[Freeswitch-users] FreeSWITCH HA question&talk

[Seven Du]

Hi,

I tested FS HA, it works like a charm. however, I'd like to talk about
some more:

I tested on my local computer by enabling tack-calls.  make a call and
kill FS, restart FS and execute sofia recover.

1) originate user/1000 &record(/tmp/blah.wav)

Recover works, however, it overwrote the recording. Added
RECORD_APPEND=true made it work.

But for other APPs like socket, it might make a new socket to a
outbound socket, can it send a var like channel_recovered so a
outbound socket can aware of a recover?


2) call from 1000 to 1001

Recover works. If I also do uuid_record, the recording cannot recover.
Seems there are no infos in sip_recovery table.

Neither does record_session, though there's an entry in sip_recovery
table. I guess it might be possible to get this work.

If it can send a channel_recovered event, then it might be possible to
continue the record in outside logic(say, event socket).

3) call from 1000 to a gateway outside (NAT-ed), leg-a can recover,
leg-b failed with:

2010-10-16 10:28:17.162241 [ERR] sofia_glue.c:3190 AUDIO RTP REPORTS
ERROR: [Bind Error!]

I guess it is caused by NAT. (since I kill & restart FS manually, it
may take 10 seconds). Is it possible to change the router settings? Or
could FS send a re-Invite to re-establish media?


Below are some fantastic thoughts:

4) N + M redundancy

Default FS HA is designed to 1+1 redundancy, it might be interesting
for N(active)+M(standby) in large farms. If one active FS fails, bind
the float IP to a standby server and go.

5) N (N>2) servers without float IP

Let's say each client UA register to 2 (or more) servers, If one
server fails, a FS controller then re-call the client in an active
call through a pair-server and the client responsible to hangup the
originate call and automatically answer the new one. This need client
support of course.  :)


-- 
Blog: http://www.dujinfang.com
Proj:  http://www.freeswitch.org.cn