strings.TrimSpace(h)) |
| 358 | if h == "" || seen[h] || strings.HasPrefix(h, "*.") { |
| 359 | return |
| 360 | } |
| 361 | if appsDomain != "" && (h == appsDomain || strings.HasSuffix(h, "."+appsDomain)) { |
| 362 | return |
| 363 | } |
| 364 | if covered(h) { |
| 365 | return |
| 366 | } |
| 367 | seen[h] = true |
| 368 | out = append(out, h) |
| 369 | } |
| 370 | for h := range e.current() { |
| 371 | consider(h) |
| 372 | } |
| 373 | for _, h := range e.homes.Load().ServedHosts() { |
| 374 | consider(h) |
| 375 | } |
| 376 | sort.Strings(out) |
| 377 | return out |
| 378 | } |
| 379 | |
| 380 | // wildcardCoverer reports whether a host is exactly one label under a |
| 381 | // wildcard parent this box has configured or derived — by SHAPE, never by |
| 382 | // whether the parent's cert exists yet (see warmSet). A nil parents func |
| 383 | // covers nothing, which is the correct answer for a deployment with no |
| 384 | // wildcard manager at all. |
| 385 | func wildcardCoverer(parents func() []string) func(string) bool { |
| 386 | return func(host string) bool { |
| 387 | if parents == nil { |
| 388 | return false |
| 389 | } |
| 390 | for _, p := range parents() { |
| 391 | rest, ok := strings.CutSuffix(host, "."+p) |
| 392 | if ok && rest != "" && !strings.Contains(rest, ".") { |
| 393 | return true |
| 394 | } |
| 395 | } |
| 396 | return false |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | // dnsGate builds the "does this name point at us?" test the warm applies |
| 401 | // before placing a NEW order. |
| 402 | // |
| 403 | // Let's Encrypt caps failed validations at 5 per account per hostname per |
| 404 | // hour, and the ACME account is shared by every name this issuer mints. So a |
| 405 | // half-configured customer domain that we order for anyway does not just fail |
| 406 | // its own order — it spends budget every other tenant needs. The gate is what |
| 407 | // keeps one customer's unfinished DNS from being everyone's problem. |
| 408 | // |
| 409 | // ALL resolved answers must be in the edge set, not merely one. A name in |
| 410 | // mid-migration resolves to both the old host and ours; an any-match test |
| 411 | // passes it, the order goes out, and validation fails from the CA's vantage |
| 412 | // anyway — burning exactly the budget this gate exists to protect. Requiring |
| 413 | // all of them means the gate only opens once the customer has actually |
| 414 | // finished. |
| 415 | // |
| 416 | // An empty edge set fails CLOSED (no host points anywhere), matching the |
| 417 | // domains pass's stand-down: a set we could not read is not evidence about |
| 418 | // anybody's DNS, and the cost of waiting a pass is one hour of delay against |
| 419 | // the cost of burning a shared rate limit. |
| 420 | func dnsGate(resolve func(string) ([]string, error), edgeIPs func() []string) func(string) bool { |
| 421 | return func(host string) bool { |
| 422 | set := map[string]bool{} |
| 423 | for _, ip := range edgeIPs() { |
| 424 | set[ip] = true |
| 425 | } |
| 426 | if len(set) == 0 { |
| 427 | return false |
| 428 | } |
| 429 | ips, err := resolve(host) |
| 430 | if err != nil || len(ips) == 0 { |
| 431 | return false |
| 432 | } |
| 433 | for _, ip := range ips { |
| 434 | if !set[ip] { |
| 435 | return false |
| 436 | } |
| 437 | } |
| 438 | return true |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | // warmResolver is the DNS gate's lookup: the system resolver, bounded so a |
| 443 | // hung one costs this host its turn in the pass and not the pass itself. |
| 444 | func warmResolver(host string) ([]string, error) { |
| 445 | ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second) |
| 446 | defer cancel() |
| 447 | return net.DefaultResolver.LookupHost(ctx, host) |
| 448 | } |
| 449 | |
| 450 | // ---- the durable outcome records (#299 §3e) --------------------------------- |
| 451 | // |
| 452 | // Every one of these degrades to a no-op rather than to a failure. A bucket |
| 453 | // blip must never stop the warm from warming: the records are how the |
| 454 | // console explains a failure, not how the fix happens, and a pass that |
| 455 | // cannot write one has still ordered the certificate. |
| 456 | |
| 457 | // warmStoreTimeout budgets the record I/O for one pass. Generous relative to |
| 458 | // the work (a listing plus a handful of small objects) and tiny relative to |
| 459 | // the hourly cadence, so a slow bucket delays a pass rather than wedging it. |
| 460 | const warmStoreTimeout = 30 * time.Second |
| 461 | |
| 462 | // loadRecords reads every outcome record in the deployment. |
| 463 | // |
| 464 | // The cost is one object per BROKEN host, not one per domain — see the |
| 465 | // package comment on internal/certwarm — so this stays a listing plus a |
| 466 | // small handful of Gets in any deployment that is mostly working, and grows |
| 467 | // only in one that is mostly not. |
| 468 | // |
| 469 | // A read failure returns what it has. Partial knowledge is the right answer |
| 470 | // here in both directions: a host whose record failed to load is treated as |
| 471 | // having none, so its attempt count restarts at one (an undercount, which |
| 472 | // only ever delays the console reaching "failed") and its record is left |
| 473 | // alone by the prune (never a delete on the strength of a read we could not |
| 474 | // complete). |
| 475 | func (c *certWarm) loadRecords(ctx context.Context) map[string]certwarm.Record { |
| 476 | out := map[string]certwarm.Record{} |
| 477 | if c.store == nil { |
| 478 | return out |
| 479 | } |
| 480 | ctx, cancel := context.WithTimeout(ctx, warmStoreTimeout) |
| 481 | defer cancel() |
| 482 | keys, err := c.store.List(ctx, certwarm.Prefix) |
| 483 | if err != nil { |
| 484 | log.Printf("edge: cert warm: could not list %s: %v — attempt counts fall back to this process's own", certwarm.Prefix, err) |
| 485 | return out |
| 486 | } |
| 487 | for _, key := range keys { |
| 488 | host := certwarm.HostFromKey(key) |
| 489 | if host == "" { |
| 490 | continue |
| 491 | } |
| 492 | body, _, err := c.store.Get(ctx, key) |
| 493 | if err != nil { |
| 494 | log.Printf("edge: cert warm: could not read %s: %v", key, err) |
| 495 | continue |
| 496 | } |
| 497 | rec, err := certwarm.Parse(body) |
| 498 | if err != nil { |
| 499 | // A record this build cannot read is left in place, not deleted: |
| 500 | // it may be a NEWER issuer's, and a rolling fleet has both. |
| 501 | log.Printf("edge: cert warm: could not parse %s: %v", key, err) |
| 502 | continue |
| 503 | } |
| 504 | out[host] = rec |
| 505 | } |
| 506 | return out |
| 507 | } |
| 508 | |
| 509 | // writeRecord publishes one host's outcome, CAS so a transient double-holder |
| 510 | // cannot half-overwrite the other's record. |
| 511 | func (c *certWarm) writeRecord(ctx context.Context, rec certwarm.Record, now time.Time) { |
| 512 | if c.store == nil { |
| 513 | return |
| 514 | } |
| 515 | body, err := rec.Encode(now) |
| 516 | if err != nil { |
| 517 | log.Printf("edge: cert warm: could not encode the record for %s: %v", rec.Host, err) |
| 518 | return |
| 519 | } |
| 520 | ctx, cancel := context.WithTimeout(ctx, warmStoreTimeout) |
| 521 | defer cancel() |
| 522 | if err := putCAS(ctx, c.store, certwarm.Key(rec.Host), body); err != nil { |
| 523 | log.Printf("edge: cert warm: could not record the failure for %s: %v", rec.Host, err) |
| 524 | } |
| 525 | } |
| 526 | |
| 527 | // clearRecord removes one host's record. why names the occasion in the log — |
| 528 | // "recovered" or "detached" — because those are opposite news and a single |
| 529 | // deletion line would hide which one happened. |
| 530 | func (c *certWarm) clearRecord(ctx context.Context, host, why string) { |
| 531 | if c.store == nil { |
| 532 | return |
| 533 | } |
| 534 | ctx, cancel := context.WithTimeout(ctx, warmStoreTimeout) |
| 535 | defer cancel() |
| 536 | if err := c.store.Delete(ctx, certwarm.Key(host)); err != nil { |
| 537 | log.Printf("edge: cert warm: could not clear the %s record for %s: %v", why, host, err) |
| 538 | return |
| 539 | } |
| 540 | log.Printf("edge: cert warm: cleared the outcome record for %s (%s)", host, why) |
| 541 | } |
| 542 | |
| 543 | // prune drops records for hosts that are no longer in the warm set — a |
| 544 | // detached custom domain, or one whose route went away. |
| 545 | // |
| 546 | // This deliberately DOES delete on the strength of an inference ("not in the |
| 547 | // set"), which §3f forbids for certificates. The distinction is ownership, |
| 548 | // not caution: a certificate is the member's property and a wrong deletion |
| 549 | // takes their site down, whereas a warm record is the platform's own note |
| 550 | // about its own attempts and a wrong deletion costs an attempt count that |
| 551 | // the next failure rebuilds. The prune is also fenced by pass()'s early |
| 552 | // return on an empty set, so the one case that would delete EVERYTHING — |
| 553 | // a warm set that briefly folds to nothing — never reaches here. |
| 554 | func (c *certWarm) prune(ctx context.Context, records map[string]certwarm.Record, inSet map[string]bool) { |
| 555 | if c.store == nil || len(records) == 0 { |
| 556 | return |
| 557 | } |
| 558 | for host := range records { |
| 559 | if inSet[host] { |
| 560 | continue |
| 561 | } |
| 562 | c.clearRecord(ctx, host, "detached") |
| 563 | } |
| 564 | } |
| 565 | |
amadan