Conference eps::oracle

1573.0. "OPS + MTS problem" by TRNOIS::ALMONDO () Thu Jun 05 1997 06:43

    
    Hi all,
    
    We have a proble moving an Oracle DB to OPS at customer site, the
    clients of the DB connects the database using MTS and it seem that 
    when a numebr of clients greather tha the number of servers try to 
    modify the db, then all the servers hangs (on any kind of access,
    also read of an "untouched" table!) until some of the updating client
    issue a rollback or a commit.
    
    The problem is easily reproducible:
    
    * set the number of server for MTS to 1,
    * connect from one terminal using sqlplus and issue am update
      statement on a table, example: delete from m;
    * connect from a second terminal and try to read anything from
      a different table, example:  select * from cat;
    
    The second sqlplus will hung until the first issue a commit or exits
    with rollback                         
    
    This happen only whit OPS.
    
    Oracle answered to this saying that the problem is due to our DLM
    that use a "process based" locking istead of a "session based"
    locking (whatever it means!!).... and this will be fixed in oracle 8
    that has its own DLM.
    They report also that this was already seen on Sun but there has been
    fixed bya a new issue of Sun's DLM.
    
    Do anyone know if there is a workaround/patch or anything else that
    fix this problem?
    
    We are using Digital UNIX V4.0B TCR 1.4A Oracle 7.3.2.3 OPS.
    
    Regards
    Mario Almondo
    Torino - Italy
    

What is your GC_LOCK definitions? Are you using same locks for both the tables?
BTW, I don't recall anybody recommends the use of MTS this days. I wouldn't
surprise if I found that OPS/DLM has never been tested (qual-ed) in MTS env.

/KK/