On Mon, Apr 23, 2001 at 10:01:53PM -0700, Mr . Bad wrote: > >>>>> "TC" == Tavin Cole <tavin at mailandnews.com> writes: > > TC> Folks, apparently we have a problem again with the datastore > TC> eating infinite CPU cycles while doing key searches. Mark > TC> J. Roberts has also confirmed this. > > Is there a reason you think it's the tree? Besides that it was the > tree before? Are we sure it's key searches? This guy seems to be > having problems with connections, not searches. The authentications > keep timing out. > > If there's some other evidence that the tree is at fault, well, there > are some possibilities why a bug would display itself only after a > long time. One is that the tree is getting wildly out of balance and > the re-balancing code never gets called. However, it's SUPPOSED to get > called every time the store file is written, so I don't think that's > the case. > > Another possibility is that there's a weird data state (like before, > with the key-less, data-less data items) that's putting the > tree-search code into a weird loop. That would explain why something > wouldn't show up until after the node had been running a while -- it > might just take some time to get to that state, or it might be a rare > state that doesn't get persisted to the store file. > > Do you have any other leads? This is kind of intriguing. > > ~Mr. Bad
I think it's the tree because Mjr found a datastore key-search loop in his logs and he said it only started happening after the tree went in. Purely circumstantial of course.. and perhaps unrelated to the other poster's problem. --
