HashJoinCost Class Reference

This class represents the cost of a hash join (excluding the cost of sub-paths). More...

#include <cost_model.h>

Public Member Functions
	HashJoinCost (THD *thd, const HashJoinMetrics &metrics, bool allow_spill_to_disk=true)
double	spill_to_disk_probability () const
double	init_cost () const
double	cost () const
double	probe_iterations () const
	The number of times the probe input must be scanned. More...
double	overflow_probability () const
	Estimated probability that the build input overflows one join buffer. More...

Private Attributes
double	m_spill_to_disk_probability
	The probability (range [0.0, 1.0]) of needing spill to disk. More...
double	m_overflow_probability {0.0}
	The probability (range [0.0, 1.0]) that the build input overflows one join buffer, regardless of whether spill-to-disk is allowed. More...
double	m_init_cost
	The cost of preparing to produce the first result row. More...
double	m_cost
	The cost of the hash join. More...
double	m_probe_iterations
	The number of iterations over the probe input. More...

Detailed Description

This class represents the cost of a hash join (excluding the cost of sub-paths).

Constructor & Destructor Documentation

HashJoinCost()

HashJoinCost::HashJoinCost	(	THD *	thd,
		const HashJoinMetrics &	metrics,
		bool	allow_spill_to_disk = true
	)

Member Function Documentation

cost()

double HashJoinCost::cost ( ) const

inline

init_cost()

double HashJoinCost::init_cost ( ) const

inline

overflow_probability()

double HashJoinCost::overflow_probability ( ) const

inline

Estimated probability that the build input overflows one join buffer.

Unlike spill_to_disk_probability(), this is independent of allow_spill_to_disk: a no-spill plan with overflow_probability()>0 will refill the hash table rather than write chunk files, but the in-memory buffer is still not reusable across re-Init() in that case.

probe_iterations()

double HashJoinCost::probe_iterations ( ) const

inline

The number of times the probe input must be scanned.

Greater than 1 when the build input does not fit in the join buffer and we need multiple passes (with hash table refills).

spill_to_disk_probability()

double HashJoinCost::spill_to_disk_probability ( ) const

inline

Member Data Documentation

m_cost

double HashJoinCost::m_cost

private

The cost of the hash join.

m_init_cost

double HashJoinCost::m_init_cost

private

The cost of preparing to produce the first result row.

m_overflow_probability

double HashJoinCost::m_overflow_probability {0.0}

private

The probability (range [0.0, 1.0]) that the build input overflows one join buffer, regardless of whether spill-to-disk is allowed.

m_probe_iterations

double HashJoinCost::m_probe_iterations

private

The number of iterations over the probe input.

Used by ProposeHashJoin() to account for probe re-scanning cost.

m_spill_to_disk_probability

double HashJoinCost::m_spill_to_disk_probability

private

The probability (range [0.0, 1.0]) of needing spill to disk.

---

The documentation for this class was generated from the following files:

• sql/join_optimizer/cost_model.h
• sql/join_optimizer/cost_model.cc