| 1 | import { Container, type StopParams } from "@cloudflare/containers"; |
| 2 | import { WorkerEntrypoint } from "cloudflare:workers"; |
| 3 | |
| 4 | import { |
| 5 | type BumpArgs, |
| 6 | UPDATE_BRANCH_PREFIX, |
| 7 | type RunKind, |
| 8 | agentsClient, |
| 9 | type G1tEvent, |
| 10 | type QueueJob, |
| 11 | type RepoPath, |
| 12 | type Result, |
| 13 | type ServiceBinding, |
| 14 | type User, |
| 15 | type Viewer, |
| 16 | type ComputeEntitlements, |
| 17 | type ComputeKind, |
| 18 | ComputeGate, |
| 19 | actualMicros, |
| 20 | eventsClient, |
| 21 | refusalMessage, |
| 22 | sandboxEstimateMicros, |
| 23 | billingClient, |
| 24 | fail, |
| 25 | ok, |
| 26 | reposClient, |
| 27 | workClient, |
| 28 | type InstanceType, |
| 29 | STANDARD_INSTANCE, |
| 30 | instanceNamed, |
| 31 | priceScale, |
| 32 | runnerMinutes, |
| 33 | } from "@g1t/contracts"; |
| 34 | |
| 35 | import { BUMP_MINUTES, BUMP_TOKEN_TTL_SECONDS, bumpEnv, bumpProblem, bumpSandboxName, systemActor } from "./bump"; |
| 36 | import { BACKUP_MINUTES, backupEnv, backupPace, backupSandboxName } from "./backup"; |
| 37 | import { holdCredentials, revokeCredentials, runCredential } from "./credentials"; |
| 38 | import { cancelTask, enqueueTask, handedOverStep, selfHostedRoute, taskEnv, taskRepo } from "./self-hosted"; |
| 39 | import { answered, capReached, describeError, type Owed, recordOwed, stoppedShort, tellStopped, withinTimeCap } from "./lifecycle"; |
| 40 | import { |
| 41 | ABUSE_EXIT_CODE, |
| 42 | ABUSE_HOST, |
| 43 | ABUSE_MESSAGE, |
| 44 | ALARM_GRACE_SECONDS, |
| 45 | type PlanLimits, |
| 46 | type RunGuard, |
| 47 | abuse, |
| 48 | buildGuard, |
| 49 | dockerFor, |
| 50 | egress, |
| 51 | guardFor, |
| 52 | harnessEnv, |
| 53 | reportRun, |
| 54 | SANDBOX_BINDINGS, |
| 55 | sandboxNamespace, |
| 56 | throughWorker, |
| 57 | timeCapMessage, |
| 58 | withPlanLimits, |
| 59 | } from "./guard"; |
| 60 | import { newlyMisdirected } from "./broker"; |
| 61 | import { relay, relayDeps } from "./broker-relay"; |
| 62 | import { forward } from "./forward"; |
| 63 | import { EGRESS_ENV, jobNamespaceOf } from "./egress"; |
| 64 | |
| 65 | // Outbound interception, which brokered secrets and abuse reports use, |
| 66 | // needs this exported. |
| 67 | export { ContainerProxy } from "@cloudflare/containers"; |
| 68 | |
| 69 | export interface RunnerEnv { |
| 70 | SANDBOX: DurableObjectNamespace<AttemptSandbox>; |
| 71 | /** |
| 72 | * Workflow jobs that ask with `runs-on` for `g1t-2core` (the standard |
| 73 | * machine, kept as a name so the label keeps working) or `g1t-4core` (a |
| 74 | * larger instance type): the same image, in classes of their own. |
| 75 | */ |
| 76 | SANDBOX_2CORE?: DurableObjectNamespace<Sandbox2Core>; |
| 77 | SANDBOX_4CORE?: DurableObjectNamespace<Sandbox4Core>; |
| 78 | IDENTITY: ServiceBinding; |
| 79 | REPOS: ServiceBinding; |
| 80 | WORK: ServiceBinding; |
| 81 | BILLING: ServiceBinding; |
| 82 | /** GitHub Actions jobs: told when a job's sandbox dies without reporting. */ |
| 83 | ACTIONS: ServiceBinding; |
| 84 | /** Told when a deploy sandbox dies without reporting. */ |
| 85 | DEPLOYMENTS: ServiceBinding; |
| 86 | /** The event bus: `abuse.flagged`, for g1t's staff. */ |
| 87 | EVENTS?: ServiceBinding; |
| 88 | CLOUDFLARE_ACCOUNT_ID: string; |
| 89 | /** |
| 90 | * `off` stops passing any sandbox's HTTP and HTTPS through the Worker: a |
| 91 | * switch for the operator, should it misbehave. Brokered secrets' stand- |
| 92 | * ins then go out as they are, unfilled. Anything else passes the |
| 93 | * requests of workflow jobs that may be given brokered secrets. |
| 94 | */ |
| 95 | EGRESS?: string; |
| 96 | /** |
| 97 | * `off` stops sandboxes watching themselves for mining (crates/runner |
| 98 | * abuse.rs): a switch for the operator, should it stop real work. |
| 99 | * Anything else leaves it on. Miners named in commands are refused |
| 100 | * either way. |
| 101 | */ |
| 102 | ABUSE_WATCH?: string; |
| 103 | /** |
| 104 | * `off` leaves workflow jobs without a Docker Engine of their own |
| 105 | * (crates/runner docker/): a switch for the operator. Anything else |
| 106 | * gives each job one, started the first time it is used. |
| 107 | */ |
| 108 | DOCKER?: string; |
| 109 | /** |
| 110 | * Nightly backups (backup.ts): how many queued backups one sweep starts |
| 111 | * (`0`: none, backups off here), and how many may run at once. |
| 112 | */ |
| 113 | BACKUPS_PER_SWEEP?: string; |
| 114 | BACKUPS_RUNNING?: string; |
| 115 | } |
| 116 | |
| 117 | /** |
| 118 | * What a sandbox is doing. Runs of an issue's commands (`checks`) are from |
| 119 | * before checks were workflows; a sandbox that was started for work g1t no |
| 120 | * longer does is told apart by its kind and only cleaned up. |
| 121 | */ |
| 122 | type Run = |
| 123 | | { kind: "checks"; runId: string; token: string } |
| 124 | /** One combined state of a merge queue, being built and checked. */ |
| 125 | | { kind: "queue"; entryId: string; token: string } |
| 126 | /** Whether a pull request merges cleanly: two commits merged, nothing pushed. */ |
| 127 | | { kind: "mergecheck"; pullId: string; token: string } |
| 128 | /** One job of a GitHub Actions workflow. */ |
| 129 | | { kind: "actions"; jobId: string; token: string } |
| 130 | /** A build of one commit, deployed to g1t.page. */ |
| 131 | | { kind: "deploy"; deployId: string; token: string } |
| 132 | /** |
| 133 | * A security update: one package raised in its lockfiles and pushed to |
| 134 | * its branch. The security service opens the pull request when it hears |
| 135 | * the push, so a failure has no one to tell. |
| 136 | */ |
| 137 | | { kind: "bump"; repo: RepoPath; branch: string } |
| 138 | /** |
| 139 | * A repository's nightly backup: a bundle cut and sent to the repos |
| 140 | * service. g1t's own work, never charged to the workspace. |
| 141 | */ |
| 142 | | { kind: "backup"; jobId: string; token: string }; |
| 143 | /** |
| 144 | * Whose sandbox time it is, reported when the sandbox stops, and the |
| 145 | * machine it ran on when it was not the standard one. |
| 146 | */ |
| 147 | type Meter = { |
| 148 | workspace: string; |
| 149 | repo: string; |
| 150 | description: string; |
| 151 | instance?: string | null; |
| 152 | }; |
| 153 | /** A stopped sandbox's time, as billing records it (`recordSandbox`). */ |
| 154 | type SandboxUsage = Parameters<ReturnType<typeof billingClient>["recordSandbox"]>[0]; |
| 155 | /** |
| 156 | * What billing reserved for a sandbox's work (`ComputeGate.admit`), settled |
| 157 | * when it stops at what it cost: its seconds. |
| 158 | */ |
| 159 | type Held = { id: string; workspace: string; microsPerSecond: number }; |
| 160 | /** |
| 161 | * A sandbox that is not tracked as a run but still has a time cap: a |
| 162 | * workflow job, a deploy build or an update, for `minutes` at most. |
| 163 | */ |
| 164 | type Build = { |
| 165 | kind: "actions" | "deploy" | "bump"; |
| 166 | minutes: number; |
| 167 | /** |
| 168 | * A workflow job that may be given brokered secrets: its HTTP and HTTPS |
| 169 | * pass the Worker, which fills them in. |
| 170 | */ |
| 171 | brokered?: boolean; |
| 172 | }; |
| 173 | /** Deploy builds are metered by the Deployments plan, not here. */ |
| 174 | type RunRequest = Run & { |
| 175 | envVars: Record<string, string>; |
| 176 | meter?: Meter; |
| 177 | track?: Track; |
| 178 | /** The workspace's plan's caps, applied under its guardrails' (lower of each). */ |
| 179 | limits?: PlanLimits; |
| 180 | reservation?: Held | null; |
| 181 | build?: Build; |
| 182 | /** Whose sandbox it is, when it has no meter: for `abuse.flagged`. */ |
| 183 | owner?: { workspace: string; repo: string }; |
| 184 | /** |
| 185 | * The labels of the workspace's self-hosted runners this work goes to |
| 186 | * instead of a container (self-hosted.ts). Null or absent: a container. |
| 187 | */ |
| 188 | selfHosted?: string[] | null; |
| 189 | }; |
| 190 | |
| 191 | /** What a sandbox is, as billing meters it. */ |
| 192 | function computeKindOf(kind: Run["kind"]): ComputeKind | null { |
| 193 | switch (kind) { |
| 194 | case "checks": |
| 195 | case "mergecheck": |
| 196 | // A security update resolves lockfiles, as cheap as a check. |
| 197 | case "bump": |
| 198 | return "check"; |
| 199 | case "queue": |
| 200 | return "queue"; |
| 201 | case "actions": |
| 202 | return "workflow"; |
| 203 | case "deploy": |
| 204 | return "deploy"; |
| 205 | // Not metered: a backup is g1t's own cost; nor is a sandbox of a kind |
| 206 | // g1t no longer starts, which ran before it stopped starting them. |
| 207 | default: |
| 208 | return null; |
| 209 | } |
| 210 | } |
| 211 | |
| 212 | /** One gate per isolate, so entitlements and prices are kept between calls. */ |
| 213 | let gate: ComputeGate | null = null; |
| 214 | function gateFor(env: { BILLING: ServiceBinding }): ComputeGate { |
| 215 | gate ??= new ComputeGate(env.BILLING); |
| 216 | return gate; |
| 217 | } |
| 218 | |
| 219 | /** |
| 220 | * What to record the sandbox as, so people can watch it: a run of the |
| 221 | * merge queue. |
| 222 | */ |
| 223 | type Track = { |
| 224 | actor: User; |
| 225 | repo: RepoPath; |
| 226 | kind: RunKind; |
| 227 | number?: number | null; |
| 228 | pullId?: string | null; |
| 229 | title?: string | null; |
| 230 | startedBy?: string | null; |
| 231 | }; |
| 232 | /** The run a sandbox reports to, kept so it can be closed when it stops. */ |
| 233 | type TrackedRun = { runId: string; token: string }; |
| 234 | |
| 235 | function meter(repo: RepoPath, description: string): Meter { |
| 236 | return { workspace: repo.namespace, repo: `${repo.namespace}/${repo.name}`, description }; |
| 237 | } |
| 238 | |
| 239 | /** What the deployments service asks a sandbox to build. */ |
| 240 | type DeployJob = { |
| 241 | deployId: string; |
| 242 | /** The workspace the project is in, which pays. */ |
| 243 | workspace?: string; |
| 244 | /** What the deployments service reserved for the build, settled when it stops. */ |
| 245 | reservation?: string | null; |
| 246 | /** The price it reserved at, per second. */ |
| 247 | microsPerSecond?: number | null; |
| 248 | /** The plan's longest run, in minutes; the build gets the lower of this and its own. */ |
| 249 | maxRunMinutes?: number | null; |
| 250 | /** Lets the sandbox, and nothing else, report this build. */ |
| 251 | token: string; |
| 252 | /** Whose access reads the commit. */ |
| 253 | actor: User; |
| 254 | /** The repository the commit is in: the pull request's fork, or the repository. */ |
| 255 | source: RepoPath; |
| 256 | /** |
| 257 | * The project's repository, whose guardrails the build runs under, and |
| 258 | * its id. A preview's `source` is its pull request's working copy, so |
| 259 | * the two differ. Older callers send only `source`. |
| 260 | */ |
| 261 | repo?: RepoPath | null; |
| 262 | repoId?: string | null; |
| 263 | commit: string; |
| 264 | /** Where in the repository the project lives; empty for all of it. */ |
| 265 | rootDir?: string; |
| 266 | buildCommand?: string | null; |
| 267 | outputDir?: string | null; |
| 268 | /** What it is built as (`auto`, `static`, `workers` or `docker`), and its Dockerfile's path when set. */ |
| 269 | buildAs?: string | null; |
| 270 | dockerfile?: string | null; |
| 271 | /** The repository's variables for deploy builds. */ |
| 272 | buildEnv?: Record<string, string>; |
| 273 | /** Its secrets for deploy builds: set like variables, and redacted from the log. */ |
| 274 | buildSecrets?: Record<string, string>; |
| 275 | }; |
| 276 | |
| 277 | /** Long enough to install and build; then the read token stops working. */ |
| 278 | const DEPLOY_TOKEN_TTL_SECONDS = 30 * 60; |
| 279 | |
| 280 | /** Long enough to clone, build and test a merge queue's state; then the token stops working. */ |
| 281 | const CHECKS_TOKEN_TTL_SECONDS = 45 * 60; |
| 282 | |
| 283 | /** Long enough to clone and merge two commits; then the read token stops working. */ |
| 284 | const MERGECHECK_TOKEN_TTL_SECONDS = 10 * 60; |
| 285 | |
| 286 | /** |
| 287 | * One sandbox, for one piece of work: a workflow job, a build, a state of |
| 288 | * the merge queue, a merge check, an update or a backup. The image's entrypoint |
| 289 | * is the g1t runner, which does the work and exits; this class only starts |
| 290 | * it and cleans up if it dies without reporting. |
| 291 | */ |
| 292 | export class AttemptSandbox extends Container<RunnerEnv> { |
| 293 | // Past the longest default time cap (90 minutes) and its alarm. A run whose guardrails allow longer (up to 240 minutes) is kept |
| 294 | // past it by `onActivityExpired`, so only its own cap ends it. A finished |
| 295 | // run's process exits and stops the sandbox well before this. |
| 296 | sleepAfter = "100m"; |
| 297 | // HTTPS to an intercepted host goes through its handler too (guard.ts). |
| 298 | interceptHttps = true; |
| 299 | static { |
| 300 | // Assigned, not declared: a class field would hide the setter that |
| 301 | // registers the handler with the containers library. Which one a |
| 302 | // sandbox uses is chosen as it starts (`run`). |
| 303 | AttemptSandbox.outboundHandlers = { egress, abuse }; |
| 304 | } |
| 305 | |
| 306 | async run(request: RunRequest): Promise<void> { |
| 307 | const { envVars, meter, track, limits, reservation, build, owner, selfHosted, ...run } = request; |
| 308 | // What billing reserved is settled however this ends, once. |
| 309 | if (reservation) await this.ctx.storage.put("reservation", reservation); |
| 310 | let guard: RunGuard | null; |
| 311 | try { |
| 312 | // A tracked run gets its project's guardrails; no sandbox for one |
| 313 | // starts without them. Workflow jobs, deploy builds and updates get |
| 314 | // the time cap they were given. The plan's caps apply under them: |
| 315 | // the lower of each. |
| 316 | guard = track |
| 317 | ? withPlanLimits(await guardFor(this.env.WORK, track.repo, track.kind), limits) |
| 318 | : build |
| 319 | ? withPlanLimits(buildGuard(build.minutes), limits) |
| 320 | : null; |
| 321 | } catch (error) { |
| 322 | // Thrown to the caller, which says why the work did not start; logged |
| 323 | // here too, so a sandbox that never started is traceable on its own. |
| 324 | console.error("sandbox not started", run.kind, describeError(error)); |
| 325 | await this.settle(0); |
| 326 | throw error; |
| 327 | } |
| 328 | await this.ctx.storage.put("run", run); |
| 329 | // What an earlier attempt in this object left: its reason, its cap |
| 330 | // and its cap's alarm, which would otherwise stop this attempt when |
| 331 | // the earlier one's time ran out (re-runs use the same object). |
| 332 | await this.ctx.storage.delete(["abuse", "stopReason", "remote", "misdirected", "timeCap"]); |
| 333 | this.deleteSchedules("timeUp"); |
| 334 | if (meter) await this.ctx.storage.put("meter", { ...meter, started: Date.now() }); |
| 335 | await this.ctx.storage.put("started", Date.now()); |
| 336 | const who = meter ? { workspace: meter.workspace, repo: meter.repo } : owner; |
| 337 | if (who) await this.ctx.storage.put("owner", { ...who, kind: track?.kind ?? run.kind }); |
| 338 | const tracked = track ? await this.openRun(track, envVars, guard) : null; |
| 339 | // Its credentials are tied to the run, and revoked when it stops. |
| 340 | await holdCredentials(this.env.IDENTITY, this.ctx.storage, envVars, tracked?.runId ?? null); |
| 341 | try { |
| 342 | const vars = tracked ? { ...envVars, AGENT_RUN: tracked.runId, AGENT_RUN_TOKEN: tracked.token } : envVars; |
| 343 | // The workspace's own runner, not a container: the same environment, |
| 344 | // handed over as a task. |
| 345 | const repo = selfHosted?.length ? taskRepo(track, meter, owner) : null; |
| 346 | if (selfHosted?.length && repo) { |
| 347 | const harness = guard ? harnessEnv(guard, vars) : {}; |
| 348 | const minutes = guard?.minutes ?? limits?.minutes ?? 60; |
| 349 | await enqueueTask(this.env.ACTIONS, { |
| 350 | sandbox: this.ctx.id.toString(), |
| 351 | repo, |
| 352 | kind: track?.kind ?? run.kind, |
| 353 | title: track?.title ?? meter?.description ?? `${run.kind} in ${repo.namespace}/${repo.name}`, |
| 354 | labels: selfHosted, |
| 355 | env: taskEnv({ ...vars, ...harness }), |
| 356 | timeoutMinutes: minutes, |
| 357 | runId: tracked?.runId ?? null, |
| 358 | }); |
| 359 | await this.ctx.storage.put("remote", true); |
| 360 | if (tracked) await reportRun(this.env.WORK, tracked, { steps: [handedOverStep(selfHosted)] }); |
| 361 | if (guard) { |
| 362 | await this.ctx.storage.put("timeCap", guard.minutes); |
| 363 | await this.schedule(guard.minutes * 60 + ALARM_GRACE_SECONDS, "timeUp"); |
| 364 | } |
| 365 | return; |
| 366 | } |
| 367 | // Every sandbox reaches the internet. A workflow job that may be |
| 368 | // given brokered secrets sends its HTTP and HTTPS through `egress`, |
| 369 | // which fills them in from the job's stand-ins (asked for with the |
| 370 | // job's token) and sends every request on, to any host; any other |
| 371 | // sandbox has only `ABUSE_HOST` intercepted, so it can report that it |
| 372 | // stopped itself for mining. Both are chosen here, as it starts: hosts |
| 373 | // named on the class up front (`outboundByHost` in a static block) |
| 374 | // left sandboxes that never picked up their jobs (rolled back |
| 375 | // 2026-10-13). |
| 376 | const brokering = run.kind === "actions" && throughWorker(build, this.env.EGRESS); |
| 377 | this.enableInternet = true; |
| 378 | if (brokering && run.kind === "actions") { |
| 379 | await this.setOutboundHandler("egress", { job: { id: run.jobId, token: run.token } }); |
| 380 | } else if (this.env.EGRESS?.trim().toLowerCase() !== "off") { |
| 381 | await this.setOutboundByHost(ABUSE_HOST, "abuse").catch((error: unknown) => |
| 382 | console.log("abuse reports not routed", String(error)), |
| 383 | ); |
| 384 | } |
| 385 | const harness = guard && !build ? harnessEnv(guard, vars) : {}; |
| 386 | // Once any of its traffic is intercepted, Cloudflare passes all of it |
| 387 | // through this Worker, and HTTPS is re-signed on the way |
| 388 | // (`interceptHttps`); so a sandbox with either handler trusts that |
| 389 | // certificate, or every HTTPS request it makes fails |
| 390 | // (SELF_SIGNED_CERT_IN_CHAIN, as on 2026-10-13). |
| 391 | const intercepted = brokering || this.env.EGRESS?.trim().toLowerCase() !== "off"; |
| 392 | const trust = intercepted ? EGRESS_ENV : {}; |
| 393 | const watch: Record<string, string> = this.env.ABUSE_WATCH === "off" ? { G1T_ABUSE: "off" } : {}; |
| 394 | await this.start({ envVars: { ...vars, ...harness, ...trust, ...watch }, enableInternet: true }); |
| 395 | // A backup has no guardrails, but still a time cap. |
| 396 | const cap = guard?.minutes ?? (run.kind === "backup" ? BACKUP_MINUTES : null); |
| 397 | if (cap) { |
| 398 | await this.ctx.storage.put("timeCap", cap); |
| 399 | await this.schedule(cap * 60 + ALARM_GRACE_SECONDS, "timeUp"); |
| 400 | } |
| 401 | } catch (error) { |
| 402 | console.error("sandbox not started", run.kind, describeError(error)); |
| 403 | await revokeCredentials(this.env.IDENTITY, this.ctx.storage); |
| 404 | if (tracked) await this.closeRun("failed", `The sandbox could not start: ${String(error)}`); |
| 405 | await this.settle(0); |
| 406 | throw error; |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | /** Settles what billing reserved for this sandbox at `micros`, once. */ |
| 411 | private async settle(micros: number): Promise<void> { |
| 412 | const held = await this.ctx.storage.get<Held>("reservation"); |
| 413 | if (!held) return; |
| 414 | await this.ctx.storage.delete("reservation"); |
| 415 | await gateFor(this.env).settle(held.id, micros); |
| 416 | } |
| 417 | |
| 418 | /** Settles the reservation at what the sandbox cost: its seconds at the price billing reserved at. */ |
| 419 | private async settleStopped(started: number | undefined): Promise<void> { |
| 420 | const held = await this.ctx.storage.get<Held>("reservation"); |
| 421 | if (!held) return; |
| 422 | const seconds = started ? Math.max(1, Math.ceil((Date.now() - started) / 1000)) : 0; |
| 423 | await this.settle(actualMicros(seconds, held.microsPerSecond)); |
| 424 | } |
| 425 | |
| 426 | /** |
| 427 | * The sandbox stopped itself because it looked like it was mining |
| 428 | * (crates/runner abuse.rs), or exited saying so. Stops the run with |
| 429 | * `ABUSE_MESSAGE`, tells g1t's staff with `abuse.flagged`, and destroys |
| 430 | * the sandbox. Once. |
| 431 | */ |
| 432 | async flagAbuse(verdict: unknown): Promise<void> { |
| 433 | if (await this.ctx.storage.get<boolean>("abuse")) return; |
| 434 | await this.ctx.storage.put("abuse", true); |
| 435 | const tracked = await this.ctx.storage.get<TrackedRun>("agentRun"); |
| 436 | if (tracked) await reportRun(this.env.WORK, tracked, { halt: "abuse", error: ABUSE_MESSAGE }); |
| 437 | const owner = await this.ctx.storage.get<{ workspace: string; repo: string | null; kind: string }>("owner"); |
| 438 | console.log("abuse flagged", owner?.workspace, owner?.repo, owner?.kind, JSON.stringify(verdict)); |
| 439 | if (this.env.EVENTS && owner) { |
| 440 | await eventsClient(this.env.EVENTS) |
| 441 | .publish([ |
| 442 | { |
| 443 | type: "abuse.flagged", |
| 444 | source: "runner", |
| 445 | // Never on a repository's timeline or its webhooks. |
| 446 | repoId: null, |
| 447 | actor: null, |
| 448 | data: { |
| 449 | workspace: owner.workspace, |
| 450 | repo: owner.repo ?? null, |
| 451 | run: tracked?.runId ?? null, |
| 452 | kind: owner.kind, |
| 453 | sandbox: this.ctx.id.toString(), |
| 454 | metrics: verdict && typeof verdict === "object" ? (verdict as Record<string, unknown>) : null, |
| 455 | }, |
| 456 | }, |
| 457 | ]) |
| 458 | .catch((error: unknown) => console.log("abuse.flagged not published", String(error))); |
| 459 | } |
| 460 | await this.destroy().catch((error: unknown) => console.log("sandbox not destroyed for abuse", String(error))); |
| 461 | } |
| 462 | |
| 463 | /** |
| 464 | * Records the run, which the sandbox then reports its steps to. Never |
| 465 | * stops the sandbox from starting: without a record it just goes unseen. |
| 466 | */ |
| 467 | private async openRun(track: Track, envVars: Record<string, string>, guard: RunGuard | null): Promise<TrackedRun | null> { |
| 468 | const opened = await agentsClient(this.env.WORK) |
| 469 | .openRun({ |
| 470 | ...track, |
| 471 | model: null, |
| 472 | sandbox: this.ctx.id.toString(), |
| 473 | timeCapMinutes: guard?.minutes ?? null, |
| 474 | }) |
| 475 | .catch((error: unknown) => ({ ok: false as const, error: { message: String(error) } })); |
| 476 | if (!opened.ok) { |
| 477 | console.log("run not recorded", track.kind, opened.error.message); |
| 478 | return null; |
| 479 | } |
| 480 | await this.ctx.storage.put("agentRun", opened.value); |
| 481 | return opened.value; |
| 482 | } |
| 483 | |
| 484 | /** |
| 485 | * A workflow job sent a brokered secret's stand-in to a host it is not |
| 486 | * for: said once per secret and host, in the job's log. |
| 487 | */ |
| 488 | async noteMisdirected(name: string, host: string): Promise<void> { |
| 489 | const run = await this.ctx.storage.get<Run>("run"); |
| 490 | if (run?.kind !== "actions") return; |
| 491 | const noted = newlyMisdirected((await this.ctx.storage.get<string[]>("misdirected")) ?? [], name, host); |
| 492 | if (!noted) return; |
| 493 | await this.ctx.storage.put("misdirected", noted.seen); |
| 494 | try { |
| 495 | const response = await this.env.ACTIONS.fetch("https://actions/rpc/job_report", { |
| 496 | method: "POST", |
| 497 | headers: { "content-type": "application/json" }, |
| 498 | body: JSON.stringify({ job: run.jobId, token: run.token, report: { kind: "note", text: noted.note } }), |
| 499 | }); |
| 500 | await answered("job_report", response); |
| 501 | } catch (error) { |
| 502 | console.log("misdirected secret not noted", name, String(error)); |
| 503 | } |
| 504 | } |
| 505 | |
| 506 | /** The run's time cap has passed: stop it, as stopped for time. */ |
| 507 | async timeUp(): Promise<void> { |
| 508 | // It already stopped (nothing to stop), or the run in it now is a later |
| 509 | // attempt whose cap has not passed: this alarm is an earlier attempt's, |
| 510 | // left by a version of g1t that did not clear it, and the run now has |
| 511 | // its own. |
| 512 | const started = await this.ctx.storage.get<number>("started"); |
| 513 | const minutes = (await this.ctx.storage.get<number>("timeCap")) ?? 0; |
| 514 | if (!capReached(started, minutes, Date.now())) { |
| 515 | if (started) console.log("an earlier attempt's time cap passed; the run now keeps going", this.ctx.id.toString()); |
| 516 | return; |
| 517 | } |
| 518 | const tracked = await this.ctx.storage.get<TrackedRun>("agentRun"); |
| 519 | // A workflow job or a build has no run to halt: it fails saying why. |
| 520 | await this.ctx.storage.put("stopReason", timeCapMessage(minutes)); |
| 521 | if (tracked) await reportRun(this.env.WORK, tracked, { halt: "time", error: timeCapMessage(minutes) }); |
| 522 | if (await this.ctx.storage.get<boolean>("remote")) { |
| 523 | await cancelTask(this.env.ACTIONS, this.ctx.id.toString(), timeCapMessage(minutes)); |
| 524 | await this.remoteEnded(1, null); |
| 525 | return; |
| 526 | } |
| 527 | await this.destroy().catch((error: unknown) => console.log("sandbox not destroyed at its time cap", String(error))); |
| 528 | } |
| 529 | |
| 530 | /** |
| 531 | * Stops this sandbox's work: its container, or the task a self-hosted |
| 532 | * runner holds, which it hears about on its next poll. |
| 533 | */ |
| 534 | async halt(reason: string | null): Promise<void> { |
| 535 | if (await this.ctx.storage.get<boolean>("remote")) { |
| 536 | await cancelTask(this.env.ACTIONS, this.ctx.id.toString(), reason); |
| 537 | await this.remoteEnded(1, reason); |
| 538 | return; |
| 539 | } |
| 540 | // A container already gone has nothing to stop; one that will not stop |
| 541 | // is logged, and its time cap still ends it. |
| 542 | await this.destroy().catch((error: unknown) => console.error("sandbox not destroyed", reason, describeError(error))); |
| 543 | } |
| 544 | |
| 545 | /** |
| 546 | * The library's `sleepAfter` has passed. The runner never fetches its |
| 547 | * container, so to the library every sandbox looks idle: a run inside its |
| 548 | * time cap keeps going, and the cap's own alarm (`timeUp`) ends it. Only |
| 549 | * a sandbox with no cap is stopped for inactivity. |
| 550 | */ |
| 551 | override async onActivityExpired(): Promise<void> { |
| 552 | const started = await this.ctx.storage.get<number>("started"); |
| 553 | const cap = await this.ctx.storage.get<number>("timeCap"); |
| 554 | if (withinTimeCap(started, cap, Date.now(), ALARM_GRACE_SECONDS)) return; |
| 555 | await super.onActivityExpired(); |
| 556 | } |
| 557 | |
| 558 | /** |
| 559 | * A container that crashed or could not be reached, as the library tells |
| 560 | * it. Logged at error level with the sandbox, never thrown: the library |
| 561 | * ignores what this throws, and the stop that follows is handled by |
| 562 | * `onStop` or by `run`, which says why the work did not start. |
| 563 | */ |
| 564 | override onError(error: unknown): void { |
| 565 | console.error("sandbox container error", this.ctx.id.toString(), describeError(error)); |
| 566 | } |
| 567 | |
| 568 | /** |
| 569 | * The library's alarm: scheduled callbacks, the container's keep-alive, |
| 570 | * and `onStop` once it has stopped. A failure is logged with the retry it |
| 571 | * was, then thrown so Cloudflare tries the alarm again. |
| 572 | */ |
| 573 | override async alarm(alarmProps?: AlarmInvocationInfo): Promise<void> { |
| 574 | try { |
| 575 | await super.alarm(alarmProps); |
| 576 | } catch (error) { |
| 577 | console.error("sandbox alarm failed", this.ctx.id.toString(), `retry ${alarmProps?.retryCount ?? 0}`, describeError(error)); |
| 578 | throw error; |
| 579 | } |
| 580 | } |
| 581 | |
| 582 | /** |
| 583 | * A self-hosted runner's task ended (the actions service says so, or g1t |
| 584 | * stopped it): everything a container's stop does, once. |
| 585 | */ |
| 586 | async remoteEnded(exitCode: number, reason: string | null): Promise<void> { |
| 587 | if (!(await this.ctx.storage.get<boolean>("remote"))) return; |
| 588 | await this.ctx.storage.delete("remote"); |
| 589 | await this.ctx.storage.put("selfHostedEnded", true); |
| 590 | if (exitCode !== 0 && reason && !(await this.ctx.storage.get<string>("stopReason"))) { |
| 591 | await this.ctx.storage.put("stopReason", reason); |
| 592 | } |
| 593 | await this.onStop({ exitCode, reason: "exit" } as StopParams); |
| 594 | await this.ctx.storage.delete("selfHostedEnded"); |
| 595 | } |
| 596 | |
| 597 | /** |
| 598 | * Ends the run's record, once. Returns the status it ended with: |
| 599 | * `stopped` when a person stopped it first. |
| 600 | */ |
| 601 | private async closeRun(outcome: "succeeded" | "failed", error?: string): Promise<string | null> { |
| 602 | const tracked = await this.ctx.storage.get<TrackedRun>("agentRun"); |
| 603 | if (!tracked) return null; |
| 604 | await this.ctx.storage.delete("agentRun"); |
| 605 | const closed = await agentsClient(this.env.WORK) |
| 606 | .closeRun(tracked.runId, tracked.token, outcome, error) |
| 607 | .catch(() => null); |
| 608 | return closed?.ok ? closed.value : null; |
| 609 | } |
| 610 | |
| 611 | /** |
| 612 | * Reports how long the sandbox ran, once, whatever it exited with. The |
| 613 | * usage is fixed now and kept (`owed:<reference>`) until billing |
| 614 | * confirms it, tried again from the object's alarm while it does not: a |
| 615 | * re-run in this object writes a new `meter`, never over what is owed. |
| 616 | * Billing records a reference once, so a try whose answer was lost |
| 617 | * cannot charge twice. The reservation is settled apart (`settleStopped`). |
| 618 | */ |
| 619 | private async meterStop(): Promise<void> { |
| 620 | const metered = await this.ctx.storage.get<Meter & { started: number }>("meter"); |
| 621 | if (!metered) return; |
| 622 | const seconds = Math.max(1, Math.ceil((Date.now() - metered.started) / 1000)); |
| 623 | const run = await this.ctx.storage.get<Run>("run"); |
| 624 | // On the workspace's own runner: its minutes, at $0. |
| 625 | const selfHosted = (await this.ctx.storage.get<boolean>("selfHostedEnded")) ?? false; |
| 626 | const usage: SandboxUsage = { |
| 627 | workspace: metered.workspace, |
| 628 | seconds, |
| 629 | description: selfHosted ? `${metered.description} on a self-hosted runner` : metered.description, |
| 630 | repo: metered.repo, |
| 631 | reference: `sandbox/${this.ctx.id.toString()}/${metered.started}`, |
| 632 | // Whether g1t's open-source pool may pay for it. |
| 633 | kind: run ? computeKindOf(run.kind) : null, |
| 634 | selfHosted, |
| 635 | instance: metered.instance ?? null, |
| 636 | }; |
| 637 | // One write: the meter becomes what is owed. |
| 638 | await this.ctx.storage.put<Owed<SandboxUsage>>(`owed:${usage.reference}`, { usage, attempts: 0, since: Date.now() }); |
| 639 | await this.ctx.storage.delete("meter"); |
| 640 | await this.recordOwedTime(); |
| 641 | } |
| 642 | |
| 643 | /** |
| 644 | * Records the sandbox time billing has not confirmed, and schedules the |
| 645 | * next try for what it still has not. Never throws: it runs in the stop |
| 646 | * hook and in the object's alarm. |
| 647 | */ |
| 648 | async recordOwedTime(): Promise<void> { |
| 649 | const owed = [...(await this.ctx.storage.list<Owed<SandboxUsage>>({ prefix: "owed:" })).values()]; |
| 650 | if (!owed.length) return; |
| 651 | const billing = billingClient(this.env.BILLING); |
| 652 | const result = await recordOwed( |
| 653 | owed, |
| 654 | async (usage) => { |
| 655 | const recorded = await billing.recordSandbox(usage); |
| 656 | if (!recorded.ok) console.log("sandbox time not recorded yet", usage.workspace, usage.seconds, recorded.error.message); |
| 657 | return recorded.ok; |
| 658 | }, |
| 659 | Date.now(), |
| 660 | ); |
| 661 | for (const entry of result.recorded) await this.ctx.storage.delete(`owed:${entry.usage.reference}`); |
| 662 | for (const entry of result.kept) await this.ctx.storage.put(`owed:${entry.usage.reference}`, entry); |
| 663 | for (const entry of result.lost) { |
| 664 | console.error("sandbox time not recorded, given up", entry.usage.reference, entry.usage.workspace, entry.usage.seconds, `after ${entry.attempts} tries`); |
| 665 | await this.ctx.storage.delete(`owed:${entry.usage.reference}`); |
| 666 | } |
| 667 | this.deleteSchedules("recordOwedTime"); |
| 668 | if (result.retryInSeconds !== null) { |
| 669 | await this.schedule(result.retryInSeconds, "recordOwedTime").catch((error: unknown) => |
| 670 | console.error("sandbox time retry not scheduled", this.ctx.id.toString(), describeError(error)), |
| 671 | ); |
| 672 | } |
| 673 | } |
| 674 | |
| 675 | override async onStop({ exitCode, reason }: StopParams): Promise<void> { |
| 676 | await revokeCredentials(this.env.IDENTITY, this.ctx.storage); |
| 677 | const started = await this.ctx.storage.get<number>("started"); |
| 678 | await this.meterStop(); |
| 679 | // It stopped itself for mining, and could not say so before it went. |
| 680 | if (exitCode === ABUSE_EXIT_CODE && !(await this.ctx.storage.get<boolean>("abuse"))) { |
| 681 | await this.flagAbuse(null); |
| 682 | } |
| 683 | const flagged = (await this.ctx.storage.get<boolean>("abuse")) ?? false; |
| 684 | // Why it stopped, when g1t stopped it: said in place of a plain failure. |
| 685 | const why = flagged ? ABUSE_MESSAGE : ((await this.ctx.storage.get<string>("stopReason")) ?? null); |
| 686 | // Stopped by g1t, it failed whatever exit code the library passed on. |
| 687 | const short = stoppedShort(exitCode, why); |
| 688 | await this.closeRun( |
| 689 | short ? "failed" : "succeeded", |
| 690 | short ? (why ?? `The sandbox exited with ${exitCode}.`) : undefined, |
| 691 | ); |
| 692 | await this.settleStopped(started); |
| 693 | await this.ctx.storage.delete("started"); |
| 694 | if (!short) return; |
| 695 | const run = await this.ctx.storage.get<Run>("run"); |
| 696 | console.log("sandbox stopped", run?.kind, "exit", exitCode, reason); |
| 697 | if (!run) return; |
| 698 | // Never thrown: this runs in the sandbox's alarm, which a throw would |
| 699 | // fail, retry and count as an error, running all of the above again. |
| 700 | // What could not be told is logged, and the sweep catches it up. |
| 701 | await tellStopped(run.kind, () => this.reportStopped(run, why, exitCode)); |
| 702 | } |
| 703 | |
| 704 | /** |
| 705 | * Tells whoever is waiting on the sandbox's work that it stopped without |
| 706 | * finishing it. Each is refused harmlessly when the sandbox reported its |
| 707 | * end before it stopped. Throws when the service could not be reached. |
| 708 | */ |
| 709 | private async reportStopped(run: Run, why: string | null, exitCode: number): Promise<void> { |
| 710 | if (run.kind === "actions") { |
| 711 | // Refused harmlessly if the job reported its end before it stopped. |
| 712 | const response = await this.env.ACTIONS.fetch("https://actions/rpc/job_report", { |
| 713 | method: "POST", |
| 714 | headers: { "content-type": "application/json" }, |
| 715 | body: JSON.stringify({ |
| 716 | job: run.jobId, |
| 717 | token: run.token, |
| 718 | report: { kind: "done", conclusion: "failure", reason: why ?? "The runner stopped before the job finished." }, |
| 719 | }), |
| 720 | }); |
| 721 | await answered("job_report", response); |
| 722 | return; |
| 723 | } |
| 724 | // Nothing was pushed, so no pull request opens; why is in its log. |
| 725 | if (run.kind === "bump") return; |
| 726 | if (run.kind === "backup") { |
| 727 | // Refused harmlessly if the sandbox reported before it stopped; the |
| 728 | // job is otherwise tried again later tonight. |
| 729 | await reposClient(this.env.REPOS).failBackup(run.jobId, run.token, why ?? `The sandbox exited with ${exitCode}.`); |
| 730 | return; |
| 731 | } |
| 732 | if (run.kind === "deploy") { |
| 733 | // Refused harmlessly if the build reported its end before it stopped. |
| 734 | const response = await this.env.DEPLOYMENTS.fetch(`https://deployments/jobs/${run.deployId}/fail`, { |
| 735 | method: "POST", |
| 736 | headers: { "content-type": "application/json" }, |
| 737 | body: JSON.stringify({ token: run.token, message: why ?? "The build stopped before it finished." }), |
| 738 | }); |
| 739 | await answered("deploy fail", response); |
| 740 | return; |
| 741 | } |
| 742 | const work = workClient(this.env.WORK); |
| 743 | if (run.kind === "checks") { |
| 744 | // Refused harmlessly if the run did report before it stopped. |
| 745 | await work.reportChecks(run.runId, run.token, { |
| 746 | error: why ?? "The sandbox stopped before the checks finished.", |
| 747 | }); |
| 748 | return; |
| 749 | } |
| 750 | if (run.kind === "queue") { |
| 751 | // Refused harmlessly if the state was reported before it stopped. |
| 752 | await work.failQueue(run.entryId, run.token, why ?? "The sandbox stopped before the state was checked."); |
| 753 | return; |
| 754 | } |
| 755 | if (run.kind === "mergecheck") { |
| 756 | // Refused harmlessly if the probe reported before it stopped. |
| 757 | await work.failMergecheck(run.pullId, run.token, why ?? "The sandbox stopped before the merge check finished."); |
| 758 | return; |
| 759 | } |
| 760 | // A sandbox of a kind g1t no longer starts, which ran from before: it |
| 761 | // has nobody left to tell. |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | /** |
| 766 | * What the compute gate decided for one start: go, with what billing |
| 767 | * reserved and the plan's caps; or not, refused with what to tell people. |
| 768 | */ |
| 769 | type Granted = { |
| 770 | ok: true; |
| 771 | held: Held | null; |
| 772 | limits: PlanLimits; |
| 773 | /** The labels of the self-hosted runners it goes to; null for a sandbox. */ |
| 774 | route: string[] | null; |
| 775 | }; |
| 776 | type Admitted = Granted | { ok: false; code: string; message: string }; |
| 777 | |
| 778 | /** |
| 779 | * The guardrails' default time cap of each kind of run, for estimating what |
| 780 | * it may cost before it starts; the sandbox applies the project's own. |
| 781 | */ |
| 782 | const DEFAULT_MINUTES: Record<"queue" | "mergecheck", number> = { |
| 783 | queue: 45, |
| 784 | mergecheck: 10, |
| 785 | }; |
| 786 | |
| 787 | /** A plan's cap on one run, for the sandbox to apply under its guardrails'. */ |
| 788 | function limitsOf(ent: ComputeEntitlements | null): PlanLimits { |
| 789 | return { minutes: ent && ent.maxRunMinutes > 0 ? ent.maxRunMinutes : null }; |
| 790 | } |
| 791 | |
| 792 | /** The shorter of a kind's time cap and the plan's, for an estimate. */ |
| 793 | function estimateMinutes(minutes: number, ent: ComputeEntitlements | null): number { |
| 794 | return ent && ent.maxRunMinutes > 0 ? Math.min(minutes, ent.maxRunMinutes) : minutes; |
| 795 | } |
| 796 | |
| 797 | /** A start the gate did not let through, as a result for whoever asked. */ |
| 798 | function notAdmitted(admitted: Exclude<Admitted, Granted>): Result<never> { |
| 799 | return fail("payment_required", admitted.message); |
| 800 | } |
| 801 | |
| 802 | /** |
| 803 | * Larger sandboxes for workflow jobs that ask for one in `runs-on`: the |
| 804 | * same image and behaviour on a larger Containers instance type, each a |
| 805 | * class of its own (wrangler.jsonc). Outbound handlers are registered by |
| 806 | * class, so each registers its own. |
| 807 | */ |
| 808 | export class Sandbox2Core extends AttemptSandbox { |
| 809 | static { |
| 810 | Sandbox2Core.outboundHandlers = { egress, abuse }; |
| 811 | } |
| 812 | } |
| 813 | export class Sandbox4Core extends AttemptSandbox { |
| 814 | static { |
| 815 | Sandbox4Core.outboundHandlers = { egress, abuse }; |
| 816 | } |
| 817 | } |
| 818 | |
| 819 | /** |
| 820 | * What the actions service sends to start a job (`StartJobArgs`). Callers |
| 821 | * from before may also send `workflow`, `environment` and `trusted`, for |
| 822 | * workflow-only domains, which are gone: they are ignored. |
| 823 | */ |
| 824 | type ActionsJobArgs = { |
| 825 | job: string; |
| 826 | token: string; |
| 827 | repo: RepoPath; |
| 828 | timeoutMinutes: number; |
| 829 | /** The machine its `runs-on` asked for, by label; absent, the standard one. */ |
| 830 | instance?: string | null; |
| 831 | /** Whether it may be given brokered secrets, which its requests need filled in. */ |
| 832 | brokered?: boolean; |
| 833 | }; |
| 834 | |
| 835 | export default class RunnerService extends WorkerEntrypoint<RunnerEnv> { |
| 836 | /** |
| 837 | * The JSON protocol the Rust services speak: `POST /rpc/<method>` with the |
| 838 | * arguments as the body. Only bound services can. |
| 839 | */ |
| 840 | async fetch(request: Request): Promise<Response> { |
| 841 | const { pathname } = new URL(request.url); |
| 842 | if (request.method === "POST" && pathname === "/rpc/start_actions_job") { |
| 843 | return Response.json(await this.startActionsJob((await request.json()) as ActionsJobArgs)); |
| 844 | } |
| 845 | if (request.method === "POST" && pathname === "/rpc/stop_actions_job") { |
| 846 | const args = (await request.json()) as { job: string }; |
| 847 | // Whichever machine it asked for: the job's object in every namespace. |
| 848 | await Promise.all( |
| 849 | Object.keys(SANDBOX_BINDINGS).map((className) => { |
| 850 | const namespace = sandboxNamespace(this.env, className) as unknown as DurableObjectNamespace<AttemptSandbox>; |
| 851 | return namespace |
| 852 | .get(namespace.idFromName(`actions:${args.job}`)) |
| 853 | .destroy() |
| 854 | .catch(() => undefined); |
| 855 | }), |
| 856 | ); |
| 857 | return Response.json(ok(true)); |
| 858 | } |
| 859 | // A self-hosted runner's task ended: the sandbox that handed it over |
| 860 | // does what it does when a container stops. |
| 861 | if (request.method === "POST" && pathname === "/rpc/task_ended") { |
| 862 | const args = (await request.json()) as { sandbox: string; exitCode: number; reason?: string | null }; |
| 863 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromString(args.sandbox)); |
| 864 | await sandbox.remoteEnded(args.exitCode, args.reason ?? null); |
| 865 | return Response.json(ok(true)); |
| 866 | } |
| 867 | // A self-hosted job's request to a brokered secret's host, which the |
| 868 | // API passes on as it came (broker-relay.ts). |
| 869 | if (request.method === "POST" && pathname === "/rpc/broker") { |
| 870 | try { |
| 871 | return await relay(request, relayDeps(this.env.ACTIONS, (upstream) => forward(upstream))); |
| 872 | } catch (error) { |
| 873 | console.log("broker relay failed", String(error)); |
| 874 | return Response.json({ error: { code: "unavailable", message: "g1t could not relay the request." } }, { status: 503 }); |
| 875 | } |
| 876 | } |
| 877 | if (request.method === "POST" && pathname === "/rpc/bump") { |
| 878 | return Response.json(await this.startBump(await request.json())); |
| 879 | } |
| 880 | if (request.method === "POST" && pathname === "/rpc/start_deploy") { |
| 881 | return Response.json(await this.startDeploy((await request.json()) as DeployJob)); |
| 882 | } |
| 883 | return new Response("Not found\n", { status: 404 }); |
| 884 | } |
| 885 | |
| 886 | // ---- The compute gate (@g1t/contracts compute.ts) ------------------------- |
| 887 | |
| 888 | /** Whether `repo` is public: what g1t's open-source pool can pay for. */ |
| 889 | private async isPublic(repo: RepoPath): Promise<boolean> { |
| 890 | const found = await reposClient(this.env.REPOS) |
| 891 | .get(repo, null) |
| 892 | .catch(() => null); |
| 893 | return Boolean(found?.ok && !found.value.isPrivate); |
| 894 | } |
| 895 | |
| 896 | /** |
| 897 | * Whether a sandbox (the merge queue, a |
| 898 | * merge check, a workflow job) may start in `repo`, with what it may cost |
| 899 | * for `minutes` reserved. Public repositories' checks, workflows and |
| 900 | * queue can be paid by the open-source pool. Never throws. |
| 901 | */ |
| 902 | private async admitSandbox( |
| 903 | kind: ComputeKind, |
| 904 | repo: RepoPath, |
| 905 | minutes: number, |
| 906 | instance: InstanceType = STANDARD_INSTANCE, |
| 907 | { selfHosted = true }: { selfHosted?: boolean } = {}, |
| 908 | ): Promise<Admitted> { |
| 909 | const workspace = repo.namespace.toLowerCase(); |
| 910 | const compute = gateFor(this.env); |
| 911 | const ent = await compute.entitlements(workspace); |
| 912 | if (ent?.paused) return { ok: false, code: "paused", message: refusalMessage("paused", workspace, kind, ent.paused) }; |
| 913 | // Checks and the merge queue go to the workspace's own runners when it |
| 914 | // says so, and cost nothing there. Workflow jobs choose with `runs-on`. |
| 915 | const route = selfHosted && (kind === "check" || kind === "queue") ? await selfHostedRoute(this.env.ACTIONS, repo) : null; |
| 916 | if (route) return { ok: true, held: null, limits: limitsOf(ent), route }; |
| 917 | const [standardMicros, isPublic] = await Promise.all([compute.microsPerSecond(), this.isPublic(repo)]); |
| 918 | // Reserved for its time limit at the larger of what it may cost (the |
| 919 | // sandbox price, scaled to its machine) and its minutes at the |
| 920 | // machine's rate, which is what it is charged. |
| 921 | const limit = estimateMinutes(minutes, ent); |
| 922 | const microsPerSecond = standardMicros * priceScale(instance); |
| 923 | const estimateMicros = Math.max( |
| 924 | sandboxEstimateMicros(limit, microsPerSecond), |
| 925 | runnerMinutes(limit * 60) * instance.microsPerMinute, |
| 926 | ); |
| 927 | const admission = await compute.admit({ workspace, repo, public: isPublic, kind, estimateMicros }, ent); |
| 928 | if (!admission.ok) return { ok: false, code: admission.code, message: admission.message }; |
| 929 | return { |
| 930 | ok: true, |
| 931 | held: admission.reservation ? { id: admission.reservation.id, workspace, microsPerSecond } : null, |
| 932 | limits: limitsOf(ent), |
| 933 | route: null, |
| 934 | }; |
| 935 | } |
| 936 | |
| 937 | /** Gives back what was reserved for a start that never reached its sandbox. */ |
| 938 | private async release(held: Held | null): Promise<void> { |
| 939 | if (held) await gateFor(this.env).settle(held.id, 0); |
| 940 | } |
| 941 | |
| 942 | /** |
| 943 | * Runs `start`, giving back what was reserved if it fails. A sandbox that |
| 944 | * could not start has given it back already; settling twice at nothing |
| 945 | * is harmless. |
| 946 | */ |
| 947 | private async holding<T>(granted: Granted, start: () => Promise<T>): Promise<T> { |
| 948 | try { |
| 949 | return await start(); |
| 950 | } catch (error) { |
| 951 | await this.release(granted.held); |
| 952 | throw error; |
| 953 | } |
| 954 | } |
| 955 | |
| 956 | /** |
| 957 | * Starts one job of a GitHub Actions workflow in a sandbox of its own. |
| 958 | * The sandbox fetches the job, its contexts and its secrets with the |
| 959 | * job's token, and reports back to the actions service through the API. |
| 960 | * Jobs run on g1t's machines, so only for workspaces that may use them. |
| 961 | */ |
| 962 | private async startActionsJob(args: ActionsJobArgs): Promise<Result<true>> { |
| 963 | // The machine its `runs-on` asked for; the standard one otherwise. |
| 964 | const instance = instanceNamed(args.instance); |
| 965 | // Workflow jobs run on g1t's machines: only as the workspace's plan |
| 966 | // allows, or on a public repository, from the open-source pool. |
| 967 | const admitted = await this.admitSandbox("workflow", args.repo, args.timeoutMinutes, instance); |
| 968 | if (!admitted.ok) return fail("payment_required", `Not started: ${admitted.message}`); |
| 969 | const namespace = this.jobNamespace(instance); |
| 970 | if (!namespace) { |
| 971 | await this.release(admitted.held); |
| 972 | return fail("invalid", `Not started: ${instance.label} machines are not available here.`); |
| 973 | } |
| 974 | const sandbox = namespace.get(namespace.idFromName(`actions:${args.job}`)); |
| 975 | const on = instance === STANDARD_INSTANCE ? "" : ` on ${instance.label}`; |
| 976 | try { |
| 977 | await sandbox.run({ |
| 978 | kind: "actions", |
| 979 | jobId: args.job, |
| 980 | token: args.token, |
| 981 | reservation: admitted.held, |
| 982 | limits: admitted.limits, |
| 983 | // The job's own time limit, and whether its requests pass the |
| 984 | // Worker for its brokered secrets. |
| 985 | build: { |
| 986 | kind: "actions", |
| 987 | minutes: Math.max(1, args.timeoutMinutes), |
| 988 | brokered: args.brokered === true, |
| 989 | }, |
| 990 | meter: { |
| 991 | ...meter(args.repo, `A workflow job in ${args.repo.namespace}/${args.repo.name}${on}`), |
| 992 | instance: instance === STANDARD_INSTANCE ? null : instance.label, |
| 993 | }, |
| 994 | envVars: { |
| 995 | MODE: "actions", |
| 996 | G1T_API: "https://api.g1t.sh", |
| 997 | ACTIONS_JOB: args.job, |
| 998 | ACTIONS_TOKEN: args.token, |
| 999 | // Docker of the job's own, inside its sandbox (crates/runner docker/). |
| 1000 | G1T_DOCKER: dockerFor(this.env.DOCKER), |
| 1001 | }, |
| 1002 | }); |
| 1003 | } catch (error) { |
| 1004 | // A sandbox that could not start, or stopped at once: the job fails |
| 1005 | // with why, rather than waiting to be noticed. |
| 1006 | return { |
| 1007 | ok: false, |
| 1008 | error: { code: "conflict", message: `The runner could not start the job: ${String(error).replace(/^Error: /, "")}` }, |
| 1009 | }; |
| 1010 | } |
| 1011 | // `true`, not null: an outcome needs a value. |
| 1012 | return ok(true); |
| 1013 | } |
| 1014 | |
| 1015 | /** The sandboxes of a label: the class the catalogue names for it, by its binding. */ |
| 1016 | private jobNamespace(instance: InstanceType): DurableObjectNamespace<AttemptSandbox> | null { |
| 1017 | return jobNamespaceOf(this.env, instance) as DurableObjectNamespace<AttemptSandbox> | null; |
| 1018 | } |
| 1019 | |
| 1020 | /** |
| 1021 | * Builds one commit in a sandbox of its own and deploys it to g1t.page. |
| 1022 | * Asked by the deployments service, which has already checked that the |
| 1023 | * workspace pays for Deployments; that plan is what |
| 1024 | * lets a build use g1t's machines. |
| 1025 | */ |
| 1026 | private async startDeploy(job: DeployJob): Promise<Result<true>> { |
| 1027 | // To read the commit, which may be private, as whoever pushed it. |
| 1028 | const token = await runCredential(this.env.IDENTITY, { |
| 1029 | onBehalfOf: job.actor, |
| 1030 | repo: job.source, |
| 1031 | kind: "deploy", |
| 1032 | use: "runner", |
| 1033 | read: [job.source], |
| 1034 | ttlSeconds: DEPLOY_TOKEN_TTL_SECONDS, |
| 1035 | }); |
| 1036 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(`deploy:${job.deployId}`)); |
| 1037 | const workspace = (job.workspace ?? job.source.namespace).toLowerCase(); |
| 1038 | // The project the build is for, and who it is charged to. |
| 1039 | const project = job.repo ?? job.source; |
| 1040 | try { |
| 1041 | await sandbox.run({ |
| 1042 | kind: "deploy", |
| 1043 | deployId: job.deployId, |
| 1044 | token: job.token, |
| 1045 | reservation: job.reservation |
| 1046 | ? { id: job.reservation, workspace, microsPerSecond: job.microsPerSecond ?? 0 } |
| 1047 | : null, |
| 1048 | limits: { minutes: job.maxRunMinutes ?? null }, |
| 1049 | // For as long as its read token lasts. |
| 1050 | build: { kind: "deploy", minutes: DEPLOY_TOKEN_TTL_SECONDS / 60 }, |
| 1051 | owner: { workspace, repo: `${project.namespace}/${project.name}` }, |
| 1052 | envVars: { |
| 1053 | MODE: "deploy", |
| 1054 | G1T_API: "https://api.g1t.sh", |
| 1055 | DEPLOY_ID: job.deployId, |
| 1056 | DEPLOY_TOKEN: job.token, |
| 1057 | G1T_USER: job.actor.username, |
| 1058 | G1T_TOKEN: token, |
| 1059 | GIT_REMOTE: `https://g1t.sh/${job.source.namespace}/${job.source.name}.git`, |
| 1060 | GIT_COMMIT: job.commit, |
| 1061 | ROOT_DIR: job.rootDir ?? "", |
| 1062 | BUILD_COMMAND: job.buildCommand ?? "", |
| 1063 | OUTPUT_DIR: job.outputDir ?? "", |
| 1064 | BUILD_AS: job.buildAs ?? "auto", |
| 1065 | DOCKERFILE: job.dockerfile ?? "", |
| 1066 | // A container build's own Docker Engine (crates/runner docker/), |
| 1067 | // started only if the project is built as one. |
| 1068 | G1T_DOCKER: dockerFor(this.env.DOCKER), |
| 1069 | BUILD_ENV: JSON.stringify(job.buildEnv ?? {}), |
| 1070 | BUILD_SECRETS: JSON.stringify(job.buildSecrets ?? {}), |
| 1071 | }, |
| 1072 | }); |
| 1073 | } catch (error) { |
| 1074 | return { |
| 1075 | ok: false, |
| 1076 | error: { code: "conflict", message: `The runner could not start the build: ${String(error).replace(/^Error: /, "")}` }, |
| 1077 | }; |
| 1078 | } |
| 1079 | return ok(true); |
| 1080 | } |
| 1081 | |
| 1082 | /** |
| 1083 | * Makes a security update in a sandbox of its own (crates/runner |
| 1084 | * bump.rs): raises one package to a fixed version in the lockfiles |
| 1085 | * named, commits that as g1t and pushes it to its `g1t/security/…` |
| 1086 | * branch. A version update (`kind: "version"`) raises one or more |
| 1087 | * packages the same way, to the branch its dependency update file names, |
| 1088 | * which is never the default one. Asked by the security service, which |
| 1089 | * opens the pull request when it hears the push; nothing here opens one. |
| 1090 | * Admitted, reserved and metered like checks, always in g1t's sandbox (a |
| 1091 | * self-hosted runner may not know the mode), under the project's network |
| 1092 | * list plus the package registries. Returns whether the sandbox started. |
| 1093 | */ |
| 1094 | private async startBump(input: unknown): Promise<Result<boolean>> { |
| 1095 | const problem = bumpProblem(input, UPDATE_BRANCH_PREFIX); |
| 1096 | if (problem) return fail("invalid", problem); |
| 1097 | const args = input as BumpArgs; |
| 1098 | const repo = args.repo; |
| 1099 | const what = args.kind === "version" ? "version update" : "security update"; |
| 1100 | const actor = systemActor(repo.namespace); |
| 1101 | const closed = await this.closedRepo(actor, repo); |
| 1102 | if (closed) return closed; |
| 1103 | const admitted = await this.admitSandbox("check", repo, BUMP_MINUTES, STANDARD_INSTANCE, { selfHosted: false }); |
| 1104 | if (!admitted.ok) return notAdmitted(admitted); |
| 1105 | try { |
| 1106 | const defaultBranch = await this.defaultBranch(repo, actor); |
| 1107 | // From its `target-branch`, which its pull request merges into, or |
| 1108 | // the default branch. |
| 1109 | const base = args.base ?? defaultBranch; |
| 1110 | // A version update names its own branch, which is never the one it |
| 1111 | // starts from, nor the default one. |
| 1112 | if (args.kind === "version" && (args.branch === defaultBranch || args.branch === base)) { |
| 1113 | await this.release(admitted.held); |
| 1114 | return fail("invalid", `A version update cannot push to ${args.branch}, the branch it starts from.`); |
| 1115 | } |
| 1116 | // As g1t, for the workspace: reads the repository and pushes this |
| 1117 | // branch only, with no API operations. |
| 1118 | const token = await runCredential(this.env.IDENTITY, { |
| 1119 | onBehalfOf: actor, |
| 1120 | repo, |
| 1121 | kind: "bump", |
| 1122 | use: "runner", |
| 1123 | read: [repo], |
| 1124 | push: [{ repo, branch: args.branch }], |
| 1125 | ttlSeconds: BUMP_TOKEN_TTL_SECONDS, |
| 1126 | }); |
| 1127 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(bumpSandboxName(args))); |
| 1128 | await sandbox.run({ |
| 1129 | kind: "bump", |
| 1130 | repo, |
| 1131 | branch: args.branch, |
| 1132 | reservation: admitted.held, |
| 1133 | limits: admitted.limits, |
| 1134 | build: { kind: "bump", minutes: BUMP_MINUTES }, |
| 1135 | meter: meter(repo, `${args.kind === "version" ? "Version" : "Security"} update in ${repo.namespace}/${repo.name}`), |
| 1136 | envVars: bumpEnv(args, base, token), |
| 1137 | }); |
| 1138 | } catch (error) { |
| 1139 | await this.release(admitted.held); |
| 1140 | return fail("conflict", `The runner could not start the ${what}: ${String(error).replace(/^Error: /, "")}`); |
| 1141 | } |
| 1142 | return ok(true); |
| 1143 | } |
| 1144 | |
| 1145 | /** |
| 1146 | * Events from the bus. Each one that could change what a pull request |
| 1147 | * needs next moves it along: the work service decides whether it can |
| 1148 | * merge by the repository's settings, and merges it if so. Merge checks |
| 1149 | * and the merge queue's states start here too. |
| 1150 | */ |
| 1151 | async queue(batch: MessageBatch<G1tEvent>): Promise<void> { |
| 1152 | for (const message of batch.messages) { |
| 1153 | const event = message.body; |
| 1154 | switch (event.type) { |
| 1155 | case "pull.opened": |
| 1156 | case "pull.ready": |
| 1157 | case "pull.updated": |
| 1158 | case "checks.completed": |
| 1159 | case "review.completed": |
| 1160 | case "pull.mergeability": |
| 1161 | await this.advance(event.data.pullId); |
| 1162 | break; |
| 1163 | // Its head or its target moved and both changed the same files: |
| 1164 | // find out whether it still merges cleanly. |
| 1165 | case "pull.mergecheck": |
| 1166 | await this.startMergecheck(event.data.pullId); |
| 1167 | break; |
| 1168 | // Something joined, left or landed: test the next batch if none is. |
| 1169 | case "queue.changed": |
| 1170 | await this.buildQueue(event.data.repoId); |
| 1171 | break; |
| 1172 | // A person approved or asked for changes, which may let it merge. |
| 1173 | case "comment.created": |
| 1174 | if (event.data.pullId && event.data.verdict) await this.advance(event.data.pullId); |
| 1175 | break; |
| 1176 | // The branch the others would land on has moved. |
| 1177 | case "pull.merged": |
| 1178 | await this.advanceAll(event.data.repoId); |
| 1179 | break; |
| 1180 | // Read-only or gone: the runs there stop. |
| 1181 | case "repo.archived": |
| 1182 | case "repo.deleted": |
| 1183 | await this.stopRunsIn(event.data.repoId); |
| 1184 | break; |
| 1185 | } |
| 1186 | message.ack(); |
| 1187 | } |
| 1188 | } |
| 1189 | |
| 1190 | /** A sweep, for steps whose trigger was missed or whose sandbox died, and the nightly backups. */ |
| 1191 | async scheduled(): Promise<void> { |
| 1192 | await this.advanceAll(); |
| 1193 | await this.startBackups().catch((error: unknown) => console.log("backups not started", String(error))); |
| 1194 | } |
| 1195 | |
| 1196 | /** |
| 1197 | * Starts a few of the nightly backups the repos service queued, each in |
| 1198 | * a sandbox of its own that holds only its job's token: the sandbox asks |
| 1199 | * for a read-only git credential itself, when it is ready to clone. No |
| 1200 | * plan is asked and nothing is metered: backups are g1t's own work. |
| 1201 | */ |
| 1202 | private async startBackups(): Promise<void> { |
| 1203 | const pace = backupPace(this.env.BACKUPS_PER_SWEEP, this.env.BACKUPS_RUNNING); |
| 1204 | if (pace.perSweep === 0) return; |
| 1205 | const repos = reposClient(this.env.REPOS); |
| 1206 | for (const claim of await repos.claimBackups(pace.perSweep, pace.running)) { |
| 1207 | try { |
| 1208 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(backupSandboxName(claim))); |
| 1209 | await sandbox.run({ |
| 1210 | kind: "backup", |
| 1211 | jobId: claim.jobId, |
| 1212 | token: claim.token, |
| 1213 | // For `abuse.flagged`: whose repository it was. |
| 1214 | owner: { workspace: claim.path.namespace, repo: `${claim.path.namespace}/${claim.path.name}` }, |
| 1215 | envVars: backupEnv(claim, "https://api.g1t.sh"), |
| 1216 | }); |
| 1217 | } catch (error) { |
| 1218 | await repos.failBackup(claim.jobId, claim.token, `The sandbox could not start: ${String(error)}`).catch(() => null); |
| 1219 | } |
| 1220 | } |
| 1221 | } |
| 1222 | |
| 1223 | private async advanceAll(repoId?: string): Promise<void> { |
| 1224 | const pulls = await workClient(this.env.WORK).managedPulls(repoId); |
| 1225 | for (const pullId of pulls) await this.advance(pullId); |
| 1226 | } |
| 1227 | |
| 1228 | /** |
| 1229 | * Merges a pull request by the repository's settings when it is ready: |
| 1230 | * the work service decides and merges, and claims the step, so calling |
| 1231 | * this twice merges nothing twice. |
| 1232 | */ |
| 1233 | private async advance(pullId: string): Promise<void> { |
| 1234 | const next = await workClient(this.env.WORK).advance(pullId); |
| 1235 | // g1t no longer reviews, revises or catches up pull requests itself. |
| 1236 | if (next.action !== "none") console.log("lifecycle step not taken", pullId, next.action); |
| 1237 | } |
| 1238 | |
| 1239 | /** |
| 1240 | * Starts the next batch of a repository's merge queue, if it has one |
| 1241 | * ready: a sandbox per entry, all at once, each building the default |
| 1242 | * branch with that entry and everything ahead of it. |
| 1243 | */ |
| 1244 | private async buildQueue(repoId: string): Promise<void> { |
| 1245 | const work = workClient(this.env.WORK); |
| 1246 | const jobs = await work.queueBuild(repoId); |
| 1247 | // Merge queue sandboxes, like any other, only as the workspace's plan |
| 1248 | // allows: refused states fail at once, saying why. A state whose |
| 1249 | // sandbox could not start fails at once too, rather than holding the |
| 1250 | // queue until it times out. |
| 1251 | await Promise.all( |
| 1252 | jobs.map(async (job) => { |
| 1253 | const admitted = await this.admitSandbox("queue", job.repo, DEFAULT_MINUTES.queue); |
| 1254 | if (!admitted.ok) { |
| 1255 | await work.failQueue(job.entryId, job.token, `Not started: ${admitted.message}`); |
| 1256 | return; |
| 1257 | } |
| 1258 | await this.holding(admitted, () => this.startQueueRun(job, admitted)).catch((error: unknown) => |
| 1259 | work.failQueue(job.entryId, job.token, `Its sandbox could not start: ${String(error)}`), |
| 1260 | ); |
| 1261 | }), |
| 1262 | ); |
| 1263 | } |
| 1264 | |
| 1265 | private async startQueueRun(job: QueueJob, granted: Granted): Promise<void> { |
| 1266 | // To read the changes and push the tested state, as a member. |
| 1267 | // Reads each queued change; pushes only the queue's own branch. |
| 1268 | const token = await runCredential(this.env.IDENTITY, { |
| 1269 | onBehalfOf: job.actor, |
| 1270 | repo: job.repo, |
| 1271 | kind: "queue", |
| 1272 | use: "runner", |
| 1273 | number: job.stack.at(-1)?.number ?? null, |
| 1274 | read: job.stack.map((item) => item.source), |
| 1275 | push: [{ repo: job.repo, branch: job.branch }], |
| 1276 | ttlSeconds: CHECKS_TOKEN_TTL_SECONDS, |
| 1277 | }); |
| 1278 | const remote = (path: RepoPath) => `https://g1t.sh/${path.namespace}/${path.name}.git`; |
| 1279 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(`queue-${job.entryId}-${job.baseCommit}`)); |
| 1280 | await sandbox.run({ |
| 1281 | kind: "queue", |
| 1282 | entryId: job.entryId, |
| 1283 | token: job.token, |
| 1284 | reservation: granted.held, |
| 1285 | limits: granted.limits, |
| 1286 | selfHosted: granted.route, |
| 1287 | track: { |
| 1288 | actor: job.actor, |
| 1289 | repo: job.repo, |
| 1290 | kind: "queue", |
| 1291 | number: job.stack.at(-1)?.number ?? null, |
| 1292 | title: `Merge queue: ${job.stack.map((item) => `#${item.number}`).join(" + ")}`, |
| 1293 | }, |
| 1294 | meter: meter(job.repo, `Merge queue on ${job.repo.namespace}/${job.repo.name}`), |
| 1295 | envVars: { |
| 1296 | MODE: "queue", |
| 1297 | G1T_API: "https://api.g1t.sh", |
| 1298 | QUEUE_ENTRY: job.entryId, |
| 1299 | QUEUE_TOKEN: job.token, |
| 1300 | G1T_USER: job.actor.username, |
| 1301 | G1T_TOKEN: token, |
| 1302 | BASE_REMOTE: remote(job.repo), |
| 1303 | BASE_COMMIT: job.baseCommit, |
| 1304 | QUEUE_BRANCH: job.branch, |
| 1305 | STACK: JSON.stringify( |
| 1306 | job.stack.map((item) => ({ |
| 1307 | number: item.number, |
| 1308 | title: item.title, |
| 1309 | remote: remote(item.source), |
| 1310 | branch: item.branch, |
| 1311 | commit: item.commit, |
| 1312 | })), |
| 1313 | ), |
| 1314 | CHECKS: JSON.stringify(job.checks), |
| 1315 | CONTRACT_CHECKS: JSON.stringify(job.contractChecks), |
| 1316 | }, |
| 1317 | }); |
| 1318 | } |
| 1319 | |
| 1320 | /** |
| 1321 | * Merges a pull request's head into its target in a sandbox of its own, |
| 1322 | * pushing nothing, to find the files that conflict. |
| 1323 | * The work service decides when one is needed and how many may run. |
| 1324 | */ |
| 1325 | private async startMergecheck(pullId: string): Promise<void> { |
| 1326 | const work = workClient(this.env.WORK); |
| 1327 | const started = await work.startMergecheck(pullId); |
| 1328 | if (!started.ok) return; |
| 1329 | const job = started.value; |
| 1330 | let granted: Granted | null = null; |
| 1331 | try { |
| 1332 | // Like any sandbox, only as the workspace's plan allows. |
| 1333 | const admitted = await this.admitSandbox("check", job.repo, DEFAULT_MINUTES.mergecheck); |
| 1334 | if (!admitted.ok) throw new Error(`Not started: ${admitted.message}`); |
| 1335 | granted = admitted; |
| 1336 | // To read the change, which may be private, as whoever opened it. |
| 1337 | const token = await runCredential(this.env.IDENTITY, { |
| 1338 | onBehalfOf: job.author, |
| 1339 | repo: job.repo, |
| 1340 | kind: "mergecheck", |
| 1341 | use: "runner", |
| 1342 | number: job.number, |
| 1343 | read: [job.repo, job.source], |
| 1344 | ttlSeconds: MERGECHECK_TOKEN_TTL_SECONDS, |
| 1345 | }); |
| 1346 | const remote = (path: RepoPath) => `https://g1t.sh/${path.namespace}/${path.name}.git`; |
| 1347 | // One sandbox per pair of commits: asking twice starts nothing twice. |
| 1348 | const sandbox = this.env.SANDBOX.get( |
| 1349 | this.env.SANDBOX.idFromName(`mergecheck-${job.pullId}-${job.head}-${job.base}`), |
| 1350 | ); |
| 1351 | await sandbox.run({ |
| 1352 | kind: "mergecheck", |
| 1353 | pullId: job.pullId, |
| 1354 | token: job.token, |
| 1355 | reservation: granted.held, |
| 1356 | limits: granted.limits, |
| 1357 | selfHosted: granted.route, |
| 1358 | meter: meter(job.repo, `Merge check of ${job.repo.namespace}/${job.repo.name}#${job.number}`), |
| 1359 | envVars: { |
| 1360 | MODE: "mergecheck", |
| 1361 | G1T_API: "https://api.g1t.sh", |
| 1362 | MERGECHECK_PULL: job.pullId, |
| 1363 | MERGECHECK_TOKEN: job.token, |
| 1364 | G1T_USER: job.author.username, |
| 1365 | G1T_TOKEN: token, |
| 1366 | BASE_REMOTE: remote(job.repo), |
| 1367 | BASE_COMMIT: job.base, |
| 1368 | HEAD_REMOTE: remote(job.source), |
| 1369 | HEAD_BRANCH: job.branch, |
| 1370 | HEAD_COMMIT: job.head, |
| 1371 | }, |
| 1372 | }); |
| 1373 | } catch (error) { |
| 1374 | if (granted) await this.release(granted.held); |
| 1375 | await work.failMergecheck(job.pullId, job.token, error instanceof Error ? error.message : String(error)); |
| 1376 | } |
| 1377 | } |
| 1378 | |
| 1379 | /** |
| 1380 | * A refusal if `repo` takes no work from anyone: it is archived, so |
| 1381 | * read-only, or it was deleted (repos hides a deleted one, so it is not |
| 1382 | * found). Null when repos cannot answer now; the other checks still run. |
| 1383 | */ |
| 1384 | private async closedRepo(actor: User, repo: RepoPath): Promise<Result<never> | null> { |
| 1385 | const repos = reposClient(this.env.REPOS); |
| 1386 | const found = await repos.get(repo, actor).catch(() => null); |
| 1387 | if (!found) return null; |
| 1388 | if (!found.ok) { |
| 1389 | return found.error.code === "not_found" |
| 1390 | ? fail("not_found", `There is no repository at ${repo.namespace}/${repo.name}, or it was deleted.`) |
| 1391 | : null; |
| 1392 | } |
| 1393 | const status = await repos.statusById(found.value.id).catch(() => null); |
| 1394 | if (status?.deleted) { |
| 1395 | return fail("not_found", `${found.value.namespace}/${found.value.name} was deleted. An owner can restore it from the workspace's settings.`); |
| 1396 | } |
| 1397 | if (status?.archived || found.value.archivedAt) { |
| 1398 | return fail( |
| 1399 | "forbidden", |
| 1400 | `${found.value.namespace}/${found.value.name} is archived, so it is read-only. An owner can unarchive it in its settings.`, |
| 1401 | ); |
| 1402 | } |
| 1403 | return null; |
| 1404 | } |
| 1405 | |
| 1406 | /** |
| 1407 | * Stops every run in a repository that was archived or deleted: the |
| 1408 | * work service marks them stopped when it hears of it, and lists them |
| 1409 | * here (`runs_in_repo`, by id, so a deleted repository's runs are found |
| 1410 | * too), and each sandbox is destroyed. Never throws. |
| 1411 | */ |
| 1412 | private async stopRunsIn(repoId: string): Promise<void> { |
| 1413 | try { |
| 1414 | const response = await this.env.WORK.fetch("https://work/rpc/runs_in_repo", { |
| 1415 | method: "POST", |
| 1416 | headers: { "content-type": "application/json" }, |
| 1417 | body: JSON.stringify({ repoId }), |
| 1418 | }); |
| 1419 | if (!response.ok) return; |
| 1420 | const runs = (await response.json()) as { runId: string; sandbox: string | null }[]; |
| 1421 | for (const run of runs) { |
| 1422 | if (!run.sandbox) continue; |
| 1423 | try { |
| 1424 | await this.env.SANDBOX.get(this.env.SANDBOX.idFromString(run.sandbox)).halt("The repository was archived or deleted."); |
| 1425 | } catch (error) { |
| 1426 | // Already gone, or never started. |
| 1427 | console.log("sandbox not destroyed", run.runId, String(error)); |
| 1428 | } |
| 1429 | } |
| 1430 | } catch (error) { |
| 1431 | console.error("could not stop the runs in", repoId, error); |
| 1432 | } |
| 1433 | } |
| 1434 | |
| 1435 | private async defaultBranch(repo: RepoPath, viewer: Viewer): Promise<string> { |
| 1436 | const found = await reposClient(this.env.REPOS).get(repo, viewer); |
| 1437 | return found.ok ? found.value.defaultBranch : "main"; |
| 1438 | } |
| 1439 | } |