Quanlong Huang has posted comments on this change. ( http://gerrit.cloudera.org:8080/24993 )
Change subject: IMPALA-15415: [DOCS] Documentation for HBO ...................................................................... Patch Set 1: (9 comments) Thanks for adding the docs quickly! http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml File docs/topics/impala_history_based_optimization.xml: http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@38 PS1, Line 38: cardinality "execution" or "runtime" is better. Cardinality is one kind of stats and we will support more stats types in the future. http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@42 PS1, Line 42: <p>HBO support is implemented incrementally by plan node types. Impala currently supports HBO : for the following plan nodes:</p> : <ul> : <li>ScanNode</li> : <li>AggregationNode</li> : <li>UnionNode</li> : <li>JoinNode</li> : </ul> This will be stale quickly since we are actively adding new node types. Might be better to say "HBO support is implemented incrementally by plan node types and statistics types. Impala currently supports HBO cardinality for HdfsScanNode, AggregationNode, UnionNode, JoinNode and SortNode, etc." http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@55 PS1, Line 55: <p outputclass="toc inpage all"/> Can we show some examples? e.g. use some outputs from https://github.com/apache/impala/blob/a64de02/testdata/workloads/functional-query/queries/QueryTest/hbo-multiple-scans.test http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@70 PS1, Line 70: STORE_HBO_STATS</codeph>=<codeph>true</codeph>;</li> Let's mention the historical execution stats are stored asynchronously after the query finishes, so this won't add significant latency. http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@101 PS1, Line 101: query The key can be shared by similar queries. Using "hash key" might be more specific. Also add "When exceeded, the oldest run is evicted." http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_history_based_optimization.xml@105 PS1, Line 105: <entry>Specifies the maximum memory size, in bytes, allocated for the in-memory backend HBO cache.</entry> Let's mention the default values for these flags. We can also reuse some descriptions here: https://github.com/apache/impala/blob/51e080d2933eae166c1cb50e2e0d1785bea5e25b/be/src/common/global-flags.cc#L254-L265 http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_store_hbo_stats.xml File docs/topics/impala_store_hbo_stats.xml: http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_store_hbo_stats.xml@40 PS1, Line 40: upon completion Let's emphasize its asynchronous. http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_store_hbo_stats.xml@49 PS1, Line 49: (ScanNode, AggregationNode, UnionNode, and JoinNode) This is stale since SortNode is also supported now. Let's use "e.g. HdfsScanNode, AggregationNode, JoinNode". http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_use_hbo_stats.xml File docs/topics/impala_use_hbo_stats.xml: http://gerrit.cloudera.org:8080/#/c/24993/1/docs/topics/impala_use_hbo_stats.xml@40 PS1, Line 40: (ScanNode, : AggregationNode, UnionNode, or JoinNode) use "e.g. ..." -- To view, visit http://gerrit.cloudera.org:8080/24993 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: I75fc76691ceed4e74e5c1aec3a409e1c41d4db7d Gerrit-Change-Number: 24993 Gerrit-PatchSet: 1 Gerrit-Owner: Sanjana Malhotra <[email protected]> Gerrit-Reviewer: Aleksandr Efimov <[email protected]> Gerrit-Reviewer: Impala Public Jenkins <[email protected]> Gerrit-Reviewer: Quanlong Huang <[email protected]> Gerrit-Reviewer: Sanjana Malhotra <[email protected]> Gerrit-Comment-Date: Thu, 08 Oct 2026 09:34:57 +0000 Gerrit-HasComments: Yes
