Open hantangwangd opened 2 weeks ago
Is any documentation update needed for this?
@steveburnett Thanks for the reminder, I have updated the description for history_input_table_statistics_matching_threshold
in docs history-based-optimization.rst
to amend the description of default value and the behavior when set to 0.0
, please take a look. Additionally, should we open a dedicate issue for fixing the description of other session properties in history-based-optimization.rst
? The description for these session properties should combine with the corresponding config properties.
cc @feilong-liu @jaystarshot
@steveburnett Thanks for the reminder, I have updated the description for
history_input_table_statistics_matching_threshold
in docshistory-based-optimization.rst
to amend the description of default value and the behavior when set to0.0
, please take a look. Additionally, should we open a dedicate issue for fixing the description of other session properties inhistory-based-optimization.rst
? The description for these session properties should combine with the corresponding config properties.cc @feilong-liu @jaystarshot
If you find inconsistencies or gaps in the doc for the HBO session properties, I would welcome updates. @feilong-liu @jaystarshot, what do you think?
Just to double confirm, the default behavior will not change, i.e. the default match threshold stays the same?
Yes, the default behavior will not change, the actual default value before and after this change are all 0.1
.
Description
This PR integrate config property into session property for history-input-table-statistics-matching-threshold, so that we can get the property value uniformly through method
getHistoryInputTableStatisticsMatchingThreshold()
inSystemSessionProperties
.Besides, there exists a problem in the previous implementation logic when we trying to explicitly set the threshold to
0.0
through session property. By setting the value to 0.0 through session property, we intended to indicate that we will use historical statistics only when the input tables have exactly the same stats with those in the historical stats, but that's not the case in previous implement logic as we would then get the value from configuration property in this scenario, which is default not 0.0. This PR fix the problem as well.Test Plan
TestHiveHistoryBasedStatsTracking
including the scenario described aboveContributor checklist
Release Notes