Unique blobs allow duplicate values upon concurrent INSERT

CREATE DATABASE test;

CREATE TABLE t1 (

f text DEFAULT NULL,

UNIQUE KEY `f` (`f`) USING HASH

) ENGINE=InnoDB;

Run jmeter script UniqueHashTest.jmx

1. adjust database configuration variables (see image).
2. script creates a numeric counter, on each iteration launch 10 threads, that attempt to make the same insert concurrently with the counter value. Scrip finish after 200 seconds. Execute like:
   jmeter -n -t UniqueHashTest.jmx -l UniqueHashTest.out

SELECT count(1) recordCount, count(distinct f) uniqueCount, count(1)- count(distinct f) duplicatedCount

FROM t1;

Should present something like that:

Reproducible on 10.4, 10.5.