Background

A follower's checkpoint pick-up behaviour is currently unobservable in the field: there is no way to tell how far behind a follower's adopted checkpoint is from the newest one delivered to it, nor whether pick-ups are being held back by active readers.

WT-18156 introduces that behaviour (a pick-up may be deferred while a transaction snapshot that predates it is active) and needs these statistics to make it diagnosable. They are defined here on their own so that the generated statistic renumbering — inserting a statistic shifts every WT_STAT_CONN_* value after it, around 1,200 lines of wiredtiger.h.in — does not obscure the code review of that change.

Statistics

Connection:

• disagg_checkpoint_delivered_lsn — most recently delivered checkpoint metadata LSN (no_clear, no_scale).
• disagg_checkpoint_meta_lsn — most recently adopted checkpoint metadata LSN (no_clear, no_scale). Together with the delivered LSN this gives the follower's adoption lag.
• disagg_checkpoint_defer — pick-ups deferred for active transaction snapshots.
• disagg_snapshot_rebuild — snapshots rebuilt after racing a checkpoint pick-up or a role change.

Data source:

• layered_curs_open_stable_refused — layered cursor stable opens refused to preserve a transaction snapshot.

Note

The statistics are unused until WT-18156 lands, which sets and increments them.