fix(hopper): diagnose MCP provider failures with stage, state, and request context (#1044)

* fix(hopper): diagnose MCP provider failures with stage, state, and request context

* fix(hopper): retain health at the actual provider lifecycle boundary

* test(hopper): distinguish socket loss from later observed process exit

* test(hopper): cover observed exit separately from socket-first replies

---------

Co-authored-by: N0zoM1z0 <1705004398@qq.com>
This commit is contained in:
Alan Vo
2026-10-09 00:33:56 +08:00
committed by GitHub
co-authored by N0zoM1z0
parent b5c2954ad6
commit 5a67c7fc54
20 changed files with 990 additions and 62 deletions
@@ -14,6 +14,11 @@ import {
PROVIDER_REJECTION_CODES,
type ProviderRejectionCode,
} from "./providerSelection.js";
import {
PROVIDER_FAILURE_STAGES,
PROVIDER_OPERATION_STATES,
PROVIDER_RETRY_ACTIONS,
} from "../domain/providerOperationHealth.js";
import { analysisErrorProjectionSchema } from "./errorSchemas.js";
import { prefixedDigestSchema } from "./../domain/digests.js";
@@ -169,6 +174,22 @@ export const analysisActivity = z.object({
),
});
const providerOperationRequest = z.object({
request_id: z.number().int().min(1),
operation: z.string().min(1),
stage: z.enum(PROVIDER_FAILURE_STAGES),
});
export const providerOperationHealth = z
.object({
state: z.enum(PROVIDER_OPERATION_STATES),
stage: z.enum(PROVIDER_FAILURE_STAGES).nullable(),
retry_action: z.enum(PROVIDER_RETRY_ACTIONS).nullable(),
exit_code: z.number().int().nullable(),
requests: z.array(providerOperationRequest),
})
.nullable();
const providerRejectionCode: z.ZodType<ProviderRejectionCode> = z.enum(
PROVIDER_REJECTION_CODES,
);
@@ -315,6 +336,7 @@ export const sessionProvider = z
capabilities: z.array(providerCapability),
analysis_run: analysisRun,
analysis_activity: analysisActivity,
provider_operation_health: providerOperationHealth,
analysis_provider_binding: z
.object({
provider: providerIdentity,