cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Lohfink <clohf...@apple.com>
Subject Re: system_auth replication strategy
Date Sat, 01 Apr 2017 19:28:10 GMT
You should use a network topology strategy with high RF in each DC or something like the everywhere
strategy.

You should never really use SimpleStrategy, especially if you have multiple DCs and are using
LOCAL or EACH consistencies. Its more for test and dev setups then a prod environment.

The problem is that it DOES ensure a LOCAL consistency level will be targeted in the same
DC as the coordinator but it doesn ensure there will be data on each DC. This means that if
there are no nodes in your DC that are a replica of the data you can get an unavailable exception
even if 100% of your nodes are up and healthy.

So if you have 10 nodes, 5 per dc, 2 dcs and a RF of 3 with simple strategy

DC1
[0] [10] [30] [40] [45]

DC2
[1] [11] [15] [21] [41]

Especially if you have random token assignment like above. A partition with a token of 11
can end up on

DC1
[ ] [ ] [ ] [ ] [ ]

DC2
[ ] [*] [*] [*] [ ]

In which case an insert or read to a node on DC1 with LOCAL_ONE or LOCAL_QUORUM will result
in an unavailable exception.

Chris

> On Apr 1, 2017, at 10:51 AM, Vlad <qa23d-vvd@yahoo.com> wrote:
> 
> Hi,
> 
> what is the suitable replication strategy for system_auth keyspace?
> As I understand factor should be equal to total nodes number, so can we use SimpleStrategy?
Does it ensure that queries with LOCAL_ONE consistency level will be targeted to local DC
(or the same node)?
> 
> Thanks.


Mime
View raw message