Skip to content

feat(web): resume a run that failed with a provider error #188

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
yordis merged 2 commits into from
Oct 31, 2026
Merged

feat(web): resume a run that failed with a provider error #188

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5e826ef
feat(web): resume a run that failed with a provider error
yordis Oct 31, 2026
4e1f3b7
fix(server): keep workspace preparation failures on their own retry path
yordis Oct 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 3 additions & 6 deletions apps/server/src/orchestration-v2/Orchestrator.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@ import type {
import {
latestExecutedRun,
latestRootProviderFailure,
manuallyResumableRun,
runRanAfter,
usageLimitBlockedRun,
} from "@t3tools/shared/orchestrationV2ThreadError";
Expand Down Expand Up @@ -5240,14 +5241,10 @@ const makeOrchestrator = Effect.fn("orchestrationV2.Orchestrator.layer")(functio
) {
let projection = yield* getProjectionWithPendingEvents(command.threadId, events);
if (command.manualContinuationOfRunId !== undefined) {
const source = projection.runs.find((run) => run.id === command.manualContinuationOfRunId);
const limited = latestRootProviderFailure(source ?? null, projection.turnItems);
if (
command.dispatchMode.type !== "start_immediately" ||
source === undefined ||
(source.status !== "interrupted" &&
!(source.status === "failed" && limited?.class === "usage_limit")) ||
latestExecutedRun(projection.runs)?.id !== source.id ||
manuallyResumableRun(projection.runs, projection.turnItems)?.id !==
command.manualContinuationOfRunId ||
projection.thread.archivedAt !== null ||
projection.thread.deletedAt !== null ||
projection.runtimeRequests.some((request) => request.status === "pending")
Expand Down
308 changes: 166 additions & 142 deletions apps/server/src/orchestration-v2/runtimeLayer.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ import {
RunId,
ScheduledTaskId,
ThreadId,
ORCHESTRATION_V2_WORKSPACE_PREPARATION_FAILURE_CODE,
} from "@t3tools/contracts";
import * as Effect from "effect/Effect";
import * as DateTime from "effect/DateTime";
Expand Down Expand Up @@ -7841,162 +7842,185 @@ it.layer(layerSharedApplicationDataPlaneTest)("shared application data plane", (
});

it.layer(layerTest)("usage-limit recovery", (it) => {
it.effect.each(["interrupted", "usage_limit"] as const)(
"manually resumes an %s run ahead of its queued message only once",
(reason) =>
Effect.gen(function* () {
const orchestrator = yield* Orchestrator.OrchestratorV2;
const events = yield* EventSink.EventSinkV2;
const threadId = ThreadId.make(`manual-resume:${reason}`);
const projectId = ProjectId.make(`manual-resume:project:${reason}`);
const now = yield* DateTime.now;
const createdAt = DateTime.formatIso(now);
yield* seedProject({
projectId,
title: "Resume project",
workspaceRoot: process.cwd(),
defaultModelSelection: modelSelection,
createdAt,
});
yield* orchestrator.dispatch({
type: "thread.create",
commandId: CommandId.make(`manual-resume:create:${reason}`),
threadId,
projectId,
title: "Interrupted thread",
modelSelection,
runtimeMode: "full-access",
interactionMode: "default",
branch: null,
worktreePath: null,
createdBy: "user",
creationSource: "web",
});
yield* orchestrator.dispatch({
type: "message.dispatch",
commandId: CommandId.make(`manual-resume:start:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:start:${reason}`),
text: "Start work.",
attachments: [],
dispatchMode: { type: "defer_start" },
createdBy: "user",
creationSource: "web",
});
yield* orchestrator.dispatch({
type: "message.dispatch",
commandId: CommandId.make(`manual-resume:queue:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:queue:${reason}`),
text: "Follow up.",
attachments: [],
dispatchMode: { type: "queue_after_active" },
createdBy: "user",
creationSource: "web",
});
const [source, queued] = (yield* orchestrator.getThreadProjection(threadId)).runs as [
OrchestrationV2Run,
OrchestrationV2Run,
];
// Hold the queue like user Stop so the terminal worker cannot start it before explicit resume.
it.effect.each([
"interrupted",
"usage_limit",
"provider_error",
"workspace_preparation",
] as const)("guards a manual resume of an %s run ahead of its queued message", (reason) =>
Effect.gen(function* () {
const orchestrator = yield* Orchestrator.OrchestratorV2;
const events = yield* EventSink.EventSinkV2;
const threadId = ThreadId.make(`manual-resume:${reason}`);
const projectId = ProjectId.make(`manual-resume:project:${reason}`);
const now = yield* DateTime.now;
const createdAt = DateTime.formatIso(now);
yield* seedProject({
projectId,
title: "Resume project",
workspaceRoot: process.cwd(),
defaultModelSelection: modelSelection,
createdAt,
});
yield* orchestrator.dispatch({
type: "thread.create",
commandId: CommandId.make(`manual-resume:create:${reason}`),
threadId,
projectId,
title: "Interrupted thread",
modelSelection,
runtimeMode: "full-access",
interactionMode: "default",
branch: null,
worktreePath: null,
createdBy: "user",
creationSource: "web",
});
yield* orchestrator.dispatch({
type: "message.dispatch",
commandId: CommandId.make(`manual-resume:start:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:start:${reason}`),
text: "Start work.",
attachments: [],
dispatchMode: { type: "defer_start" },
createdBy: "user",
creationSource: "web",
});
yield* orchestrator.dispatch({
type: "message.dispatch",
commandId: CommandId.make(`manual-resume:queue:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:queue:${reason}`),
text: "Follow up.",
attachments: [],
dispatchMode: { type: "queue_after_active" },
createdBy: "user",
creationSource: "web",
});
const [source, queued] = (yield* orchestrator.getThreadProjection(threadId)).runs as [
OrchestrationV2Run,
OrchestrationV2Run,
];
// Hold the queue like user Stop so the terminal worker cannot start it before explicit resume.
yield* events.write({
events: [
{
id: EventId.make(`manual-resume:hold:${reason}`),
type: "run.updated",
threadId,
occurredAt: now,
payload: { ...queued, queueHeld: true },
},
{
id: EventId.make(`manual-resume:stop:${reason}`),
type: "run.updated",
threadId,
occurredAt: now,
payload: {
...source,
status: reason === "interrupted" ? "interrupted" : "failed",
completedAt: now,
},
},
],
});
let scheduledResume: ReturnType<typeof limitRecoveryCommand> = null;
const resetAt = DateTime.formatIso(DateTime.add(now, { minutes: 1 }));
if (reason !== "interrupted") {
yield* events.write({
events: [
{
id: EventId.make(`manual-resume:hold:${reason}`),
type: "run.updated",
threadId,
occurredAt: now,
payload: { ...queued, queueHeld: true },
},
{
id: EventId.make(`manual-resume:stop:${reason}`),
type: "run.updated",
id: EventId.make(`manual-resume:error:${reason}`),
type: "turn-item.updated",
threadId,
occurredAt: now,
payload: {
...source,
status: reason === "interrupted" ? "interrupted" : "failed",
id: TurnItemId.make(`manual-resume:error:${reason}`),
type: "error",
threadId,
runId: source.id,
nodeId: source.rootNodeId,
providerThreadId: null,
providerTurnId: null,
nativeItemRef: null,
parentItemId: null,
ordinal: 2,
status: "failed",
title: "Provider error",
startedAt: now,
completedAt: now,
updatedAt: now,
failure:
reason === "usage_limit"
? {
class: "usage_limit",
message: "Plan limit reached.",
code: "usageLimitExceeded",
retryable: null,
resetAt,
}
: {
class: "provider_error",
message: "Usage balance exhausted.",
code:
reason === "workspace_preparation"
? ORCHESTRATION_V2_WORKSPACE_PREPARATION_FAILURE_CODE
: null,
retryable: null,
},
},
},
],
});
let scheduledResume: ReturnType<typeof limitRecoveryCommand> = null;
if (reason === "usage_limit") {
const resetAt = DateTime.formatIso(DateTime.add(now, { minutes: 1 }));
yield* events.write({
events: [
{
id: EventId.make(`manual-resume:error:${reason}`),
type: "turn-item.updated",
threadId,
occurredAt: now,
payload: {
id: TurnItemId.make(`manual-resume:error:${reason}`),
type: "error",
threadId,
runId: source.id,
nodeId: source.rootNodeId,
providerThreadId: null,
providerTurnId: null,
nativeItemRef: null,
parentItemId: null,
ordinal: 2,
status: "failed",
title: "Usage limit reached",
startedAt: now,
completedAt: now,
updatedAt: now,
failure: {
class: "usage_limit",
message: "Plan limit reached.",
code: "usageLimitExceeded",
retryable: null,
resetAt,
},
},
},
],
});
const shell = (yield* orchestrator.getShellSnapshot()).threads.find(
(thread) => thread.id === threadId,
)!;
yield* orchestrator.dispatch(
limitRecoveryCommand(shell, true, DateTime.toEpochMillis(now))!,
);
const armed = (yield* orchestrator.getShellSnapshot()).threads.find(
(thread) => thread.id === threadId,
)!;
scheduledResume = limitRecoveryCommand(armed, true, Date.parse(resetAt));
assert.isNotNull(scheduledResume);
}
const resume = (suffix: string) => ({
type: "message.dispatch" as const,
commandId: CommandId.make(`manual-resume:${suffix}:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:${suffix}:${reason}`),
manualContinuationOfRunId: source.id,
text: "Continue where you left off.",
attachments: [],
dispatchMode: { type: "start_immediately" as const },
createdBy: "user" as const,
creationSource: "web" as const,
});
yield* orchestrator.dispatch(resume("first"));
const after = yield* orchestrator.getThreadProjection(threadId);
assert.lengthOf(after.runs, 3);
assert.equal(after.runs[1]?.status, "queued");
assert.equal(after.runs[2]?.status, "starting");
}
if (reason === "usage_limit") {
const shell = (yield* orchestrator.getShellSnapshot()).threads.find(
(thread) => thread.id === threadId,
)!;
yield* orchestrator.dispatch(
limitRecoveryCommand(shell, true, DateTime.toEpochMillis(now))!,
);
const armed = (yield* orchestrator.getShellSnapshot()).threads.find(
(thread) => thread.id === threadId,
)!;
scheduledResume = limitRecoveryCommand(armed, true, Date.parse(resetAt));
assert.isNotNull(scheduledResume);
}
const resume = (suffix: string) => ({
type: "message.dispatch" as const,
commandId: CommandId.make(`manual-resume:${suffix}:${reason}`),
threadId,
messageId: MessageId.make(`manual-resume:${suffix}:${reason}`),
manualContinuationOfRunId: source.id,
text: "Continue where you left off.",
attachments: [],
dispatchMode: { type: "start_immediately" as const },
createdBy: "user" as const,
creationSource: "web" as const,
});
if (reason === "workspace_preparation") {
assert.equal(
(yield* orchestrator.dispatch(resume("second")).pipe(Effect.exit))._tag,
(yield* orchestrator.dispatch(resume("first")).pipe(Effect.exit))._tag,
"Failure",
);
if (scheduledResume !== null) {
yield* TestClock.adjust("1 minute");
yield* orchestrator.dispatch(scheduledResume);
}
assert.lengthOf((yield* orchestrator.getThreadProjection(threadId)).runs, 3);
}),
return;
}
yield* orchestrator.dispatch(resume("first"));
const after = yield* orchestrator.getThreadProjection(threadId);
assert.lengthOf(after.runs, 3);
assert.equal(after.runs[1]?.status, "queued");
assert.equal(after.runs[2]?.status, "starting");
assert.equal(
(yield* orchestrator.dispatch(resume("second")).pipe(Effect.exit))._tag,
"Failure",
);
if (scheduledResume !== null) {
yield* TestClock.adjust("1 minute");
yield* orchestrator.dispatch(scheduledResume);
}
assert.lengthOf((yield* orchestrator.getThreadProjection(threadId)).runs, 3);
}),
);

it.effect.each([
Expand Down
19 changes: 5 additions & 14 deletions apps/web/src/components/ChatView.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,10 +42,7 @@ import { RightPanelSurfaceContent } from "./chat/RightPanelSurfaceContent";
import { usePendingThreadRequests } from "./chat/usePendingThreadRequests";
import { isScratchProject } from "@t3tools/client-runtime/state/projects";
import { visibleThreadPullRequests } from "@t3tools/shared/threadPullRequests";
import {
latestExecutedRun,
latestRootProviderFailure,
} from "@t3tools/shared/orchestrationV2ThreadError";
import { manuallyResumableRun } from "@t3tools/shared/orchestrationV2ThreadError";
import type { UsageLimitSourceSnapshots } from "@t3tools/contracts";
import { feedbackBannerItem } from "./chat/ComposerFeedback";
import { useUsageLimitsPanel } from "./chat/useUsageLimitsPanel";
Expand Down Expand Up @@ -814,16 +811,10 @@ export default function ChatView(props: ChatViewProps) {
isServerThread &&
serverProjection?.runs.some((run) => run.status === "queued" && run.queueHeld === true) ===
true;
const resumableRunId = (() => {
if (!isServerThread || serverProjection === null) return null;
const run = latestExecutedRun(serverProjection.runs);
if (run?.status === "interrupted") return run.id;
return run?.status === "failed" &&
serverRuntime?.lastErrorClass === "usage_limit" &&
latestRootProviderFailure(run, serverProjection.turnItems)?.class === "usage_limit"
? run.id
: null;
})();
const resumableRunId =
!isServerThread || serverProjection === null
? null
: (manuallyResumableRun(serverProjection.runs, serverProjection.turnItems)?.id ?? null);
const parentSubagentThreadId =
activeThread?.lineage.relationshipToParent === "subagent"
? activeThread.lineage.parentThreadId
Expand Down
Loading
Loading