> > > > I'll need to check more carefully. If that's true, the tests should show > data corruption (and they did a few times during development). Take a look > at the tests for the chunked item support? > > IE: If I allocate a new page to a slab class, the sequential bytes get > chopped up into a linked list. The first item chunk of a fresh boot will > naturally get linked to the next contiguous chunk of memory. > > So if you boot up a new server, write a random pattern, and the first > chunk is offset, it should overwrite the header of the next one if what > you said is true. That should leave to a crash, or incorrect results, or > etc. A few bytes of the chunk can be shifted due to alignment but being > off by an entire header is tougher. > > I also ran the code in a 12 hour torture test > setting/unsetting/overwriting while moving slab classes at the same time. > > but yes, it's written as a layer violation. my intent was to come back the > week after and refactor it more cleanly but I haven't done that yet. I'll > try to look at this soon but I have a few pressing bugs to cut a release > for.
That all said; are you looking into a particular bug or weirdness or anything? What's gotten you into this? -Dormando -- --- You received this message because you are subscribed to the Google Groups "memcached" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
