java.lang.Object | |
↳ | org.apache.helix.util.HelixUtil |
Public Methods | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
Even though this is simple we want to have the mechanism of bucketing the
partitions.
| |||||||||||
get the last part of the zk-path
| |||||||||||
get the parent-path of given path
return "/" string if path = "/xxx", null if path = "/"
| |||||||||||
Attempts to load the class and delegates to TCCL if class is not found.
| |||||||||||
parse a csv-formated key-value pairs
|
[Expand]
Inherited Methods | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
From class
java.lang.Object
|
Even though this is simple we want to have the mechanism of bucketing the partitions. If we have P partitions and N nodes with K replication factor and D databases. Then on each node we will have (P/N)*K*D partitions. And cluster manager neeeds to maintain watch on each of these nodes for every node. So over all cluster manager will have P*K*D watches which can be quite large given that we over partition. The other extreme is having one znode per storage per database. This will result in N*D watches which is good. But data in every node might become really big since it has to save partition Ideally we want to balance between the two models
get the parent-path of given path return "/" string if path = "/xxx", null if path = "/"
Attempts to load the class and delegates to TCCL if class is not found. Note: The approach is used as a last resort for environments like OSGi.@return
ClassNotFoundException |
---|
parse a csv-formated key-value pairs
keyValuePairs | : csv-formatted key-value pairs. e.g. k1=v1,k2=v2,... |
---|