[Asterisk-Users] realtime caching

[Damon Estep]

> > It seems that some options are not re-read when caching is on, for
> > example, changing the caller ID value in the sip table has no effect
> > until a reload (or expiration), so at least in some cases
rtcahcefriends
> > makes realtime notsorealtime.
> 
> 	No. It is doing exactly what it says it will, "cacheing". If you
> have
> rtcachefriends turned on, when a peer/user registers the info is
pulled
> from DB and added to the internal (a la 'in memory') list that
chan_sip
> maintains. If you change something in DB after this occurs then your
> changes won't take affect because chan_sip has no need to re-lookup
your
> phones info since the info is already present in memory.
> 
> 	What you can do is use "sip prune realtime <name>" to remove
just
> the
> single peer/user from memory. And you can force a reload of that peer
> from realtime by using "sip show peer <name> load".
> 
> 	If you want pure realtime where chan_sip always pulls from db,
then
> turn caching off. Keep in mind that turning caching off will remove
MWI
> and NAT functionality.
> 
> -Matthew
> 
What would it take (you, $) to add functionality that is a cross between
caching and not, that is it caches with a flag in the extension, so if
the flag is present realtime will be queried even though the extension
is in cache when a new call comes IN TO that extension.

Outgoing calls would not really need a re-query unless something about
the provisioning of the phone changes, at which point it would
re-register anyways, right?

The goal is caching for MWI and NAT but realtime for calling, so the
database is checked on every inbound call in case the dialplan changed,
and the cache updated accordingly.

Maybe a TTL flag, and when the TTL expires the cache entry stays, but is
re-queried when a dialplan match is found. The admin could then tune the
performance by setting different TTLs, maybe 15 minutes for lightly
loaded systems, 4 hours for heavy loaded systems.

Dynamic updates take place in whatever timeframe is specified on the TTL
or less.

Have I missed something, is this functionality already present?

Damon