[tahoe-lafs-trac-stream] [tahoe-lafs] #1641: fix regressions in convergent uncoordinated write detection

[tahoe-lafs]

#1641: fix regressions in convergent uncoordinated write detection
-------------------------+--------------------------------------
     Reporter:  kevan    |      Owner:  nobody
         Type:  defect   |     Status:  new
     Priority:  major    |  Milestone:  undecided
    Component:  unknown  |    Version:  1.9.0
   Resolution:           |   Keywords:  upload mutable mdmf sdmf
Launchpad Bug:           |
-------------------------+--------------------------------------

Comment (by kevan):

 The only example I've thought of so far involves storage servers with
 shares associated with a file that's being updated joining the grid in
 between the map update step and the publish step (so the servermap doesn't
 know about their shares). The old publisher would interpret the unknown
 shares as evidence of an uncoordinated write and cause the upload to fail.
 The new publisher might not; if each of the new servers happens to have a
 subset of the shares allocated to it by the publisher, or if the file is
 SDMF and the unknown shares are from a certain previous version of the
 mutable file, the new publisher will ignore them. That's bad because we
 want the publisher to have as much information as possible before
 publishing, and we should interpret shares that we didn't know about
 before publishing as evidence that we need to gather more information
 before pushing new shares.

 (is that unclear? I could try to make a concrete example if that'd help.)

-- 
Ticket URL: <https://tahoe-lafs.org/trac/tahoe-lafs/ticket/1641#comment:5>
tahoe-lafs <https://tahoe-lafs.org>
secure decentralized storage