Cosmos DB Continuation Token Validity Under Session Consistency With Cross-Partition Writes

Pagination token stability during concurrent modifications

Azure Cosmos DB SQL API uses opaque continuation tokens to paginate cross-partition query results. Each token encodes the query state across all target partitions and does not expire, but its logical position can shift when the underlying data changes or when partition topology evolves.

The SDKs automatically handle partition splits and merges, yet the continuation token's logical position may shift, leading to non-deterministic ordering if the query lacks an explicit ORDER BY. The service does not provide a built-in point-in-time snapshot for cross-partition queries unless the account consistency level is Strong, and even then, new matching documents inserted after the initial request may not appear in subsequent pages.

An unresolved behavior is whether a continuation token captured at consistency level Session remains valid if the session token advances due to writes in the same session but targeting a different partition.

Specific questions

• Does a Session-consistency continuation token remain valid for resuming pagination when the session token advances from writes to a different partition?
• What guarantees exist for token stability when concurrent writes modify matching documents across partitions during pagination?
• How should applications bound query execution time when continuation tokens may return duplicate or missing items under Eventual consistency?