Cassandra multinode

[smile]

Hi All

I'm going to set up ORS in production environment and faced with question of Cassandra HA. If i want deploy ORS in primary+backup mode, what the minimum number of Cassandra nodes do i need?
Looks like the minimum is 3,but is it possible to set up only 2 like standard primary/backup or this config can't provide desired consistency?

[Kubig]

I think that Cassanda cannot be in standard primary-backup mode, but "only" in multinode deployment (Cassandra cluster), where each cluster is in "active" mode and the data are replicated trough whole cluster for SPOF (single point of failure) protection. Cassandra(Apache) is not a Genesys product, so, does not search any common behavior or configuration. So, my opinion is that you can create how many nodes you want and no limits or pre-requisite are there, except min. number of nodes in multi-node deploy are two :-)

[smile]

i know about it. but i'm trying to understand how to set it up in HA environment.  What is minimal Cassandra nodes enough to make ORS redundant?

[Kubig]

For ORS redundany you do not deal with Cassandra, but if you want to have Cassandra in no SPOF deployment too, you have to create two nodes at least.

[smile]

two nodes is ok, thank you!

[smile]

another one question in ORS context integration with context services in HA mode.
how to understand in SCXML to which of HA pair UCS (context services) should i connect? UCS doesn't provide virtual IP or something similar, the one way that i see is connecting to UCS_p and in case of error connect to backup UCS_b. Any other suggestions?

[GMG]

Smile,

First ORS works in load balancing mode(Active + Active),  N+1 architecture style.

Second minimum cassandra nodes required is 1, if you have 2 ORS, you can use same/single cassandra node in both ORS.


--
GMG

[smile]

Thanx, GMG, understood.

What about integration with UCS and OCS via http interface? How can i know which node is primary?

[Kubig]

I think that you do not need determine which node is primary, because all nodes should be with same role and each node should return same type of data. If you want to know more about Cassandra, forgot on Genesys doc and look at Cassandra Apache project web sites.

[smile]

ok, moved question to new topic http://www.sggu.com/smf/index.php/topic,8001.0.html

[smile]

Found answer on my question.

A quorum is calculated as (rounded down to a whole number): (replication_factor / 2) + 1

if we have 2 nodes cluster, then we will have to set replication_factor (RF) = 2: 2/2+1 = 2. quorum=2, i.e. every write transaction must be stored into 2 nodes. In case of one node failure, Cassandra can't do this.

If we have 3 nodes cluster and RF=2, quorum=2 as well, but in this case if one node failure, we still have two nodes for data writing.

Finally i decide to set up 4 Cassandra nodes and set RF=2.

[Kubig]

Great, very useful for others including me :-) The ORS is not deployed in high number, to tell truth, I have no customer with ORS in production env, so I never solve your needs, so do not know real-life experience. may be, in next years, will the count of customers with ORS increase. Personally, I do no see any adantages of using ORS in real-lie, but this is my opinion :-)

[smile]

ORS looks a little strange, but now we have a number of requests from our customers, first of all in proactive eservices, outbound area and complex IVR with external web services (without GVP). I found that using direct scxml request like ixn:createmessage WITH javascript give us more flexibility to render content of messages (sms and email) than URS can provide.
Migration to ORS is unavoidable, it's just a question of time

[smile]

[quote author=Kubig link=topic=8000.msg35141#msg35141 date=1379428254]
Yes, it is partly right. But for execution SCXML files you do not need ORS
[/quote]

what do you mean? 

[bublepaw]

If this is production env I would wait 2 weeks because new ORS will be released soon and it will have different architecture for HA (cassandra will still be there). Most of customers I've seen are using 2 nodes of cassandra per data centre. In practise with 2 node setup system still works with just one cassandra up.