[
https://issues.apache.org/jira/browse/HADOOP-17463?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17358581#comment-17358581
]
Steve Loughran commented on HADOOP-17463:
-----------------------------------------
CPU cores have changed since the previous discussion and on recent intel x86
parts the TSC clock will be consistent across all cores in the same socket, so
risk of going backwords on rescheduling is 0 for a single-socket device.
But: this absolutely doesn't hold for multisocket servers, and I don't know
about AMD or ARM parts.
I do think we need some form of monotonic clock for our wait loops, or at least
those where an NTP update could trigger a false failure. VM suspend/resume,
well, that's where currentTimeMillis() is better than CPU counters -the virtual
CPU clock may not have increased, but relative to the outside world, time has
passed.
> Replace currentTimeMillis with monotonicNow in elapsed time
> -----------------------------------------------------------
>
> Key: HADOOP-17463
> URL: https://issues.apache.org/jira/browse/HADOOP-17463
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Ahmed Hussein
> Assignee: Ahmed Hussein
> Priority: Major
> Labels: pull-request-available
> Time Spent: 1h 40m
> Remaining Estimate: 0h
>
> I noticed that there is a widespread incorrect usage of
> {{System.currentTimeMillis()}} throughout the hadoop code.
> For example:
> {code:java}
> // Some comments here
> long start = System.currentTimeMillis();
> while (System.currentTimeMillis() - start < timeout) {
> // Do something
> }
> {code}
> Elapsed time should be measured using `monotonicNow()`.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
