# Changelog of Google Autocomplete Scraper — Bulk (`devilscrapes/google-autocomplete-bulk`) Actor

- **URL**: https://apify.com/devilscrapes/google-autocomplete-bulk/changelog.md
- **Full Actor documentation**: https://apify.com/devilscrapes/google-autocomplete-bulk.md

## Changelog — google-autocomplete-bulk

### 0.0.2 — 2026-08-11

- **Fix (reliability)**: 30-day public success rate was ~85-94% (customer
  runs only — our own QA runs never failed). Reproduced the root cause: a
  single transient `curl_cffi` network error (connection reset, DNS
  failure, TLS handshake, timeout, or a malformed/non-JSON body from a
  blocked request) raised straight out of `session.get()`/`response.json()`
  — it was never caught by the retry loop (which only inspected an
  already-returned `response.status_code`) and was not a `RuntimeError`, so
  it bypassed `main.py`'s per-tuple guard and crashed the *entire* run via
  `asyncio.gather(..., return_exceptions=False)`. One bad seed/locale combo
  out of potentially hundreds took down the whole bulk run.
- `src/client.py`: `_get_with_retry` now wraps `session.get()` and
  `response.json()` in `try/except RequestException` and retries with the
  same exponential backoff as HTTP 429/503 — this is exactly what the
  module's own docstring already promised ("network errors") but never
  implemented.
- `src/main.py`: `_process_tuple` now isolates the parse+push phase too
  (previously unguarded) as a defense-in-depth backstop, so any per-tuple
  failure is logged and skipped rather than propagating. Cancellation
  (`BaseException`) still propagates normally.
- No change to input schema, output row shape, or PPE pricing/charging —
  verified by local `apify run` against `tests/fixtures/input.qa.json`
  (still 10 rows, identical field set) and against an adversarial bulk
  input (unicode/emoji seeds, empty seeds, multiple locales) — 78 rows
  across 16 tuples, exit 0, no crash.
- 7 new regression tests (51 → 58) pinning: retry-on-network-error,
  retry-on-malformed-JSON-body, exhausted-retries-still-raises-RuntimeError,
  and `_scrape` surviving a fetch/parse failure on one tuple among many.

### 0.0.1 — 2026-05-16

- Initial release.
- Bulk Google autocomplete scraper. Seeds × languages × countries
  fan-out per run, optional A–Z alphabet expansion.
- `client=chrome` (default, up to 15 suggestions + relevance scores)
  or `client=firefox` (up to 10 suggestions, lean payload).
- PPE pricing: `actor-start` $0.05, `result-row` $0.001.
- Pydantic v2 input + dataset row validation.
- curl-cffi `chrome131` impersonation; concurrency capped at 2;
  exponential backoff with `Retry-After` honoured.