-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: 8.2.0-rc0
-
Component/s: None
-
None
-
Catalog and Routing
-
CAR Team 2025-06-23, CAR Team 2025-07-07
-
None
-
3
-
TBD
-
🟥 DDL
-
None
-
None
-
None
-
None
-
None
-
None
Compared to legacy "viewful" timeseries, with viewless timeseries listCollections no lists a system.buckets namespace. This has the potential to break existing usages of listCollections in the server, which may have been relying on this.
Â
The main affected use case is collection cloning (chunk migration, initial sync, movePrimary, shardCollection, resharding) but we should review all usages and consider using rawData mode (SERVER-105340).
Â
The main breaking points are:
- Unless rawData is used only the main namespace is shown, with type:'timeseries'. May break code that filters by type:'collection'
- No buckets namespace is present, the main timeseries namespace now has a UUID
- Unless rawData is used, clusteredIndex is hidden (pending
SERVER-105339)
Â
Similarly, code that relied on using listIndexes over the buckets namespace should use rawData mode instead.
- is related to
-
SERVER-106191 Add sanity check to prevent creating timeseries without clustered index
-
- Backlog
-
-
SERVER-105339 listCollections should hide clusteredIndex for viewless timeseries
-
- Closed
-
-
SERVER-105340 listCollections support for rawData mode for viewless timeseries
-
- Closed
-
- related to
-
SERVER-105850 Logical initial sync does not clone viewless timeseries collections
-
- Backlog
-