File
Blob: src/worker/durable/project-do/reconciliation/sandbox-cleanup.ts
| 1 | import { asc, eq } from "drizzle-orm"; |
| 2 | |
| 3 | import { type ProjectId, RunId } from "@/contracts"; |
| 4 | import { expectTrusted } from "@/worker/contracts"; |
| 5 | import * as projectSchema from "@/worker/db/durable/schema/project-do"; |
| 6 | import { isNoContainerInstanceError } from "@/worker/sandbox/container-errors"; |
| 7 | |
| 8 | import { SANDBOX_CLEANUP_RETRY_DELAYS_MS } from "../constants"; |
| 9 | import { getSandboxCleanupRetryState, rescheduleAlarmInTransaction, sandboxCleanupRetryKey } from "../sidecar-state"; |
| 10 | import type { ProjectDoContext, SandboxCleanupRetryState } from "../types"; |
| 11 | |
| 12 | const buildRetryState = (attempt: number): SandboxCleanupRetryState => ({ |
| 13 | attempt, |
| 14 | nextAt: |
| 15 | Date.now() + SANDBOX_CLEANUP_RETRY_DELAYS_MS[Math.min(attempt - 1, SANDBOX_CLEANUP_RETRY_DELAYS_MS.length - 1)], |
| 16 | }); |
| 17 | |
| 18 | export const persistSandboxCleanupRetryState = async ( |
| 19 | context: ProjectDoContext, |
| 20 | projectId: ProjectId, |
| 21 | runId: RunId, |
| 22 | value: SandboxCleanupRetryState | null, |
| 23 | ): Promise<void> => { |
| 24 | await context.ctx.storage.transaction(async (txn) => { |
| 25 | if (value === null) { |
| 26 | await txn.delete(sandboxCleanupRetryKey(runId)); |
| 27 | } else { |
| 28 | await txn.put(sandboxCleanupRetryKey(runId), value); |
| 29 | } |
| 30 | |
| 31 | await rescheduleAlarmInTransaction(context, txn, projectId); |
| 32 | }); |
| 33 | }; |
| 34 | |
| 35 | export const seedSandboxCleanupRetry = async ( |
| 36 | context: ProjectDoContext, |
| 37 | projectId: ProjectId, |
| 38 | runId: RunId, |
| 39 | ): Promise<void> => { |
| 40 | await persistSandboxCleanupRetryState(context, projectId, runId, buildRetryState(1)); |
| 41 | }; |
| 42 | |
| 43 | export const clearSandboxCleanupRetry = async ( |
| 44 | context: ProjectDoContext, |
| 45 | projectId: ProjectId, |
| 46 | runId: RunId, |
| 47 | ): Promise<void> => { |
| 48 | await persistSandboxCleanupRetryState(context, projectId, runId, null); |
| 49 | }; |
| 50 | |
| 51 | export const attemptSandboxCleanup = async ( |
| 52 | context: ProjectDoContext, |
| 53 | projectId: ProjectId, |
| 54 | runId: RunId, |
| 55 | trigger: "alarm" | "watchdog", |
| 56 | ): Promise<boolean> => { |
| 57 | const sandboxStub = context.env.Sandbox.getByName(runId); |
| 58 | |
| 59 | try { |
| 60 | await sandboxStub.setKeepAlive(false); |
| 61 | } catch (error) { |
| 62 | context.logger.warn("sandbox_keep_alive_release_failed", { |
| 63 | projectId, |
| 64 | runId, |
| 65 | trigger, |
| 66 | error: error instanceof Error ? error.message : String(error), |
| 67 | }); |
| 68 | } |
| 69 | |
| 70 | try { |
| 71 | await sandboxStub.destroy(); |
| 72 | return true; |
| 73 | } catch (error) { |
| 74 | if (isNoContainerInstanceError(error)) { |
| 75 | return true; |
| 76 | } |
| 77 | |
| 78 | context.logger.warn("sandbox_destroy_failed", { |
| 79 | projectId, |
| 80 | runId, |
| 81 | trigger, |
| 82 | error: error instanceof Error ? error.message : String(error), |
| 83 | }); |
| 84 | return false; |
| 85 | } |
| 86 | }; |
| 87 | |
| 88 | export const reconcileSandboxCleanup = async ( |
| 89 | context: ProjectDoContext, |
| 90 | projectId: ProjectId, |
| 91 | ): Promise<RunId | null> => { |
| 92 | const rows = await context.db |
| 93 | .select({ |
| 94 | runId: projectSchema.projectRuns.runId, |
| 95 | }) |
| 96 | .from(projectSchema.projectRuns) |
| 97 | .where(eq(projectSchema.projectRuns.projectId, projectId)) |
| 98 | .orderBy(asc(projectSchema.projectRuns.createdAt), asc(projectSchema.projectRuns.runId)); |
| 99 | |
| 100 | let candidateRunId: RunId | null = null; |
| 101 | let retryState: SandboxCleanupRetryState | null = null; |
| 102 | |
| 103 | for (const row of rows) { |
| 104 | const runId = expectTrusted(RunId, row.runId, "RunId"); |
| 105 | const currentRetryState = await getSandboxCleanupRetryState(context, runId); |
| 106 | if (!currentRetryState || currentRetryState.nextAt > Date.now()) { |
| 107 | continue; |
| 108 | } |
| 109 | |
| 110 | candidateRunId = runId; |
| 111 | retryState = currentRetryState; |
| 112 | break; |
| 113 | } |
| 114 | |
| 115 | if (!candidateRunId || !retryState) { |
| 116 | return null; |
| 117 | } |
| 118 | |
| 119 | if (await attemptSandboxCleanup(context, projectId, candidateRunId, "alarm")) { |
| 120 | await clearSandboxCleanupRetry(context, projectId, candidateRunId); |
| 121 | return candidateRunId; |
| 122 | } |
| 123 | |
| 124 | const nextRetryState = buildRetryState(retryState.attempt + 1); |
| 125 | await persistSandboxCleanupRetryState(context, projectId, candidateRunId, nextRetryState); |
| 126 | context.logger.warn("sandbox_cleanup_retry_scheduled", { |
| 127 | projectId, |
| 128 | runId: candidateRunId, |
| 129 | attempt: nextRetryState.attempt, |
| 130 | nextAt: nextRetryState.nextAt, |
| 131 | }); |
| 132 | return null; |
| 133 | }; |