fix: surface Spotify quota exhaustion
Some checks are pending
ci / backend (push) Waiting to run
ci / frontend (push) Waiting to run

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Justin Visser 2026-08-11 22:07:06 +02:00
parent 358d058392
commit bfea9efb5e
6 changed files with 97 additions and 10 deletions

View file

@ -38,6 +38,8 @@ from app.ports.protocols import (
RERANK_FALLBACK_CODE = "rerank_fallback"
RERANK_FALLBACK_MESSAGE = "Ranking output was invalid, so grounded results are shown instead."
RERANK_FALLBACK_JUSTIFICATION = "Selected as a grounded match for your request."
QUOTA_ERROR_CODE = "quota_exhausted"
QUOTA_ERROR_MESSAGE = "Spotify's Development Mode quota is temporarily exhausted."
class _TrackSelection:
@ -140,10 +142,8 @@ class RecommendationPipeline:
return
except CatalogQuotaExhaustedError:
yield PipelineErrorEvent(
code="quota_exhausted",
message=(
"Spotify request quota was exhausted before recommendations could be prepared."
),
code=QUOTA_ERROR_CODE,
message=QUOTA_ERROR_MESSAGE,
)
return
except SpotifyError:
@ -158,9 +158,16 @@ class RecommendationPipeline:
candidate_count=len(intent.candidates),
)
try:
pool = await self._grounded_pool(
session_id, catalog, intent, taste, deadline_at, seeded_pool
)
except CatalogQuotaExhaustedError:
yield PipelineErrorEvent(
code=QUOTA_ERROR_CODE,
message=QUOTA_ERROR_MESSAGE,
)
return
if not pool:
yield PipelineErrorEvent(
code="no_grounded_results",
@ -209,6 +216,8 @@ class RecommendationPipeline:
self.settings.rerank_count + self.settings.rerank_pool_buffer,
deadline_at,
)
if not result.tracks and result.metrics.did_exhaust_quota:
raise CatalogQuotaExhaustedError
if result.tracks:
self.last_pools[session_id] = result.tracks
return result.tracks

View file

@ -146,6 +146,30 @@ def test_intent_stage_failures_yield_one_typed_error_event() -> None:
asyncio.run(run())
def test_grounding_quota_yields_typed_error_instead_of_empty_results() -> None:
class QuotaCatalog(FakeCatalog):
"""Exhaust the Spotify quota on the first grounding search."""
async def search_tracks(self, query: str, limit: int = 10) -> list[Track]:
self.search_call_count += 1
raise SpotifyQuotaExhaustedError(0.0, "QUOTA_EXCEEDED")
async def run() -> None:
track = _track("candidate", "Candidate Song")
catalog = QuotaCatalog(())
recommender = FakeRecommender([_intent(track)])
events = await _run_pipeline(catalog, recommender)
assert [event.type for event in events] == ["metadata", "error"]
assert isinstance(events[-1], PipelineErrorEvent)
assert events[-1].code == "quota_exhausted"
assert catalog.search_call_count == 1
assert recommender.rerank_call_count == 0
asyncio.run(run())
def test_taste_failure_cancels_sibling_fetches() -> None:
class FailingTasteCatalog(FakeCatalog):
"""Fail one taste request after all sibling requests have started."""

View file

@ -514,3 +514,27 @@ Waarom:
werkende versie 1 klik verderop hebben. Een runtime mode-switch in de
app zelf zou een derde codepad zijn; twee instanties met elk 1 mode
houden het ontwerp schoon.
### Final polishing round
Wat ik deed:
- De laatste live-test gaf op de Spotify Search endpoint opnieuw
`QUOTA_EXCEEDED`, terwijl auth en de taste-profile calls wel werkten. De
pipeline toont dit nu als een `quota_exhausted` error in plaats van als nul
resultaten; de retry button verdwijnt en de demo-link blijft staan.
- De eerdere aanname van een dagelijks resetmoment gecorrigeerd. Spotify deelt
Development Mode quota per developer-account en endpoint bucket, maar
publiceert de limieten en het reset window niet.
Waarom:
- Een quota error is geen geldige zoekopdracht met nul resultaten. Door deze
rate-limit issues lever ik op 12 augustus in: eerst één gecontroleerde
live-test, daarna direct de hand-in.
Wat ik heb laten vallen of uitgesteld:
- De structurele oplossing is een Spotify key met hogere quota. Dat is voor
deze PoC-assessment niet proportioneel; ik bouw geen slimme workaround en
live mode valt niet stil terug op fixtures.

View file

@ -17,8 +17,8 @@ below. No cloud dev environment involved.
the repo you are reading is the mirror and shows the same history and
the same green CI.
- CI runs on every push: ruff, mypy and pytest for the backend; typecheck,
lint, build and vitest for the frontend. It has been green since the
first commit.
lint, build and vitest for the frontend. One intermediate batch exposed
a test typing error; the next commit fixed it, and current main is green.
## Deployment

View file

@ -8,7 +8,9 @@ defineEmits<{ retry: [] }>()
<template>
<div class="error" role="alert" :data-error-code="code">
<p>{{ message }}</p>
<button type="button" @click="$emit('retry')">{{ messages.streamErrorRetry }}</button>
<button v-if="code !== 'quota_exhausted'" type="button" @click="$emit('retry')">
{{ messages.streamErrorRetry }}
</button>
<a v-if="demoUrl" class="demo-link" :href="demoUrl">
{{ messages.streamErrorDemoInstance }}
</a>

View file

@ -166,6 +166,34 @@ describe('result presentation', () => {
expect(onPick).toHaveBeenCalledWith('Focus')
})
it('renders a quota error without retry and keeps the demo escape hatch', () => {
const turn: AssistantTurn = {
...doneTurn(),
status: 'error',
tracks: [],
warnings: [],
error: {
type: 'error',
code: 'quota_exhausted',
message: "Spotify's Development Mode quota is temporarily exhausted.",
},
completion: null,
}
const demoUrl = 'https://demo.example.com'
const root = mountComponent(AssistantMessage, {
turn,
canSave: false,
demoUrl,
})
const alert = root.querySelector<HTMLElement>('[role="alert"]')
const demoLink = alert?.querySelector<HTMLAnchorElement>('a')
expect(alert?.dataset.errorCode).toBe('quota_exhausted')
expect(alert?.querySelector('button')).toBeNull()
expect(demoLink?.href).toBe(`${demoUrl}/`)
expect(demoLink?.textContent).toContain('Open the demo instance')
})
it('renders a first-stage stream error with its code and message', () => {
const turn: AssistantTurn = {
...doneTurn(),