[Date Prev][Date Next] [Chronological] [Thread] [Top]

replog, misc



There was a thread on the -software list about Replication failing on a
subtree delete, because the delete ops got into the replog in the wrong
order. Any thoughts on how this might happen? Seems to me that if we're
dealing with a delete of a leaf node and its parent, there's no way they can
complete in the wrong order, so how can they get into the replog in the wrong
order?

I was thinking that perhaps we should set the default lock-detect method to
DB_LOCK_YOUNGEST and try to preserve the order of operations this way, but
still that should only affect the completion of peer-level deletes. Between a
leaf and its parent, the operations must necessarily complete in only one
order. Very strange.

So should I ask how syncrepl is coming along?  ;)

Anyone running into any problems with the current back-bdb or back-hdb? I
know Luke has hit some weird behavior, but I haven't heard much from anyone
else.

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support