response to your post on LJ re zeroconf-based formation/management feature...

Lisa Seelye
Thu, 09 Oct 2003 12:51:19 -0400

Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Thu, 2003-10-09 at 11:03, Justin Matlock wrote:
> Referring to:
> I didn't want to clutter up your LJ with technobabble.  :)
> Ryan, jump in and correct me if I miss something..
> This sounds like a great idea, from a very high level.


I think the concerns that Brad raises in his "blog" are very important
-- having nodes crap out that stored important information can be a
problem.  Eventually you'd have to fetch it from the database and store
it in a cache, thus duplicating data.

However, if you set up a cluster of memcached servers behind one or two
synchronised master (that would store the key and pair it to the IP of a
machine that would actually store the real data) you could kind of get a
RAID effect going on.  One of the masters would query its network
periodically for new (and dead) servers.  The use for this would be to
allow the API to define one or two memcached [addresses], maybe on a DNS
round-robin, and for the staff to have some real scalability.  To handle
servers going offline and online, any server that didn't answer the
zeroconf "ping" would be considered to be dead, and the data stored in
it to be lost.  Any new server coming online would be ready to store new

In effect, this would be transparent to the API.

If that doesn't make sense I'll draw up a diagram. ;-) =20

<Vix ulla tam iniqua pax, quin bello vel aequissimo sit potior>

Content-Type: application/pgp-signature; name=signature.asc
Content-Description: This is a digitally signed message part

Version: GnuPG v1.2.3 (GNU/Linux)