[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[EP-tech] select and mysql 5.6.27

We're also on 3.3.10, and  have our optimizer_search_depth set to 3 (code below) which seems to work well for us.
This is useful: https://mariadb.com/blog/setting-optimizer-search-depth-mysql - and explains what a setting of 0 does too.


In: ~/perl_lib/EPrints/Database/mysql.pm
sub connect
        my( $self ) = @_;

        my $rc = $self->SUPER::connect();

        if( $rc )
                # always try to reconnect
                $self->{dbh}->{mysql_auto_reconnect} = 1;

                $self->do("SET NAMES 'utf8'");
                # JLRS 2014-03-18
                # See: https://github.com/eprints/eprints/issues/120
                # See: http://www.eprints.org/tech.php/18677.html
                $self->do('SET @@session.optimizer_search_depth = 3;');
        elsif( $DBI::err == 1040 )
                EPrints->abort( "Error connecting to MySQL server: $DBI::errstr. To fix this increase max_connections in my.cnf:\n\n[mysqld]\nmax_connections=300\n" );

        return $rc;

-----Original Message-----
From: eprints-tech-bounces at ecs.soton.ac.uk [mailto:eprints-tech-bounces at ecs.soton.ac.uk] On Behalf Of Paolo Tealdi
Sent: 07 March 2016 10:33
To: eprints-tech at ecs.soton.ac.uk
Subject: Re: [EP-tech] select and mysql 5.6.27

On 07/03/2016 09:25, Paolo Tealdi wrote:
> Dear all,
> After 5 year of normal use with mysql 5.1.XX, after the upgrade to mysql 5.6.27,
> we noticed that our eprints sometimes (1 time every two day more or less)
> suddendly blocks itself.
> Analyzing the problem we noticed that one of the mysql connections from eprints
> server (we are using a centralized version of mysql) is hanging in "statistics"
> status on a very big select (in attach).
> All the other processes are waiting for table lock.
> After  the Kill of that processs,   all the other processes flushed immediately
> and the server returns to life.
> Have you ever noticed this BIG issue ? Any idea ?
> In attach one of the sql commands.
> Our optimizer_search_depth is set to 0.
> Best regards,
> Paolo Tealdi
> P.S. One of the  problems probably is due to the eprint__rindex size ... 37M of
> records ...

Hi all.

Thank for your answers.
I'll study more deeply all your answers.
What i want to  say to you is that :

-) the problem persists if i set optimizer_search_depth  to '62' (the server 
default). I'll try with other values.
-) the problem disappears when i try it on a copy of my eprints production 
database on my linux ubuntu workstation with a 5.5.47 vanilla mysql. :-( 
(simply copying and pasting the query in mysql character client). That query 
return an answer in 2 seconds, while on the production server had been killed 
after 1 hour of down and mysql process at 100% cpu.
-) The base search is set on xapian search (almost it should be). I'll check it
-) the advanced search is on sql search
-) My eprints production  server is a modified (but not in the search engine 
parts) version of 3.3.10 version. It's on since 2011 without many changes. None 
in the indexing part.
-) I'm now comparing the system variable between the two mysql server. I'm 
noticing that there are many new variables in the optimization section between 
5.5 and 5.6.

I'll keep all you updated.

Best regards and thanks again,

P.S. grrrrr

Ing. Paolo Tealdi         Area IT - Politecnico Torino
Telefono/Phone : +39-011-0906714 , FAX : +39-011-0906625
Indirizzo/Address : C.so Duca degli Abruzzi,  24 - 10129 Torino - ITALY
Skype : tealdi.paolo
Please consider your environmental responsibility before printing this e-mail
*** Options: http://mailman.ecs.soton.ac.uk/mailman/listinfo/eprints-tech
*** Archive: http://www.eprints.org/tech.php/
*** EPrints community wiki: http://wiki.eprints.org/
*** EPrints developers Forum: http://forum.eprints.org/