-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: 9.0.0-rc0
-
Component/s: None
-
Catalog and Routing
-
Fully Compatible
-
v9.0
-
CAR Team 2026-08-17
-
200
-
馃煩 Routing and Topology
-
None
-
None
-
None
-
None
-
None
-
None
SERVER-124946 implemented creation of the shard authoritative collections during FCV upgrade, including config.shard.catalog.chunks which has secondary indexes. The creation of the indexes is done before entering kUpgrading, so that no authoritative DDLs are writing to it yet.
聽
Edge case: Before shard1 has entered kUpgrading, another shard0 could have already entered kUpgrading, ran an authoritative DDL, and committed authoritatively on shard1. In this case, shard1 ends up with documents on config.shard.catalog.chunks despite it not having the indexes. This can both:
- Trigger the foreground index creation bug described in SERVER-128952.
- Potentially cause suboptimal non-indexed queries.
聽
Proposed solution:聽Before any shard enters kUpgrading, all shards must have created config.shard.catalog.chunks with their indexes. We can do this by using the same pattern we already use for the drop, where the configsvr broadcasts the command before any shard enters kUpgrading.
- is related to
-
SERVER-128952 Rollback-to-stable causes untimestamped _mdb_catalog writes for unfinished single-phase indexes
-
- Backlog
-
-
SERVER-124946 Create/Drop authoritative indexes and collections on upgrade/downgrade
-
- Closed
-