If a single SQL query runs for 10 seconds or more, the connection is killed from the client side.
This could happen when slaving domains with a huge amount of rows - as all existing rows will be deleted in a DELETE statement at the beginning of the transaction, before re-populating the rows again.
With MySQL warnings enabled, MySQL server will log the following:
"[Warning] Aborted connection to db: '' user: '' host: ' (Unknown error)"
... while powerdns (at least in 3.4.1) will log:
"Database failed to start transaction: Failed to execute mysql_query, perhaps connection died? Err=1: Lost connection to MySQL server during query" - none of which make it obvious why the query failed.
Having a configurable timeout is obviously a better alternative, but as a temporary workaround, changing the timeout to a couple of minutes should hopefully minimise the risk in most situations.
Increase MySQL client timeouts from 10 seconds
If a query runs for 10 seconds or more, the connection is killed from the client side.
Bumping this timeout value to three minutes should hopefully be good enough for most situations, until a configurable option can be added.