Add logging to avoid loss of information when exception changes during retries

[joker1007]

If a raised exception changes during retry executions and the exception is not retryable, the retrying is aborted instantly, but the information of the last exception is lost, because RetryExecutor throws only the first exception and OnRetry hook doesn't log about the last exception.

For example, like this:

2023-09-06 18:56:18.365 +0000 [WARN] (0018:task-0010): Operation fai
led (1213:40001), retrying 1/8 after 1 seconds. Message: Deadlock found when trying to get lock; try restarting transaction
2023-09-06 18:56:19.366 +0000 [ERROR] (0018:task-0010): Operation fa
iled (1213:40001)

In this situation, it is difficult to identify the root problem, so I would like to change to log also last exception.

[dmikurube]

Thanks! and sorry to be late. It's better at least than before, then approving it now.

(I was thinking about starting to use JsonValue in these plugins, which means that it wouldn't work with Embulk v0.9. However, it seems better to release this fix before JsonValue.)

---

Let me leave a comment and a question although approving it.

• We may want to have the logging message formats aligned with onRetry's. Maybe nice to update also onRetry.
• In onRetry's logging, did you have any intention for buildExceptionMessage, not just with logger.warn(..., Throwable)? (Confirming it here as you seem the author of Implements retryable execution feature #113.)

[joker1007]

A reason why I didn't use buildExceptionMessage is that the method displays only error messages.
I wanted the stack trace of the exception.

I don't think the logging format needs to be aligned with onRetry.
Because users usually don't need the stack trace of retriable exceptions but typically want the stack trace of not retriable exceptions or given up exceptions.

[dmikurube]

Got it. Sounds reasonable! Let me merge this.