You can clone with
HTTPS or Subversion.
This is Issue 937 moved from a Google Code project.
Added by 2012-07-06T08:05:31.000Z by sylvain....@gmail.com.
Please review that bug for more context and additional comments, but update this bug.
Original labels: Type-Enhancement, Priority-Low, v2.2.0
Current state :
We can create up to 32000 clusters. These clusters are well designed for "structural clusters". I mean the clusters count is application
dependent. For example one cluster by table, etc.
The needs :
We need something like "dynamic cluster" or "live
cluster". I mean clusters depending on the life of the
application, not its structure. Examples :
- one cluster for each project in a large collaborative application :
perhaps 200 000 projects, at least one by user, etc.
- one new cluster per day or per hour on each table for logs purposes...
- No 32000 clusters limit.
- Performance. Live clusters will match very closely the application logic (with one data file per live cluster) : load / unload / delete / synchronize / browse / backup / sharding (...) a live cluster will be very easy and fast.
- Memory management. "Structural clusters" are loaded at startup. Live clusters should be loaded on demand. OMemoryWatchDog could automatically manage unloading live clusters.
- Cluster id : structural clusters use 0-32000 range and live clusters use 32001-2147483647 range.
- The biggest problem I see : RID serialization must be changed with an int for cluster id instead of a short.