It raised encoding failed when executing select statement on medium blob type

[lilinghai]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

deploy the tide cluster containing TiKV instance
download 1.txt

source 1.txt
select * from t where a=1;

2. What did you expect to see? (Required)

successfully

3. What did you see instead (Required)

ERROR 1105 (HY000): encoding failed
tidb log
[err="[tikv:10000]encoding failed"]

4. What is your TiDB version? (Required)

Release Version: v4.0.0-beta.2-2365-g46f5f2177-dirty
Edition: Community
Git Commit Hash: 46f5f21
Git Branch: release-5.0
UTC Build Time: 2021-03-13 14:50:59
GoVersion: go1.13
Race Enabled: false
TiKV Min Version: v3.0.0-60965b006877ca7234adaced7890d7b029ed1306
Check Table Before Drop: false

[zimulala]

/cc @XuHuaiyu

[zz-jason]

It's better to figure out the affected versions firstly.

[guo-shaoge]

drop table if exists t1;
create table t1(c1 blob);
insert into t1 values(unhex('f7'));
select hex(c1) from t1;
mysql> select * from t1 where c1 = 1;
ERROR 1105 (HY000): encoding failed

root cause: Copr tries to convert 0xf7 into utf8, then convert it to float, which will be compared by 1. But 0xf7 is not a valid utf8 char(one byte and first bit is 1, check wiki). So TiKV gives error. May be we should std::str::from_utf8_unchecked instead of std::str::from_utf8 when convert bytes to real or any other type.

[guo-shaoge]

We close this. Fix in tikv.