[
https://issues.apache.org/jira/browse/FLINK-12161?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
godfrey he updated FLINK-12161:
-------------------------------
Description:
To resolve data-skew for distinct aggregates on stream, we introduce a rule
named {{SplitAggregateRule}} which rewrites an aggregate query with distinct
aggregations into an expanded double aggregations. The first aggregation
compute the results in sub-partition(with bucket) and the results are combined
by the second aggregation.
if two-stage aggregation is also enabled, we find that many plans have common
pattern, looks like:
{code}
... (output)
StreamExecGlobalGroupAggregate (final global agg)
+- StreamExecExchange
+- StreamExecLocalGroupAggregate (final local agg)
+- StreamExecGlobalGroupAggregate (partial global agg)
+- .... (input)
{code}
There is no exchange between the final local aggregate and the partial global
aggregate, so they will be executed in a same JobVertex, and could share state.
We introduce a rule named {{IncrementalAggregateRule}} to do that optimization.
was:
To resolve data-skew for distinct aggregates on stream, we introduce a rule
named {{SplitAggregateRule}} which rewrites an aggregate query with distinct
aggregations into an expanded double aggregations. The first aggregation
compute the results in sub-partition(with bucket) and the results are combined
by the second aggregation.
if two-stage aggregation is also enabled, we find that many plans have common
pattern, looks like:
{code}
...
StreamExecGlobalGroupAggregate (final global agg)
+- StreamExecExchange
+- StreamExecLocalGroupAggregate (final local agg)
+- StreamExecGlobalGroupAggregate (partial global agg)
+- ....
{code}
There is no exchange between the final local aggregate and the partial global
aggregate, so they will be executed in a same JobVertex, and could share state.
We introduce a rule named {{IncrementalAggregateRule}} to do that optimization.
> Supports partial-final optimization for stream group aggregate
> ---------------------------------------------------------------
>
> Key: FLINK-12161
> URL: https://issues.apache.org/jira/browse/FLINK-12161
> Project: Flink
> Issue Type: New Feature
> Components: Table SQL / Planner
> Reporter: godfrey he
> Assignee: godfrey he
> Priority: Major
>
> To resolve data-skew for distinct aggregates on stream, we introduce a rule
> named {{SplitAggregateRule}} which rewrites an aggregate query with distinct
> aggregations into an expanded double aggregations. The first aggregation
> compute the results in sub-partition(with bucket) and the results are
> combined by the second aggregation.
> if two-stage aggregation is also enabled, we find that many plans have common
> pattern, looks like:
> {code}
> ... (output)
> StreamExecGlobalGroupAggregate (final global agg)
> +- StreamExecExchange
> +- StreamExecLocalGroupAggregate (final local agg)
> +- StreamExecGlobalGroupAggregate (partial global agg)
> +- .... (input)
> {code}
> There is no exchange between the final local aggregate and the partial global
> aggregate, so they will be executed in a same JobVertex, and could share
> state. We introduce a rule named {{IncrementalAggregateRule}} to do that
> optimization.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
