Error Handling

Finch handles these MySQL errors without reconnecting:

After handling the errors above, Finch starts a new iteration from the first assigned trx.

Other errors cause Finch to disconnect and reconnect to MySQL, then start a new iteration. Reconnect time is not directly measured or recorded, but if it’s severe it will reduce reported throughput because Finch will spend time reconnecting rather than executing queries.

Query statistics are recorded when the query returns an error. This is usually correct because, for example, a lock wait timeout is part of query response time. However, for errors that cause a fast error-retry-error loop, it will skew statistics towards zero or artificially high values.