[Bucardo-general] hub-and-spoke replication

Greg Sabino Mullane greg at endpoint.com
Mon Aug 24 17:06:41 UTC 2009


> So rows are replicated from B to A, and then back to B again. So
> performance under a scenario like this could be fairly bad. Bucardo
> could perhaps be modified to support this more gracefully, but it 
> would take some work (and some demand for the capability).

Right - we'd need to have makedelta start tracking not just what row has
changed, but *where* the change originated from. Further, we'd need to
distinguish local changes from remote ones - probably by timestamp. For
example, consider a change on B getting replicated to A, and then
Bucardo goes to replicate from A to B and C, sees that the change came
from B, and only replicates the row to C. Now what if A makes a change
to the row after it arrives from B, but before the sync to B and C? :)
We'd need to either kill the "got this from B" mark, or teach Bucardo
that any newer-non-marked row entries invalidate the don't-send-to-B hint.

Random Monday morning musings. Thanks for that patch, Josh!

-- 
Greg Sabino Mullane greg at endpoint.com
End Point Corporation
PGP Key: 0x14964AC8

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 226 bytes
Desc: OpenPGP digital signature
Url : https://mail.endcrypt.com/pipermail/bucardo-general/attachments/20090824/f6ff38e0/attachment.bin 


More information about the Bucardo-general mailing list