On Wed, 26 Oct 2011 08:31:53 -0700 Harry Mangalam <harry.mangalam at uci.edu> wrote: > Thanks very much for your input. > > I'm a bit surprised that new files would hash to the failed brick - > there isn't a check to make sure that the assigned brick is > responding and fall back to a ready brick? I can see that this would > happen in the 1st few seconds of failure, but after a short timeout, > shouldn't this feed back to the hasher? > > I'll explicitly test this when I bring up the new version today. Definitely a good idea. For all I know this behavior might have changed, but if it did then I must have been looking the other way when that happened. ;) It's not that dissimilar to the case where a brick is full, which is handled by putting files elsewhere. The only problem is that the file will then be in the "wrong" place according to the hash scheme, so lookups for it will be slightly more expensive. All of this is code that's already there for other reasons, but I wouldn't want to claim it's all seamless unless/until I've tested it myself (and I haven't for a while).
