[Xastir] Multiple Servers

[Gerry Creager]

Robbie, wa9inf wrote:
> Dan Brown wrote:
> 
>> On Wed, 25 Oct 2006, Ray Wells wrote:
>>
>>  
>>
>>> Gerry Creager wrote:
>>>
>>>   
>>>> But do you *need* to connect to each of the servers?  Connecting to 
>>>> the Core at one point, and using a rotate scheme, should accomplish 
>>>> what you would want, the ability to fail over another server when, 
>>>> albeit rarely, we have a server die.
>>>>
>>>> Multiple connections to multiple servers tends to lead to loops on 
>>>> the servers.  AE5PL's done a lot of work to stop that and the Core 
>>>> sysops have periodically gone thru and individually asked folks to 
>>>> help us stamp out redundant connections.
>>>>
>>>> gerry
>>>>
>>>>     
>>> Hi Jerry,
>>>
>>> No, I don't need to connect to each server at the same time. All I'm 
>>> looking for is a redundancy scheme whereby if one server falls over, 
>>> Xastir will attempt to connect to an alternative server in a manner 
>>> similar to UIView.
>>>
>>> Server failures are rare but, in accordance with Murphy's Law they 
>>> will happen when I'm not here, which is quite often, and I want to 
>>> cover that possibility.
>>>   
>>
>> I believe that this has been somewhat implemented by the APRS IS guys 
>> using
>> round robin DNS records with a 60 second (or so) ttl for 
>> "rotate.aprs.net"
>>
>> Using "dig" on a command line:
>> ------------------------------------------------------
>>
>> Myhost.foo$ dig rotate.aprs.net
>>
>> ; <<>> DiG 9.2.1 <<>> rotate.aprs.net
>> ;; global options:  printcmd
>> ;; Got answer:
>> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 6864
>> ;; flags: qr rd ra; QUERY: 1, ANSWER: 3, AUTHORITY: 2, ADDITIONAL: 2
>>
>> ;; QUESTION SECTION:
>> ;rotate.aprs.net.               IN      A
>>
>> ;; ANSWER SECTION:
>> rotate.aprs.net.        46      IN      A       165.91.140.28
>> rotate.aprs.net.        46      IN      A       64.58.200.20
>> rotate.aprs.net.        46      IN      A       134.173.254.38
>>
>> ;; AUTHORITY SECTION:
>> aprs.net.               5986    IN      NS      ns3.aprs.net.
>> aprs.net.               5986    IN      NS      ns7.aprs.net.
>>
>> ;; ADDITIONAL SECTION:
>> ns3.aprs.net.           5986    IN      A       24.123.66.139
>> ns7.aprs.net.           5986    IN      A       64.58.207.2
>>
>> ;; Query time: 47 msec
>> ;; SERVER: 192.168.1.7#53(192.168.1.7)
>> ;; WHEN: Tue Oct 24 19:56:43 2006
>> ;; MSG SIZE  rcvd: 149
>>
>> ------------------------------------------------------
>>
>>
>>
>> If you use dig a few times in a row, you'll see the 3 lines in the 
>> "Answer
>> section" change order, and the TTL (46 here) will decrease over time.  
>> I've
>> not looked at the Xastir code, but, As long as users use hostnames and
>> xastir (and other stuff like nscd) doesn't cache the Hostname->Ip mapping
>> any longer than specified, you should be able fail over by using
>> "rotate.aprs.net" within 60 seconds, plus or minus timeouts to see if 
>> things
>> are truely down.
>> Still, it would be nice functionality to have -- use one "interface"
>> specification and have multiple possible hostnames to connect to.  I
>> personally just have multiple interface specifications.  If one goes 
>> down,
>> I can start another manually, fairly easily.
>>  
>>
> I am new to xastir myself, and yes, UI-View and WinAPRS and others use 
> "one" server, which does all that if I understand what is being 
> attempted there..
> 
> Just set xastir up to connect to rotate.aprs2.net port 14580 and supply 
> a filter like m/50 for 50 km from your position. When one server that 
> you might be connected to drops out, within a few second you will be 
> connected to another one without any intervention by you.. Since a few 
> days ago only, I have used rotate.aprs2.net. Not long enough to tell 
> about dropping from a server. My UI-View station drops only when I get a 
> clitch in my router, or I kill the program for one reason or another.
> 
> That being said, did I miss something here? Were you discussing 
> something different... :-)  Sorry if I didn't get it...
<aprsIs:soapbox>
Those of us "in" the Core have received complaints of late when some of 
the Tier2 systems in their rotation went down.

OK:  I strongly support the concept of a tiered, distributed system of 
servers,  but from my rather self-centered, and professional, opinion, 
that implies a TOP tier that is the basis for that distribution.  The 
Tier2 folks have initiated their own "core" operation rather than 
continuing to recommend that the 2nd tier connect to the Core, and 
receive their regional data that way.  Or, gather *all* the data for 
distribution from the core and parse it out.  In other words, they've 
added a new layer of broadcast and abstraction I don't think was 
well-thought-out, beneficial, or necessary.

So: use rotate.aprs2.net and be willing to expect some delay and growing 
pains, or use rotate.aprs.net and attach to the core.

We're adding another server to the Core sometime soon... we're real 
close.  We already exchange our data using a new UDP method that's 
diminished our bandwidth requirements, allowing more scalability.  My 
bandwidth usage has never been a problem, but we do watch that sort of 
thing and work with AE5PL on mods to the JavARPSrvr code to see about 
real enhancements.  Note that that upgrade was accomplished without 
disturbing services to our users.

</aprsIs:soapbox>
-- 
Gerry Creager -- gerry.creager at tamu.edu
Texas Mesonet -- AATLT, Texas A&M University	
Cell: 979.229.5301 Office: 979.458.4020  FAX 979.862.3983
MAIL:  AATLT, 3139 TAMU
Physical: 1700 Research Parkway, Suite 160,
College Station, TX 77843-3139