HADOOP-16751: DurationInfo text parsing/formatting should be moved out of hotpath

[rbalamohan]

DurationInfo is in hotpath in S3A and it would be good to lazy init formatted text. https://issues.apache.org/jira/browse/HADOOP-16751 has the profiler output.

[hadoop-yetus]

💔 -1 overall

Vote	Subsystem	Runtime	Comment
+0 🆗	reexec	0m 0s	Docker mode activated.
-1 ❌	docker	0m 10s	Docker failed to build yetus/hadoop:104ccca9169.

Subsystem	Report/Notes
GITHUB PR	#1754
JIRA Issue	HADOOP-16751
Console output	https://builds.apache.org/job/hadoop-multibranch/job/PR-1754/1/console
versions	git=2.17.1
Powered by	Apache Yetus 0.11.1 https://yetus.apache.org

This message was automatically generated.

[steveloughran]

well spotted perf issue.

The patch is doing lazy eval, but it now does it in: start, finish and toString calls, so at least 2x current overhed.

better: just have a String text field which is set to the formatted result if the output is to be logged which is

(logAtInfo && log.isInfoEnabled() || log.isDebugEnabled()

set that in the constructor and reference as needed,

[rbalamohan]

Thanks for the review Steve.

It would be evaluated only once on need basis (textStr takes care of that).

Having String text based on (logAtInfo && log.isInfoEnabled() || log.isDebugEnabled(), it would fail in some cases. E.g invoking toString on new DurationInfo(log, false, "test") would end up with "null" for text.

[steveloughran]

I see. and reviewing it, yes, that getFormattedText() does update the cached value.

+1

[steveloughran]

merged to trunk; thanks