You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Use a seperate memory pool for input vectors in ExpressionRunner
Summary:
At present, the expression runner utilizes the identical memory pool for both
the deserialization of input vectors and the execution of the expression
evaluation. Typically, inputs are produced by a distinct operator that has a
separate pool from the FilterProject operator. Therefore, to mimic this
scenario, this modification ensures that we use separate memory pools.
Reviewed By: kagamiori
Differential Revision: D54277992
Copy file name to clipboardexpand all lines: velox/docs/develop/testing/fuzzer.rst
+4
Original file line number
Diff line number
Diff line change
@@ -284,6 +284,10 @@ ExpressionRunner supports the following flags:
284
284
285
285
* ``--store_result_path`` optional directory path for storing the results of evaluating SQL expression or query in 'common', 'simplified' or 'query' modes.
286
286
287
+
* ``--findMinimalSubExpression`` optional Whether to find minimum failing subexpression on result mismatch. Set to false by default.
288
+
289
+
* ``--useSeperatePoolForInput`` optional If true (default), expression evaluator and input vectors use different memory pools. This helps trigger code-paths that can depend on vectors having different pools. For eg, when copying a flat string vector copies of the strings stored in the string buffers need to be created. If however, the pools were the same between the vectors then the buffers can simply be shared between them instead.
0 commit comments