memcached documentation ?

[Brad Fitzpatrick]

On Tue, 2 Nov 2004, Shie Erlich wrote:

> Hi Brad,
> i'm sorry if this has been answered before, but i was unable to find
> information
> in the source package or in the mailing list archives.
> basically, i was hoping to find some general information about the way
> memcached
> works, in terms of atomicity, failure and recovery. For example, the
> following
> questions come to mind:
> 1) if a node wants to modify a value, is it locked first across all servers?
> 2) if (1), does the process of obtaining a lock is blocking the user?

No:  memcached is a hash of hashes.  The client does the first level hash,
picking a node to store data on.  The memcached nodes never talk to each
other.

> 3) is the data stored on all the nodes all the time? or is it stored on
> some of the nodes?

Just one.  Unless the client decides to failover to another.

> 4) what happens when a node shuts down? how does the system handle
> replicating the data?

It doesn't.  It's a cache, so presumably the application can generate the
data again from a slower source.

> and another question: how difficult do you think it would be to add
> support for disk storage?

Not worth it, whatever the cost... that's why we have databases.  :-)

- Brad