Where a writer records each update and insert it lands, by the operation that
asked for it, so a retry of that operation is answered from the record rather
than written again. Without one, a create of an id already taken is refused,
retry or not. A table the application migrates:
target is the row (Entity:id), so one operation that updates two rows
records each, and an operation id sent again for another row is not that
row's retry. at is milliseconds, past a Postgres integer. request is
a SHA-256 digest of what the operation asked (its mutation, keys and
input): an operation id sent again with another request is refused rather
than answered as a retry. A ledger made before it gains the column with
alter table write_ledger add column request text; a record without one
is answered as a retry, as before, until its retention passes.