On Dec 18, 2007, at 2:23 PM, Robert Hailey wrote: > > Argghh, sorry. Bad commit-log comment... > > [...] > > I thought that I had added this return statement, but instead > committed some minor cleanups/comments and added more log noise. > > -- > Robert Hailey
I think I found the bug I was tracking BTW, 'receivedCompletionNotice' was not set in one of the two completed() routines of CHKInsertSender. In fact, since it was this broken function that was being called at the end of the block transfer, I suspect that we *never* would correctly wait on such an insert. I'm not sure how this affects inserts at large, I would think that such a bug would make all inserts timeout... Since applying this patch I've seen a huge spike in inserts (at the moment inserts outweigh requests 3-to-1), I would appreciate critical review of the patches (I know they may be hard to read). -- Robert Hailey
