On 14/02/13 17:26, Greg Troxel wrote: > > thierry dijoux <[email protected]> writes: > >> After some investigations rebalance seems to work. > > Your example doesn't show that. > >> Here is the test : >> >> stop the 2nd node. >> Upload a file on node 1. >> Starting node 2 and do a "check --repair" => nothing. >> Now, in node1, rm .tahoe/storage/shares/de/derbnumtsxwu7rhydpxtxizd5y/1 and >> do a check --repair, the file is created on node2. > > But that isn't "rebalancing", which would describe the process of > checking files to see if any do not meet the balance criteria (as shown > by shares.happy). That was recovering from missing shares. I suspect > if you did the repair while node2 was down that you'd be back where you > started.
That's correct. >> I supposed the files named 0 and 1 corresponding to the index of node (in >> my case 0 => node1 and 1=>node2) > > I think the files in storage are the share #. Also correct. >> May i right or wrong ? I've only got 4 files, so it was "easy", but how to >> achieve this with tons of files ? > > What's needed is a way to have deep-check --repair consider a file > unheatlhy if it does not meet some distribution criteria. > > You might try setting shares.happy to 2 instead of 1. > > The real trouble is that this line of thinking leads to incremental > complexity. Next one wants one distribution criteria for "good enough > to place a file" and another for "good enough that we don't try to make > it better". The currently proposed algorithm is in <https://tahoe-lafs.org/trac/tahoe-lafs/ticket/1130#comment:12>. -- David-Sarah Hopwood ⚥
signature.asc
Description: OpenPGP digital signature
_______________________________________________ tahoe-dev mailing list [email protected] https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
