Checkpoints

A checkpoint is the log position of the last source transaction TwinCommit applied. It lives in the target, in the schema named by state_schema (default twincommit), and is updated in the same transaction as the data.

Checkpoint tables

Table	Contents
twincommit.checkpoint	One row per pipeline: last applied log position and commit time
twincommit.table_state	One row per replicated table: state (snapshotting, streaming, broken), snapshot position, last error

Positions are opaque to the target. Only the source connector interprets them, for example as an LSN on Db2, an SCN on Oracle or an LSN on PostgreSQL and SQL Server.

Inspecting

twincommit status -config pipeline.yaml

or, while run is active, GET /status on the metrics address.

Resetting

To start a table over, re-snapshot it:

twincommit snapshot -config pipeline.yaml -tables SALES.ORDERS

Caution

Do not edit the checkpoint tables by hand. If the stored position is older than the oldest log file the source still has, run stops with an error rather than skipping data. Re-snapshot instead.

Log retention

The source must keep its log files, either active or retrievable from the archive, until TwinCommit has read them. Size your archive retention for the longest outage you want to survive without re-snapshotting.