How long is Snowflake's string min/max? ~32 bytes undocumented (measured on 10.25.101)

gist-snowflake-minmax.md

CREATE OR REPLACE TABLE probe_kK (s VARCHAR) AS
SELECT REPEAT('A', K-1) || TO_CHAR(SEQ8())
FROM TABLE(GENERATOR(ROWCOUNT => 500000000))
ORDER BY 1;   

SET probe = (SELECT s FROM probe_kK SAMPLE (1 ROWS));
SELECT COUNT(*) FROM probe_kK WHERE s = $probe;

SELECT operator_statistics:pruning:partitions_scanned::INT AS scanned,
       operator_statistics:pruning:partitions_total::INT   AS total
FROM TABLE(GET_QUERY_OPERATOR_STATS(LAST_QUERY_ID(-1)))
WHERE operator_type = 'TableScan';

K	Scan part/Total Part	pct	pruning
32	2 / 16	12.5%	✅
33	15 / 16	93.8%	❌
34	15 / 16	93.8%	❌
36	15 / 16	93.8%	❌
40	22 / 23	95.7%	❌

Use Chinese char

table	prefix	Distinguish character positions	Distinguish byte positions	Scan / Total	pct	pruning
probe_mb6	中×5 + SEQ8	char6	byte16	1 / 16	6.3%	✅
probe_mb11	中×10 + SEQ8	char11	byte31	2 / 16	12.5%	✅
probe_bx31	中×10 + 'A' + SEQ8	char12	byte32	1 / 16	6.3%	✅
probe_bx32	中×10 + 'AA' + SEQ8	char13	byte33	15 / 16	93.8%	❌
probe_mb12	中×11 + SEQ8	char12	byte34	16 / 16	100%	❌
probe_mb33	中×32 + SEQ8	char33	byte97	22 / 24	91.7%	❌

Conclusion

• If two rows first differ within the first ≤32 bytes → partition min/max values are distinguishable → pruning works.
• If they first differ at byte ≥33 → all partitions share the same first 32 bytes in their min/max → no pruning → near full-table scan

---

Measured on Snowflake 10.25.101. Corrections welcome.