RQG Performance Comparisons
Performance testing
The performance/perfrun.pl
executes each query against a set of tw servers
and reports the outcome.
perl performance/perfrun.pl --input-directory=... --dsn1=... --dsn2=... --output-file=... --filter=...
--input-directory
contains the queries to be run, one query per file. Alternative sources for queries will be made available in the future;--dsn1
and--dsn2
specify the locations of the two servers being compared in the Perl DBI URL format. If the queries are not fully qualified, the DSNs must contain the database name. The username and the password are also provided via the DSN. For example:
--dsn1=dbi:mysql:host=127.0.0.1:port=19300:user=root:database=test \ --dsn2=dbi:mysql:host=127.0.0.1:port=19302:user=root:database=test
--output-file=...
specifies a file where matching queries will be dumped in a form suitable for follow-up filtering withperfreport
- if no
--filter
is specified, all queries from the input will be reported;
Performance reporting
The performance/perfreport.pl
script takes an ouput file created by
perfrun
and prints out its contents, possibly applying a filter in the
process:
perl performance/perfreport.pl --input-file=... --filter=...
If no --filter
is specified, all queries present in the input
file are printed.
Filters
The queries reported by the system can be filtered out using perl-based filtering expressions. A filter can be applied during data collection, at which point it determines what information is stored in the output file and at
If you do not specify a filter, it is assumed that you want to output all queries.
The following variables can participate in filter expressions:
- Variables from
SHOW SESSION STATUS
, except:- variables that are not reset at the start of the query, which
includes
Com_*
,Uptime
,Opened_files
and the like; - variables that relate to the operation of SSL encryption or the query cache;
- variables that are not reset at the start of the query, which
includes
- Variables from
SHOW GLOBAL STATUS LIKE 'Innodb_%'
For each MySQL status variable, 4 Perl variables are provided
— the value of the MySQL variable from each server,
their absolute difference and their ratio. For example, for the
Innodb_rows_read
MySQL variable, you can use Innodb_rows_read1
,
Innodb_rows_read2
, Innodb_rows_read_delta
and
Innodb_rows_read_ratio
in your Perl filter expressions.
In addition to the MySQL status variables, the framework provides the following additional variables:
$Execution_time{1|2|delta|ratio}
reports the time the query took to run from start to end in seconds. The value is truncated at milliseconds in order to prevent exorbitant performance ratios from being reported on very fast queries;$Temperature
can either becold
orwarm
depending on whether the first execution of the query is being processed, or the second.$Query
contains the text of the query, in order to enable filtering such as$Query !~ m{l_receiptDATE}
For example, the following command-line option:
--filter='($Execution_time1 > 0.1 || $Execution_time2 > 0.1) && $Temperature eq "warm"'
Will only process or report queries that took more than 0.1 sec to execute on a warm server.
On-screen Reporting Format
The ASCII on-screen reporting format is as follows:
Query: SELECT l_linenumber FROM lineitem WHERE l_suppkey IN ( 10 , 1 ) AND l_shipdate NOT IN ( '1993-06-06' , '1998-02-04' , '1992-01-08' ) AND l_linenumber NOT IN ( 8 , 7 ) AND l_quantity <> 3 AND ( l_orderkey NOT IN ( 1298 , 93 ) OR l_suppkey = 10 ) AND ( l_suppkey BETWEEN 4 AND 10 + 4 OR l_linenumber = 2 AND l_commitDATE BETWEEN '1993-06-27' AND '1993-09-05' AND l_linenumber BETWEEN 3 AND 9 + 9 ); Cache: warm 5.3.0-MariaDB- 5.2.6-MariaDB- Delta Ratio ------------------------------------------------------------------------------------------------------- Execution time 0.011s 0.004s -0.007s 0.36 Innodb_buffer_pool_read_requests 2171 1836 -335 0.85 Handler_read_rnd 583 0 -583 Handler_read_next 583 602 19 1.03 Innodb_rows_read 1166 602 -564 0.52
Only variables whose values are different between the two servers are reported.
In this particular example, the query ran slower on 5.3.0-MariaDB
with warm
cache and caused twice as many Innodb_rows_read
.
On-disk Data Storage Format
The on-disk storage format is Data::Dumper
objects, wrapped in
<![CDATA[ ... ]]>
tags, without the file being a full-blown XML. The
serialized representation is created by
GenTest::QueryPerformanceDelta::serialize()
and is read by using eval()
in performance/perfreport.pl