Repository navigation
Highlights
- Fix #186 (Long article timeout replay & billing protection):
- Persist per-
(entry, stage)retry state (EntryTaskState) with exponential backoff, auto-pause after max attempts or truncation, and estimated unbilled token tracking for timeouts. - Stop immediate in-round retries on client timeouts and
finish_reason=lengthtruncations; add Entry Tasks admin view with Pause / Resume / single-entry Retry Now actions. - Decouple
OpenAIAgent.max_chunk_tokens(default4000), per-agentrequest_timeout, andOPENAI_MAX_OUTPUT_TOKENS(default16384) from model context size; add HTML-safe DOM chunking (chunk_html) and persistent DB-backed chunk resume caching (TranslationChunkCache). - Document new environment variables (
OPENAI_MAX_OUTPUT_TOKENS,ENTRY_MAX_RETRY_ATTEMPTS,ENTRY_RETRY_BASE_DELAY_SECONDS,ENTRY_RETRY_MAX_DELAY_SECONDS,CHUNK_CACHE_TTL_SECONDS).
- Persist per-
- Admin & cache invalidation hardening: Invalidate rendered RSS/Tag/Digest caches on admin edits, filter rule changes, and deletions via
transaction.on_commit; reuse digest feeds in place ontarget_languagechanges. - Tag feed filters (#185): Add optional
apply_feed_filterson tags to apply per-feed filters when merging feeds. - Process management: Use
os.execvpin POSIX server launchers to prevent orphaned worker processes and forward CLI args indev_server.py.
What's Changed
- Add "Apply feed filters" option to tags by @thewebmasterp in #185
Full Changelog: 2026.11.17...2026.11.19