Page sync retries transient fetch failures with backoff

PageSource.fetch in Agno, which loads every page and nested index during Knowledge.sync_pages, now abandons a stalled connection after 5 seconds and retries it on a fresh one. Syncing docs.agno.com used to fail one or two random pages per run with Connection reset by peer. A hung TLS handshake ate the whole 30 second fetch deadline, so no retry could run.

The 30 second deadline still caps each fetch, and a retry that would wait past it fails right away with the transport error as its cause. On the deployment that hit the bug, the stalled handshake timed out at 5 seconds, the second attempt succeeded, and the sync finished with 0 failed pages. Class attributes on PageSource, such as fetch_attempts and attempt_timeout_seconds, tune the policy.

See the cookbook, and learn more about syncing published pages in the documentation.