On Sun, Oct 4, 2015 at 3:16 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Peter Geoghegan <pg@heroku.com> writes:
>> To be clear: I wasn't sure why you though I falsely count entries with
>> dropped texts within entry_dealloc().
>
> In the existing^H^H^Hprevious code, dropped-text entries would essentially
> act as length-zero summands in the average calculation, whereas I think
> we agree that they ought to be ignored; otherwise they decrease the
> computed mean and thereby increase the probability of (useless) GC cycles.
> In the worst case where the hashtable is mostly dropped-text entries,
> which would for instance be the prevailing situation shortly after a GC
> failure, we'd be calculating ridiculously small mean values and that'd
> prompt extra GC cycles no?
Yes, but my patch changed that, too. I suggested that first.
--
Peter Geoghegan