|
| 1 | +import { ErrorCodes, errorShape } from "../../../packages/gateway-protocol/src/index.js"; |
| 2 | +import { |
| 3 | + buildAgentRunTerminalOutcome, |
| 4 | + type AgentRunTerminalOutcome, |
| 5 | +} from "../../agents/agent-run-terminal-outcome.js"; |
| 6 | +import { isAgentRunRestartAbortReason } from "../../agents/run-termination.js"; |
| 7 | +import { |
| 8 | + normalizeAgentRunTimeoutPhase, |
| 9 | + normalizeProviderStarted, |
| 10 | +} from "../../agents/run-timeout-attribution.js"; |
| 11 | +import { agentCommandFromIngress } from "../../commands/agent.js"; |
| 12 | +import { isAbortError } from "../../infra/abort-signal.js"; |
| 13 | +import { clearAgentRunContext } from "../../infra/agent-events.js"; |
| 14 | +import { readErrorName } from "../../infra/errors.js"; |
| 15 | +import { defaultRuntime } from "../../runtime.js"; |
| 16 | +import { createRunningTaskRun } from "../../tasks/detached-task-runtime.js"; |
| 17 | +import { normalizeDeliveryContext } from "../../utils/delivery-context.shared.js"; |
| 18 | +import type { ChatAbortControllerEntry } from "../chat-abort.js"; |
| 19 | +import { formatForLog } from "../ws-log.js"; |
| 20 | +import { setGatewayDedupeEntries } from "./agent-dedupe.js"; |
| 21 | +import { |
| 22 | + resolveFailedTrackedAgentTaskStatus, |
| 23 | + tryFinalizeTrackedAgentTask, |
| 24 | + type GatewayAgentTaskTrackingMode, |
| 25 | +} from "./agent-task-tracking.js"; |
| 26 | +import type { GatewayRequestContext, GatewayRequestHandlerOptions } from "./types.js"; |
| 27 | + |
| 28 | +function readAgentRunTimeoutAttribution(meta: unknown) { |
| 29 | + const record = |
| 30 | + meta && typeof meta === "object" && !Array.isArray(meta) |
| 31 | + ? (meta as Record<string, unknown>) |
| 32 | + : undefined; |
| 33 | + return { |
| 34 | + timeoutPhase: normalizeAgentRunTimeoutPhase(record?.timeoutPhase), |
| 35 | + providerStarted: normalizeProviderStarted(record?.providerStarted), |
| 36 | + }; |
| 37 | +} |
| 38 | + |
| 39 | +function isGatewayAbortSignalReason(reason: unknown): boolean { |
| 40 | + return reason === undefined || isAbortError(reason) || readErrorName(reason) === "TimeoutError"; |
| 41 | +} |
| 42 | + |
| 43 | +function isGatewayAgentAbortRejection(error: unknown, signal: AbortSignal): boolean { |
| 44 | + if (!signal.aborted) { |
| 45 | + return false; |
| 46 | + } |
| 47 | + if (isAgentRunRestartAbortReason(signal.reason)) { |
| 48 | + return true; |
| 49 | + } |
| 50 | + if (readErrorName(signal.reason) === "TimeoutError") { |
| 51 | + return true; |
| 52 | + } |
| 53 | + if (!isGatewayAbortSignalReason(signal.reason)) { |
| 54 | + return false; |
| 55 | + } |
| 56 | + return isAbortError(error) || readErrorName(error) === "TimeoutError"; |
| 57 | +} |
| 58 | + |
| 59 | +function resolveGatewayAgentAbortStopReason(signal: AbortSignal): "restart" | "rpc" | "timeout" { |
| 60 | + if (isAgentRunRestartAbortReason(signal.reason)) { |
| 61 | + return "restart"; |
| 62 | + } |
| 63 | + return readErrorName(signal.reason) === "TimeoutError" ? "timeout" : "rpc"; |
| 64 | +} |
| 65 | + |
| 66 | +export function resolveAbortedAgentStopReason(entry?: ChatAbortControllerEntry): string { |
| 67 | + return entry?.abortStopReason?.trim() || "rpc"; |
| 68 | +} |
| 69 | + |
| 70 | +export function deleteGatewayDedupeEntries(params: { |
| 71 | + dedupe: GatewayRequestContext["dedupe"]; |
| 72 | + keys: readonly string[]; |
| 73 | +}) { |
| 74 | + for (const key of params.keys) { |
| 75 | + params.dedupe.delete(key); |
| 76 | + } |
| 77 | +} |
| 78 | + |
| 79 | +export function dispatchAgentRunFromGateway(params: { |
| 80 | + ingressOpts: Parameters<typeof agentCommandFromIngress>[0]; |
| 81 | + runId: string; |
| 82 | + dedupeKeys: readonly string[]; |
| 83 | + /** |
| 84 | + * Controller whose signal is wired into `ingressOpts.abortSignal`. Used on |
| 85 | + * completion to drop the matching `chatAbortControllers` entry without |
| 86 | + * touching a same-runId entry owned by a concurrent chat.send. |
| 87 | + */ |
| 88 | + abortController: AbortController; |
| 89 | + cleanupAbortController: () => void; |
| 90 | + respond: GatewayRequestHandlerOptions["respond"]; |
| 91 | + context: GatewayRequestHandlerOptions["context"]; |
| 92 | + taskTrackingMode: Exclude<GatewayAgentTaskTrackingMode, "plugin_subagent">; |
| 93 | + onSettled?: (outcome: { |
| 94 | + terminalOutcome: AgentRunTerminalOutcome; |
| 95 | + onRecovered?: () => void; |
| 96 | + }) => Promise<boolean> | boolean; |
| 97 | +}) { |
| 98 | + const shouldTrackTask = params.taskTrackingMode === "cli"; |
| 99 | + let taskTracked = false; |
| 100 | + if (shouldTrackTask) { |
| 101 | + try { |
| 102 | + taskTracked = Boolean( |
| 103 | + createRunningTaskRun({ |
| 104 | + runtime: "cli", |
| 105 | + sourceId: params.runId, |
| 106 | + ownerKey: params.ingressOpts.sessionKey, |
| 107 | + scopeKind: "session", |
| 108 | + requesterOrigin: normalizeDeliveryContext({ |
| 109 | + channel: params.ingressOpts.channel, |
| 110 | + to: params.ingressOpts.to, |
| 111 | + accountId: params.ingressOpts.accountId, |
| 112 | + threadId: params.ingressOpts.threadId, |
| 113 | + }), |
| 114 | + childSessionKey: params.ingressOpts.sessionKey, |
| 115 | + runId: params.runId, |
| 116 | + task: params.ingressOpts.message, |
| 117 | + deliveryStatus: "not_applicable", |
| 118 | + startedAt: Date.now(), |
| 119 | + }), |
| 120 | + ); |
| 121 | + } catch (err) { |
| 122 | + // Best-effort only: background task tracking must not block agent runs. |
| 123 | + // Still surface the swallowed error so non-transient tracking failures stay observable. |
| 124 | + params.context.logGateway.warn( |
| 125 | + `failed to start tracked agent task ${params.runId}: ${formatForLog(err)}`, |
| 126 | + ); |
| 127 | + } |
| 128 | + } |
| 129 | + const settle = async (outcome: { |
| 130 | + terminalOutcome: AgentRunTerminalOutcome; |
| 131 | + onRecovered?: () => void; |
| 132 | + }): Promise<boolean> => { |
| 133 | + try { |
| 134 | + return (await params.onSettled?.(outcome)) ?? true; |
| 135 | + } catch (error) { |
| 136 | + params.context.logGateway.warn( |
| 137 | + `failed to settle agent continuation ${params.runId}: ${formatForLog(error)}`, |
| 138 | + ); |
| 139 | + return false; |
| 140 | + } |
| 141 | + }; |
| 142 | + void agentCommandFromIngress(params.ingressOpts, defaultRuntime, params.context.deps) |
| 143 | + .then(async (result) => { |
| 144 | + const aborted = result?.meta?.aborted === true; |
| 145 | + const timeoutAttribution = readAgentRunTimeoutAttribution(result?.meta); |
| 146 | + if (taskTracked) { |
| 147 | + tryFinalizeTrackedAgentTask({ |
| 148 | + runId: params.runId, |
| 149 | + status: aborted ? "timed_out" : "succeeded", |
| 150 | + terminalSummary: aborted ? "aborted" : "completed", |
| 151 | + log: params.context.logGateway, |
| 152 | + }); |
| 153 | + } |
| 154 | + const payload = { |
| 155 | + runId: params.runId, |
| 156 | + status: aborted ? ("timeout" as const) : ("ok" as const), |
| 157 | + summary: aborted ? "aborted" : "completed", |
| 158 | + ...(aborted ? { stopReason: result?.meta?.stopReason ?? "rpc" } : {}), |
| 159 | + ...(aborted && timeoutAttribution.timeoutPhase |
| 160 | + ? { timeoutPhase: timeoutAttribution.timeoutPhase } |
| 161 | + : {}), |
| 162 | + ...(aborted && timeoutAttribution.providerStarted !== undefined |
| 163 | + ? { providerStarted: timeoutAttribution.providerStarted } |
| 164 | + : {}), |
| 165 | + result, |
| 166 | + }; |
| 167 | + const terminalOutcome = buildAgentRunTerminalOutcome({ |
| 168 | + status: |
| 169 | + aborted || result?.meta?.stopReason === "timeout" || timeoutAttribution.timeoutPhase |
| 170 | + ? "timeout" |
| 171 | + : result?.meta?.error || result?.meta?.stopReason === "error" |
| 172 | + ? "error" |
| 173 | + : "ok", |
| 174 | + error: result?.meta?.error, |
| 175 | + stopReason: result?.meta?.stopReason, |
| 176 | + livenessState: result?.meta?.livenessState, |
| 177 | + timeoutPhase: timeoutAttribution.timeoutPhase, |
| 178 | + providerStarted: timeoutAttribution.providerStarted, |
| 179 | + }); |
| 180 | + const persistTerminalDedupe = () => { |
| 181 | + setGatewayDedupeEntries({ |
| 182 | + dedupe: params.context.dedupe, |
| 183 | + keys: params.dedupeKeys, |
| 184 | + entry: { |
| 185 | + ts: Date.now(), |
| 186 | + ok: true, |
| 187 | + payload, |
| 188 | + }, |
| 189 | + }); |
| 190 | + }; |
| 191 | + const settled = await settle({ terminalOutcome, onRecovered: persistTerminalDedupe }); |
| 192 | + if (!settled) { |
| 193 | + const summary = "failed to persist cron continuation settlement"; |
| 194 | + const error = errorShape(ErrorCodes.UNAVAILABLE, summary); |
| 195 | + const failedPayload = { runId: params.runId, status: "error" as const, summary }; |
| 196 | + setGatewayDedupeEntries({ |
| 197 | + dedupe: params.context.dedupe, |
| 198 | + keys: params.dedupeKeys, |
| 199 | + entry: { ts: Date.now(), ok: false, payload: failedPayload, error }, |
| 200 | + }); |
| 201 | + params.respond(false, failedPayload, error, { runId: params.runId, error: summary }); |
| 202 | + return; |
| 203 | + } |
| 204 | + persistTerminalDedupe(); |
| 205 | + // Send a second res frame (same id) so TS clients with expectFinal can wait. |
| 206 | + // Swift clients will typically treat the first res as the result and ignore this. |
| 207 | + params.respond(true, payload, undefined, { runId: params.runId }); |
| 208 | + }) |
| 209 | + .catch(async (err: unknown) => { |
| 210 | + const aborted = isGatewayAgentAbortRejection(err, params.abortController.signal); |
| 211 | + const renderedErr = formatForLog(err); |
| 212 | + if (taskTracked) { |
| 213 | + tryFinalizeTrackedAgentTask({ |
| 214 | + runId: params.runId, |
| 215 | + status: aborted ? "timed_out" : resolveFailedTrackedAgentTaskStatus(err), |
| 216 | + error: renderedErr, |
| 217 | + terminalSummary: renderedErr, |
| 218 | + log: params.context.logGateway, |
| 219 | + }); |
| 220 | + } |
| 221 | + const error = errorShape(ErrorCodes.UNAVAILABLE, renderedErr); |
| 222 | + const stopReason = resolveGatewayAgentAbortStopReason(params.abortController.signal); |
| 223 | + const terminalOutcome = buildAgentRunTerminalOutcome({ |
| 224 | + status: aborted ? "timeout" : "error", |
| 225 | + error: renderedErr, |
| 226 | + stopReason, |
| 227 | + timeoutPhase: aborted ? "gateway_draining" : undefined, |
| 228 | + }); |
| 229 | + const payload = { |
| 230 | + runId: params.runId, |
| 231 | + status: aborted ? ("timeout" as const) : ("error" as const), |
| 232 | + summary: aborted ? "aborted" : renderedErr, |
| 233 | + ...(aborted ? { stopReason, timeoutPhase: "gateway_draining" as const } : {}), |
| 234 | + }; |
| 235 | + const persistTerminalDedupe = (settlementPersisted: boolean) => { |
| 236 | + setGatewayDedupeEntries({ |
| 237 | + dedupe: params.context.dedupe, |
| 238 | + keys: params.dedupeKeys, |
| 239 | + entry: { |
| 240 | + ts: Date.now(), |
| 241 | + ok: aborted && settlementPersisted, |
| 242 | + payload, |
| 243 | + ...(aborted ? {} : { error }), |
| 244 | + }, |
| 245 | + }); |
| 246 | + }; |
| 247 | + const settled = await settle({ |
| 248 | + terminalOutcome, |
| 249 | + onRecovered: () => persistTerminalDedupe(true), |
| 250 | + }); |
| 251 | + persistTerminalDedupe(settled); |
| 252 | + params.respond(aborted && settled, payload, aborted && settled ? undefined : error, { |
| 253 | + runId: params.runId, |
| 254 | + ...(aborted ? {} : { error: formatForLog(err) }), |
| 255 | + }); |
| 256 | + }) |
| 257 | + .finally(() => { |
| 258 | + clearAgentRunContext(params.runId, params.ingressOpts.lifecycleGeneration); |
| 259 | + params.cleanupAbortController(); |
| 260 | + }); |
| 261 | +} |
0 commit comments