Backend-defined resume position for ordered query results.
Offsets are meaningful only to the backend instance that produced them;
there is no cross-backend or cross-database global-ordering promise.
Sequence is the in-memory/RocksDB per-database append counter, and those
backends resume strictly after the supplied value. Cassandra per-id
envelopes use their journal sequence number; that capability has no
offset-resume parameter. Timestamp is the
PostgreSQL database timestamp plus journal sequence number. PostgreSQL
resumes inclusively at an equal composite position: rows sharing the exact
timestamp and sequence number may therefore be re-emitted after a restart,
but no row at that position is skipped. Consumers must persist and resume
offsets with the same backend and database that produced them.
PostgreSQL database timestamp and sequence-number resume position.
epoch_micros retains PostgreSQL’s microsecond timestamp precision.
The pair is ordered lexicographically. Since sequence numbers are only
unique within a persistence id, distinct rows can share a pair; resume
is inclusive for that pair and duplicate re-emission is intentional.