Colleagues, we’re using CouchBase for many use-cases, one of those: session storage.
And we would very much like to survive 1 node failures, so we need replication.
After 1 node fails we will rely on auto-failover to remove it and promote replicas to active.
Then we think we’ll do rebalance and continue live happily [until next problem :)].
But we do not care for complete power outage and we do not need persistence.
Persistence in 2.0 version became very HDD hungry and we don’t like that at all.
We came across several hints that persistence is integral part of CouchBase but that is totally counterintuitive. One thinks of replication queues and disk persistence queues as separate entities.
(One reference here: Stop writing data to disk in couchbase)
(With advice to consider memcached bucket type; not applicable to our use-case, since we need replication)
Please clarify, if it is possible to disable persistence in any way?
At the moment, this is not a feature of Couchbase Server. There is no way to disable persistence permanently and intended for production use.
One suggestion, if you have very “slow” disk which is usually inexpensive, that would be fine and you’d still have the protection of memory to memory replication. Replication does not flow through disk IO.
Matt, according to our estimates, our update rate is so high that slow disks will never be able to cope with it.
The compaction speed will be slower than grow speed = disk eventually will overflow.
Do you know of any plans in this direction?
Prior to 2.0 our use-case was well covered and worked almost perfectly (except for nasty bugs that were fixed in 2.0).
Will it help if we’ll finally manage to buy the support (some misunderstanding prevent us to buy it for several months now)?
We feel our use-case seems quite valid and we’re really sad to be forced to buy more expensive hardware where cheaper hardware worked fine for us for years.