[
https://issues.apache.org/jira/browse/HBASE-18573?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16129044#comment-16129044
]
Xiang Li commented on HBASE-18573:
----------------------------------
Thanks [~jerryhe].
I studied the code of ArrayList and find that
* New ArrayList with initial capacity of 1 is better (save space) when a caller
adds only one cell for a family and then processes the mutation.
* New ArrayList with no initial capacity specified is better when a caller adds
a lot of cells for a family.
{panel:title=More details}
When an ArrayList needs to inflate its backing array(elementData), the
increment when doing inflation is in proportion to the initial capacity.
{code:title=ArrayList#grow(int minCapacity)|borderStyle=solid}
int oldCapacity = elementData.length;
int newCapacity = oldCapacity + (oldCapacity >> 1);
if (newCapacity - minCapacity < 0)
newCapacity = minCapacity;
{code}
The newCapacity is max of
* min capacity required
* current capacity * 1.5
When initial capacity is not specified, default capacity 10 is used, so it is
more aggressive then having initial capacity to 1 in terms of inflation. That
is, with no initial capacity specified, an Object array with size of 10 has
been allocated even you only add 1 cell. If initial capacity is set to 1, the
array size is 1 when adding the first cell for a family, the array size is 2
when another cell is added to the same family.
{panel}
> Update Append and Delete to use Mutation#getCellList(family)
> ------------------------------------------------------------
>
> Key: HBASE-18573
> URL: https://issues.apache.org/jira/browse/HBASE-18573
> Project: HBase
> Issue Type: Improvement
> Reporter: Xiang Li
> Assignee: Xiang Li
> Priority: Minor
>
> In addxxx() of Put and Increment, Mutation#getCellList(family) is called to
> get cell list from familyMap. But in the other 2 sub-class of Mutation:
> Append and Delete, the logic like Mutation#getCellList(family) is used, like
> {code}
> List<Cell> list = familyMap.get(family);
> if(list == null) {
> list = new ArrayList<>(1);
> }
> {code}
> in
> {code}
> public Delete addColumn(byte [] family, byte [] qualifier, long timestamp)
> {code}
> of Delete
> We could make them to call Mutation#getCellList(family) to get better
> encapsulation
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
