Azure / Azure/azure-sdk-for-python
[Cosmos] feed_range query continuation token replays documents from non-cursor PK ranges
- Vorherrschende Sprache
- Python
- Sterne
- 5.6k
- Forks
- 3.4k
- Ø Merge
- 2 T.
- Gemergte PRs (30 T.)
- 217
Beschreibung
### Summary
When `query_items(feed_range=..., max_item_count=N)` is invoked with a `feed_range` that overlaps multiple physical partition key (PK) ranges (for example, after a server-side split), `__QueryFeed` issues one POST per overlapping inner range and merges the partial results.
PR #46469 fixed the page-size bug (the merged page is now truncated to `max_item_count`). However, the **continuation token surfaced to the caller** is still only the last inner range's `x-ms-continuation`. Round-tripping that token on the next page sends it to every inner range — which is undefined server-side and would produce duplicates, missing documents, or non-terminating iteration.
As a safety mitigation in PR #46469, when the merged page is actually truncated the continuation header is **suppressed** so callers observe the truncated page as terminal rather than experiencing data loss. This makes the broken pagination case visible (pagination ends early) instead of silently corrupting results, but it also means **only the first page of a multi-range `feed_range` query is reliable**.
### Proposed fix
Implement a **composite continuation token** that spans all overlapping inner PK ranges for a `feed_range` query:
- Track per-inner-range continuation tokens and remaining-cap state.
- Encode them into a single opaque continuation string surfaced to the caller.
- On resume, decode and route each per-range token to its corresponding inner POST.
- Stop issuing inner POSTs once the user-requested `max_item_count` cap is reached for the current page; carry the unconsumed inner range continuations forward.
### Related
- PR #46469 — initial fix for the K×N page-size bug + safety mitigation
- Files: `sdk/cosmos/azure-cosmos/azure/cosmos/_cosmos_client_connection.py`, `sdk/cosmos/azure-cosmos/azure/cosmos/aio/_cosmos_client_connection_async.py` (`__QueryFeed`, feed_range_epk branch)
### Affected SDK area
`Azure.Cosmos`
Beitragsleitfaden
Rechercherichtung
Lies PR #46469 und die __QueryFeed feed_range_epk-Zweige in sdk/cosmos/azure-cosmos/azure/cosmos/_cosmos_client_connection.py und sdk/cosmos/azure-cosmos/azure/cosmos/aio/_cosmos_client_connection_async.py. Lege fest, wie Fortsetzungen pro Bereich und verbleibende Obergrenzen dargestellt, codiert, decodiert und beim Fortsetzen weitergeleitet werden. Erledigt ist die Aufgabe, wenn Abfragen über mehrere Bereiche hinweg über Seiten fortgesetzt werden können, ohne Dokumente erneut auszugeben, Ergebnisse zu verlieren oder vorzeitig zu enden.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Bewertung
- Tech-Stack
- python
- Bereich
- databases
- Issue-Typ
- Bug
- Schwierigkeit
- 5/5
- Geschätzter Aufwand
- Über eine Woche
- Aktivitätsstatus
- Ruhig
- Klarheit
- Größtenteils klar
- Anfängerfreundlichkeit
- 35/100