RFR: JDK-8319662 ForkJoinPool trims worker threads too slowly

Doug Lea dl at openjdk.org
Mon Nov 20 13:34:34 UTC 2023


On Mon, 20 Nov 2023 11:23:54 GMT, Viktor Klang <vklang at openjdk.org> wrote:

>> This update cascades timeouts to trim subsequent workers after the first  keepAlive inactive period.
>
> src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java line 2140:
> 
>> 2138:             Thread.interrupted();               // clear for next park
>> 2139:             if (deadline != 0L &&               // try to trim on timeout
>> 2140:                 deadline - System.currentTimeMillis() < TIMEOUT_SLOP) {
> 
> Since `currentTimeMillis()` is fetched from system time, I think it might be more appropriate to use nanoTime() to track delays.

It would be better, but absolute-time Unsafe.park only operates at millisecond accuracy (and even at that may misfire early, requiring TIMEOUT_SLOP.)

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/16725#discussion_r1399210760


More information about the core-libs-dev mailing list