Stefan Fuhrmann wrote on Thu, Feb 10, 2011 at 00:08:46 +0100: > On 09.02.2011 03:43, Daniel Shahaf wrote: > >Stefan, > > > >> < wayita> stefan2: danielsh said: does 'svnadmin dump/verify' use the > >> caches by default? > >><@stefan2> danielsh: yes. However, the biggest performance improvement for > >>dump / verify > >> will come from delta-caching. This is like 2 merges away. I'm > >> currently > >> preparing new patch sets for merger > >Is it possible that 'verify' (and 'dump' when used for verification purposes) > >would, when using the cache, overlook a corruption that it would have > >spotted if > >the cache was disabled? > In very loose terms, that is "possible" - just like > any code may malfunction in unforeseen ways. >
Murphy's law, yes. Any such bugs can go through the normal "find, fix, backport, patch release" procedure. They weren't my concern. > But assuming the cache works correctly (storing > and returning the right data for the right keys), > it will only prevent *repeated* data reads. > Assuming further that repeated reads would > always yield the same results, the cache should > not affect the results of 'verify' nor 'dump'. > Thanks; I accept these assumptions and I agree that under them the cache shouldn't hide corruptions. > However, it could hide corruptions that occur > while the check is running - which is no different > from today's situation. > :-) > -- Stefan^2. Thanks, Daniel
