MySQL btr_search_guess_on_hash high load - mysql

Question

as I'm having some, let's say, "optimisation issues" with a multi source/channel replication slave, I was checking for "perf top" output to detect kernel function load anomalies.
The output shows a in my eyes high (17%-40%) overhead caused by btr_search_guess_on_hash.
[perf top]
Does somebody know if a high load on btr_search_guess_on_hash is a hint at mysql queries being badly optimised or an index or something missing? What am I looking for in order to optimise potential issues?
Thank you in advance!
Some query like this is most common for the server:
select table1.attr0, table1.attr1, table1.attr2, table1.attr3, table1.attr4, table1.attr5, table1.attr8, table1.attr6, table1.attr7
from table1
left outer join table2 on table1.attr6=table2.attr0
where table1.attr4='BLABLABLABLA' and table2.attr0=123456789
order by table1.attr7 desc
limit 1;
Show create table table1
CREATE TABLE `table1` (
`attr0` bigint(20) NOT NULL AUTO_INCREMENT,
`attr1` datetime DEFAULT NULL,
`attr2` datetime DEFAULT NULL,
`attr3` varchar(255) DEFAULT NULL,
`attr4` varchar(255) DEFAULT NULL,
`attr5` varchar(255) DEFAULT NULL,
`attr6` bigint(20) DEFAULT NULL,
`attr7` int(11) NOT NULL,
`attr8` bigint(20) DEFAULT NULL,
PRIMARY KEY (`attr0`),
KEY `FK2` (`attr6`),
KEY `FK1` (`attr8`),
CONSTRAINT `FK1` FOREIGN KEY (`attr8`) REFERENCES `table1` (`attr0`),
CONSTRAINT `FK2` FOREIGN KEY (`attr6`) REFERENCES `table2` (`attr0`)
) ENGINE=InnoDB AUTO_INCREMENT=12202000 DEFAULT CHARSET=utf8
Show create table table2
CREATE TABLE `table2` (
`attr0` bigint(20) NOT NULL AUTO_INCREMENT,
`attr1` datetime DEFAULT NULL,
`attr2` datetime DEFAULT NULL,
`attr3` varchar(255) DEFAULT NULL,
`attr4` varchar(255) DEFAULT NULL,
`attr5` varchar(255) DEFAULT NULL,
`attr6` bigint(20) DEFAULT NULL,
`attr7` int(11) NOT NULL,
`attr8` bigint(20) DEFAULT NULL,
PRIMARY KEY (`attr0`),
KEY `FK3` (`attr6`),
KEY `FK4` (`attr8`),
CONSTRAINT `FK4` FOREIGN KEY (`attr8`) REFERENCES `table2` (`attr0`),
CONSTRAINT `FK3` FOREIGN KEY (`attr6`) REFERENCES `table3` (`attr0`)
) ENGINE=InnoDB AUTO_INCREMENT=145366 DEFAULT CHARSET=utf8
Show create table table3
CREATE TABLE `table3` (
`attr0` bigint(20) NOT NULL AUTO_INCREMENT,
`attr1` datetime DEFAULT NULL,
`attr2` datetime DEFAULT NULL,
`attr3` varchar(255) NOT NULL,
`attr4` varchar(255) DEFAULT NULL,
`attr5` varchar(255) NOT NULL,
`attr6` bigint(20) DEFAULT NULL,
PRIMARY KEY (`attrr0`),
KEY `FK5` (`attr6`),
CONSTRAINT `FK5` FOREIGN KEY (`attr6`) REFERENCES `table4` (`attr0`)
) ENGINE=InnoDB AUTO_INCREMENT=5994 DEFAULT CHARSET=utf8
Show create table table4
CREATE TABLE `table4` (
`attr0` bigint(20) NOT NULL AUTO_INCREMENT,
`attr1` datetime DEFAULT NULL,
`attr2` datetime DEFAULT NULL,
`attr3` bit(1) NOT NULL,
`attr4` bit(1) NOT NULL,
`attr5` varchar(255) NOT NULL,
`attr6` bit(1) NOT NULL,
`attr7` bit(1) NOT NULL,
`attr8` bit(1) NOT NULL,
`attr9` varchar(255) NOT NULL,
`attr10` date NOT NULL,
`attr11` varchar(255) DEFAULT NULL,
`attr12` bit(1) NOT NULL DEFAULT b'0',
`attr13` varchar(255) DEFAULT NULL,
PRIMARY KEY (`attr0`)
) ENGINE=InnoDB AUTO_INCREMENT=5959 DEFAULT CHARSET=utf8

Answer 1

High CPU --> poor indexing and/or poor formulation. In this case, it is indexing of one of the tables.
from table1
left outer join table2 ON table1.attr6=table2.attr0
where table1.attr4='BLABLABLABLA'
and table2.attr0=123456789
order by table1.attr7 desc
We can't necessarily tell whether the Optimizer will start with table1 or table2, so let's devise optimal indexes for both and let it decide.
If t1.a0 is more selective:
t2: INDEX(a0)
t1: INDEX(a4, a6)
If t1.a4 is a better starting point, or of the Optimizer decides to include the ORDER BY in the indexing:
t1: INDEX(a4, a7)
t2: INDEX(a0)
(It is confusing to have attr0 in two tables meaning two different values. I assume it is really id?)
Seeing that t2.a0 is the PK, I recommend adding these 'composite' indexes, both to t1:
INDEX(a4, a6)
INDEX(a4, a7)
Since you have constraints on both tables, it is not really a LEFT JOIN. To avoid human confusion, please change to just JOIN. (INNER and OUTER are ignored.)
Since you are selecting no columns from t2, the JOIN could be turned into a "semi-join":
select table1.attr0, table1.attr1, table1.attr2, table1.attr3,
table1.attr4, table1.attr5, table1.attr8, table1.attr6,
table1.attr7
from table1
where table1.attr4='BLABLABLABLA'
AND EXISTS( SELECT 1 FROM table2
WHERE table1.attr6=table2.attr0
AND table2.attr0=123456789 )
order by table1.attr7 desc
limit 1;
(However, I am not sure this would speed it up. I worry that your obfuscation messed up some of the column names or logic.)