[Berkman-sysadmin] berkman mirror

[mtoups at indymedia.org]

On Fri, 17 Mar 2006, Occam wrote:

>> Well we could rsync just the uploads.
>> otherwise, we'd need to use squid.
>>
>> but, this would defeat the download counter which someone recently
>> wrote.
>>
>> theoretically you could have a distributed download counter ;)
>
>
> a _Backup_ System would need
>
>  - MySQL Replication (need the exact same version of mysql on both
> hosts)
>  - rsync of all icecast and sf-active cfg files
>  - rsync of the upload dir
>  - rsync of a daily mysqldump -A
>
> a _Mirror_ System would need
>  - a squid mirror for the website
>  - a full icecast relay with all mount points using the "Master-
> Slave Relay" Feature of Icecast2 (http://www.icecast.org/docs/
> icecast-2.2.0/icecast2_relay.html)
>
> so what do we want? :)

it sounds like lotus is interested in a way to keep services running when 
there is berkman downtime.  unfortunately both of the points in occam's 
mirror list won't work when berkman goes down - you can't "relay" a stream 
if the master is down, and a squid cache will expire if the site it is 
caching goes down.

there may be workarounds - for example, an icecast *source* could send its 
stream to two master streaming servers, but that would require double the 
bandwidth on their end, which may not be possible in many cases.  and it 
may be possible to extend the expiry times on a squid cache so that it 
covers up some amount of downtime, though publishing new files would 
certainly not work.  so neither of these solutions really solve the 
problem - we need berkman to stay up.

the sf-active code in cvs supports a mir-like static mirroring system 
which would work well for the website.  but last time i checked the 
sf-active version on berkman was a few versions out of date as it is.  and 
that won't help at all with icecast.

backups, of course, are always a good idea.

-matt