fix(credential_pool): auto-detect Z.AI endpoint via probe and cache#5575
Closed
SeeYangZhi wants to merge 1 commit into
Closed
fix(credential_pool): auto-detect Z.AI endpoint via probe and cache#5575SeeYangZhi wants to merge 1 commit into
SeeYangZhi wants to merge 1 commit into
Conversation
The credential pool seeder and runtime credential resolver hardcoded api.z.ai/api/paas/v4 for all Z.AI keys. Keys on the Coding Plan (or CN endpoint) would hit the wrong endpoint, causing 401/429 errors on the first request even though a working endpoint exists. Add _resolve_zai_base_url() that: - Respects GLM_BASE_URL env var (no probe when explicitly set) - Probes all candidate endpoints (global, cn, coding-global, coding-cn) via detect_zai_endpoint() to find one that returns HTTP 200 - Caches the detected endpoint in provider state (auth.json) keyed on a SHA-256 hash of the API key so subsequent starts skip the probe - Falls back to the default URL if all probes fail Wire into both _seed_from_env() in the credential pool and resolve_api_key_provider_credentials() in the runtime resolver, matching the pattern from the kimi-coding fix (PR NousResearch#5566). Fixes the same class of bug as NousResearch#5561 but for the zai provider.
Contributor
|
Merged via PR #5763. Your commit was cherry-picked with authorship preserved. Thanks @SeeYangZhi! |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The credential pool seeder and runtime credential resolver hardcoded
api.z.ai/api/paas/v4for all Z.AI keys. Keys on the Coding Plan (or CN endpoint) would hit the wrong endpoint, causing HTTP 429/401 errors on the first request even though a working endpoint exists.Root Cause
Same class of bug as #5561 (kimi-coding).
_seed_from_env()inagent/credential_pool.pyuses:For
zai,pconfig.inference_base_urlishttps://api.z.ai/api/paas/v4, so the pool and resolver always use that endpoint regardless of which plan the key belongs to. Meanwhiledetect_zai_endpoint()exists inhermes_cli/auth.pyto probe candidate endpoints, but was never wired up.Fix
Add
_resolve_zai_base_url()that:GLM_BASE_URLenv var (no probe when explicitly set)auth.jsonfor a previously detected endpoint (keyed on SHA-256 hash of the API key)detect_zai_endpoint()to probe all 4 candidates (global, cn, coding-global, coding-cn) if no cache hitWire into both
_seed_from_env()in the credential pool andresolve_api_key_provider_credentials()in the runtime resolver, matching the pattern from the kimi-coding fix (#5566).Changes
hermes_cli/auth.py— Add_resolve_zai_base_url()+ wire intoresolve_api_key_provider_credentials()agent/credential_pool.py— Import and call_resolve_zai_base_urlin_seed_from_env()for zaitests/test_api_key_providers.py— AddTestZaiEndpointAutoDetect(4 tests), mock probe in existing zai tests