On Fri, Apr 25, 2008 at 7:03 AM, Matthew Toseland <toad at amphibian.dyndns.org> wrote: > On Thursday 24 April 2008 17:19, nextgens at freenetproject.org wrote: > > Author: nextgens > > Date: 2008-04-24 16:19:07 +0000 (Thu, 24 Apr 2008) > > New Revision: 19544 > > > > Modified: > > trunk/freenet/src/freenet/client/FECCodec.java > > Log: > > Hopefully fix #2287: Freenet uses too many fd's when starting a 2G+ insert, > results in failure to persist > > > > Now we leave it up to the GC ASAP > > This isn't going to work. Bucket.getOutputStream() doesn't append, at least, > that's always been my assumption. And Bucket.getInputStream() *ALWAYS* starts > at zero. I expect this commit to seriously break things - did you test it? > > It is conceivable that we'd run out of threads on a quad core system (192 > fd's > per encode/decode, 4 at once), but imho it's unlikely - if we're running out > of fd's, most likely there's a leak, surely?
My blackbox testing show inserting a file can take at most ~256 FD per core. Never see any leakage in my testing.. Regards, Daniel
