[sql_lab] Improve performance, only use slow func when needed

[dpgaspar]

CATEGORY

Choose one

• Bug Fix
• Enhancement (new features, refinement)
• Refactor
• Add tests
• Build / Development Environment
• Documentation

SUMMARY

When quering the examples table wb_health_population on SQLLab including all the columns on the statement:

SELECT "NY_GNP_PCAP_CD",
       "SE_ADT_1524_LT_FM_ZS",
       "SE_ADT_1524_LT_MA_ZS",
       "SE_ADT_1524_LT_ZS",
       "SE_ADT_LITR_FE_ZS",
  ... (8551 characters truncated) ... _IN_ZS",
       "SP_UWT_TFRT",
       country_code,
       country_name,
       region,
       year
FROM public.wb_health_population

The query would take approximately 50s to 70s to complete, yet the query itself only takes 10-20ms to execute. This is due to ParsedQuery.__extract_from_token function that extracts all the tables from a query, noticed that the property tables itself is seldom used/needed. This is a simple fix, that reduces the time to execute from 50s-70s to 10s-20s so 1/6 of the total time.

Expected, since sql_json execution path instantiates ParsedQuery 6 times.

Simple fix, but the __extract_from_token recursive loop should be visited and improved (@lilykuang).

ADDITIONAL INFORMATION

• Has associated issue:
• Changes UI
• Requires DB Migration.
• Confirm DB Migration upgrade and downgrade tested.
• Introduces new feature or API
• Removes existing feature or API

REVIEWERS

@robdiciuccio @willbarrett

[villebro]

Wow 👀 Nice improvement, would be great to get that __extract_from_token method refactored, I can't imagine it needs to take that long..!

[willbarrett]

LGTM! Thanks for finding + fixing this @dpgaspar

@villebro - we have someone on it. That function will get some attention in the near future.