[
https://issues.apache.org/jira/browse/PHOENIX-1264?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14144232#comment-14144232
]
James Taylor commented on PHOENIX-1264:
---------------------------------------
[~ramkrishna] - as an alternative like we talked about offline, you can
piggyback on the UngroupedAggregateRegionObserver coprocessor instead of
creating your own. Instead of having an endpoint coprocessor to re-analyze the
stats, set a special attribute on the scan and then run a count(*) query like
this example from MetaDataUtil. Might as well return how many rows were
analyzed (instead of always 1 in the MutationState), since you'll be
calculating this value:
{code}
private MutationState buildIndex(PTable index, TableRef dataTableRef)
throws SQLException {
...
String query = "SELECT count(*) FROM " + tableName;
QueryPlan plan = statement.compileQuery(query);
Scan scan = plan.getContext().getScan();
scan.setAttribute(BaseScannerRegionObserver.MySpecialAttribute,
PDataType.TRUE_BYTES);
// Executes the QueryPlan by getting the iterator and calling
next
Cell kv = plan.iterator().next().getValue(0);
ImmutableBytesWritable tmpPtr = new
ImmutableBytesWritable(kv.getValueArray(), kv.getValueOffset(),
kv.getValueLength());
// A single Cell will be returned with the count(*) - we decode
that here
long rowCount = PDataType.LONG.getCodec().decodeLong(tmpPtr,
SortOrder.getDefault());
state = new MutationState(0, connection, rowCount);
{code}
> Add StatisticsCollector to existing tables on first connection to cluster
> -------------------------------------------------------------------------
>
> Key: PHOENIX-1264
> URL: https://issues.apache.org/jira/browse/PHOENIX-1264
> Project: Phoenix
> Issue Type: Sub-task
> Reporter: James Taylor
>
> We'll need to add the StatisticsCollector to existing tables so that stats
> are kept for these tables. We typically do that on first connection to the
> cluster in ensureTableCreated when our system table is created by walking
> through the existing Phoenix tables. We only need to do it if the system
> table already exists (otherwise, it's a new installation).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
