RFR: JDK-8150068: Log the main G1 phases at info level
thomas.schatzl at oracle.com
Fri Feb 26 10:15:11 UTC 2016
On Thu, 2016-02-25 at 17:50 +0100, Thomas Schatzl wrote:
> Hi Bengt,
> On Mon, 2016-02-22 at 13:19 +0100, Bengt Rutisson wrote:
> > Hi again,
> > Now that JDK-8149013 has been pushed the only remaining caller of
> > G1GCPhaseTimes::accounted_time_ms() was G1GCPhaseTimes::print(). It
> > makes the code easier to follow if the calculation of the accounted
> > time
> > is inlined in the print method. Here's an updated webrev where
> > accounted_time_ms() has been removed and inlined in the print()
> > method
> > instead.
> > http://cr.openjdk.java.net/~brutisso/8150068/webrev.01/
> looks okay (let me have another look later or tomorrow) except for
I had another look on it, I think is actually a good change, except
for the handling of phases that are not active, requiring lots of
conditional code everywhere this might be the case.
Maybe adding to NULL WorkerDataArray's, this can be detected in the
accessor methods for them? Like doing nothing, or returning 0.0? (Some
lowest level debug log output could say that the value has been
replaced by some default action).
I would really prefer that.
Also, in G1GCPhaseTimes::print(), this information can be used to
suppress the output instead of (again) checking random places.
More information about the hotspot-gc-dev