sql: default sql.stats.statement_fingerprint.format_mask to use special flags

[xinhaoz]

Please note only the latest commit should be reviewed.

---

By default sql.stats.statement_fingerprint.format_mask is now
set to FmtCollapseLists|FmtConstantsAsUnderscores to reduce
statement fingerprint cardinality due to long constant lists and
variations in constant formatting. Note that the default fmt flag
for statement fingerprint generation is FmtHideConstants. Any
flags set with sql.stats.statement_fingerprint.format_mask will be
OR'd with FmtHideConstants.

Closes: #120409

Release note (sql change): Users will see the following changes
in their generated statement fingerprints from sql stats:

• lists with only literals/placeholders and similar subexpressions are
  shortened to their first item followed by "more", e.g.
• constants and placeholders are all replaced with the same character,
  an underscore _

SELECT * FROM foo WHERE f IN (1, $1, 1+2) ->
SELECT * FROM foo WHERE f IN (_, __more__)

[cockroach-teamcity]

This change is 

[abarganier]

, modulo a nit & question

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @dhartunian and @xinhaoz)

---

pkg/sql/conn_executor.go line 3533 at r1 (raw file):

	settings.ApplicationLevel,
	"sql.stats.statement_fingerprint.format_mask",
	"enables setting additional fmt flags  FmtHideConstants for statement fingerprint formatting. "+

nit: grammar and spacing

Code quote:

flags  FmtHideConstants f

---

pkg/sql/conn_executor.go line 3537 at r1 (raw file):

	int64(tree.FmtCollapseLists|tree.FmtConstantsAsUnderscores),
	settings.WithValidateInt(func(i int64) error {
		if i == 0 || int64(tree.FmtCollapseLists|tree.FmtConstantsAsUnderscores)&i == i {

So we only allow the clearing of the setting to 0, or masks that include tree.FmtCollapseLists|tree.FmtConstantsAsUnderscores in the bitmask? Just want to clarify.

Code quote:

int64(tree.FmtCollapseLists|tree.FmtConstantsAsUnderscores)&i == i

[xinhaoz]

pkg/sql/conn_executor.go line 3537 at r1 (raw file):

Previously, abarganier (Alex Barganier) wrote…

So we only allow the clearing of the setting to 0, or masks that include tree.FmtCollapseLists|tree.FmtConstantsAsUnderscores in the bitmask? Just want to clarify.

Ideally yes but this bitmask is more permissible, allowing any setting that is a subset of bits in the union of bits of both flags. I think this is good enough for an undocumented setting. Otherwise we'll need to check every combination of all flags in the future (assuming we'll add more). Let me know if you think we shuold tighthen the check to be more strict though.

[abarganier]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 1 stale) (waiting on @dhartunian and @xinhaoz)

---

pkg/sql/conn_executor.go line 3537 at r1 (raw file):

Previously, xinhaoz (Xin Hao Zhang) wrote…

Ideally yes but this bitmask is more permissible, allowing any setting that is a subset of bits in the union of bits of both flags. I think this is good enough for an undocumented setting. Otherwise we'll need to check every combination of all flags in the future (assuming we'll add more). Let me know if you think we shuold tighthen the check to be more strict though.

That sgtm, just wanted to make sure I understood! Thanks

[xinhaoz]

TFTR!
bors r+

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)
• check_generated_code
• docker_image_amd64
• examples_orms
• macos_amd64_build
• macos_arm64_build
• windows_build