Until the logrotate hit in the middle of some significant processing and a client had a query. What we found was a 2 second gap in the logs. The end of the old log was the start of a transaction and the start of the new log was the end of a completely different transaction. So we have lost partial information from two transactions and it is possible that there was at least one transaction between these two that was lost completely. Each transaction takes around 1 second to process, sometimes less

This time it was not too big an issue and there were other ways to investigate the problem

Is there some other way of logging that is even less likely to lose data?