Concentrated storage load from servers of happiness placement

Wed Apr 3 12:24:16 UTC 2019

After some discussion of this on #tahoe-lafs on Freenode, I'm convinced
there really is a misbehavior here.  I've filed a ticket (<
https://tahoe-lafs.org/trac/tahoe-lafs/ticket/3022>) with essentially the
same content as my email and created a branch that includes a test case (<
https://github.com/tahoe-lafs/tahoe-lafs/blob/0e38fd27716b0cd4875e1b10829cbbdbade5a104/src/allmydata/test/test_happiness.py#L272-L317>)
that fails because of (kind of) that misbehavior.

There was a lot of head scratching over the use of maximum matchings of
bipartite graphs of peers and shares for placement decisions.  Hopefully
someone who knows more about the implementation can chime in with some
comments about that code.

Jean-Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20190403/f8255c1b/attachment.html>