ocsinventory-server-2.0.5 session.pm breaks mysql replication of ocsweb db

[Pieter-Jan Gunst]

package Apache::Ocsinventory::Server::Communication::Session

if the below code is executed...
> # To avoid race conditions
> if( !$dbh->do("INSERT INTO engine_mutex(NAME, PID, TAG) VALUES('SESSION',?,'CLEAN')", {}, $$) ){
> &_log(315,'session',"already handled") if $ENV{'OCS_OPT_LOGLEVEL'};
> return;

and replicated to a slave db, it will also attempt to execute this query. as there's a primary key on name and tag, it would be safer to check whether we already have any records with that name and tag. Here's how I worked around the issue:

< my $check_clean_exists = $dbh->prepare('SELECT * from engine_mutex where NAME="SESSION" AND TAG="CLEAN"');
< if($check_clean_exists->execute() && !$check_clean_exists->rows()){
< $dbh->do('DELETE FROM engine_mutex WHERE NAME="SESSION" AND TAG="CLEAN"', {}, $$ );
< &_log(312,'session', "deleted") if $ENV{'OCS_OPT_LOGLEVEL'};
< $dbh->do("INSERT INTO engine_mutex(NAME, PID, TAG) VALUES('SESSION',?,'CLEAN')", {}, $$);
< &_log(313,'session', "started") if $ENV{'OCS_OPT_LOGLEVEL'};

you could as well just log and return, but I leave that up to you ocs experts.

[mortheres (mortheres)]

Hi,

I was not able to reproduce the problem using a MySQL slave server. Can you tell me on which case the problem happens ? On most of the computers inventories ? In standard case, the "INSERT INTO" request should only be executed on master server not on the slave.

Kind regards,

--
Guillaume