You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For the "retry wrappers for flaky downloads" candidate: tools-dataset already has this code, in the source toolkit of spec 004 (PR tools-dataset#7), ported from dataset-names.
datasettool.sources.cache.fetch_resource: cache at <cache>/<source_id>/<file> with .meta.json (url, time, sha256, size), User-Agent per project, up to three attempts with doubling backoff, no retry on 4xx except 408 and 429, atomic writes through .part, Content-Length check, BlockedError for an HTML page served instead of a file, manual-download hint on 403, offline and refresh modes. Tested against a local HTTP server only.
Placement decision: tools-dataset ADR 0007. It stays in tools-dataset because its consumers are the dataset repos. If tools-benchmark needs it, it can move here through a tools-commons spec, and tools-dataset would then depend on tools-commons.
Activity
matbrgz commentedon Oct 12, 2026
For the "retry wrappers for flaky downloads" candidate: tools-dataset already has this code, in the source toolkit of spec 004 (PR tools-dataset#7), ported from dataset-names.
datasettool.sources.cache.fetch_resource: cache at<cache>/<source_id>/<file>with.meta.json(url, time, sha256, size), User-Agent per project, up to three attempts with doubling backoff, no retry on 4xx except 408 and 429, atomic writes through.part, Content-Length check,BlockedErrorfor an HTML page served instead of a file, manual-download hint on 403, offline and refresh modes. Tested against a local HTTP server only.