Context

As part of the lock-free serverStatus effort (SPM-4755), slow serverStatus sections contribute to FTDC gaps when collection exceeds 1 second. Analysis in SERVER-124926 and the slow serverStatus spreadsheet shows that the shardingStatistics section is a significant contributor. The spreadsheet data was collected from live Atlas clusters.

Owning team

Suggested owner (from CODEOWNERS): Catalog and Routing – Routing and Topology (@10gen/server-catalog-and-routing-routing-and-topology)

Impact (from slow serverStatus logs)

• Tab: mongos
• Total duration share: 2.7% of slow-run section time
• Avg duration when slow: 473 ms
• Rows where section exceeded 1s: 2.4% of slow runs

Code location

• Section registration: ServerStatusSectionBuilder<ShardingStatisticsServerStatus>("shardingStatistics")
• Primary files:
  • src/mongo/db/sharding_environment/s_sharding_server_status.cpp (mongos)
  • src/mongo/db/sharding_environment/sharding_server_status.cpp (mongod)

Task

Please audit the shardingStatistics serverStatus section and remove any unnecessary blocking work (locks, synchronous I/O, contended atomics, etc.).

Use SKUNK-40 branch for reference implementations and SKUNK-40-nonblocking for nonblocking annotations.

Acceptance criteria

• shardingStatistics section generation does not acquire blocking locks on the hot path
• No regression in section output for FTDC/default serverStatus collection
• Add or update tests if behavior changes

Related work

• Parent analysis: SERVER-124926
• Prior work: SERVER-74720