Character Sets

GoArchive reads and writes every text value through a utf8mb4 session, and some character sets do not convert to and from utf8mb4 exactly: two characters can read as one, or a character can read as ?. GoArchive therefore refuses a job whose columns use such a set, before any row is read.

The session character set #

Every connection is set to utf8mb4, collated utf8mb4_general_ci, after authentication, so a server’s init_connect setting cannot narrow it. Each source and destination connection then proves its session settings when it opens (SESSION_CHARSET_CHECK ). Comparisons in a job’s where and in GoArchive’s key SQL use that collation.

Columns are checked #

Before any row is read, preflight refuses a job column the run reads or writes as text whose character set is not supported (COLUMN_CHARSET_CHECK ):

ENUM and SET columns are checked. Numbers, temporal types, binary strings and JSON have no character set and are not. There is no bypass: both checks are enforced when preflight is bypassed, because the loss would be silent and could not be found after the run.

Supported character sets #

Refused character sets #

Every other character set is refused:

A refused column is converted to utf8mb4 on both servers, or its table is left out of the job.