mateo-berri 3bbd2fac33 fix(batches): fill a managed batch page past rows that will not parse ... The managed batch listing fetched one page of rows, derived has_more from that raw fetch, then dropped every row whose stored blob would not parse. last_id came from the survivors, so a page of corrupt or legacy rows came back as data [], last_id null, has_more true, and a client following last_id could not advance. The OpenAI SDK's auto-paginator, which cursors off the last item in data, stopped silently and returned a truncated list. Read chunks until page_size + 1 batches survive parsing and file-id resolution or the caller's rows run out, the way the managed file listing already does, so a page carries data and a usable cursor while parseable rows remain and has_more only says true when another one exists. The first chunk keeps the old page_size + 1 size so a healthy page still costs one query; a scan that has to continue widens to the file listing's continuation chunk and stops resolving rows once the page is full.		2026-08-28 07:54:18 -07:00
..
litellm_enterprise	fix(batches): fill a managed batch page past rows that will not parse	2026-08-28 07:54:18 -07:00
conftest.py	test: drop the cwd-relative sys.path.insert calls from the test suite (#37802)	2026-08-22 09:25:58 -07:00