develooper Front page | perl.dbi.users | Postings from January 2021

DBD::Oracle insert speed

Thread Next
From:
Peter Meszaros
Date:
January 10, 2021 00:23
Subject:
DBD::Oracle insert speed
Message ID:
CABMJSUOiWqtsW04UBtK=uOQoYOcvL127K2st8SYYVMcPLTnQOg@mail.gmail.com
Hi,
I have already asked this question at perlmonks.org but I did not get
proper answer.

Module DBD::Oracle <http://search.cpan.org/perldoc?DBD%3A%3AOracle> has
serious performance degradation at inserting rows. Except execute_array (or
bind_param_array), all other insert solutions are pretty slow (eg. bind by
name). DBD::Oracle use its own wrapper (dbimp.c, oci8.c) and seem to be
implemented in an ineffective way. Comparing to otlv4 (
http://otl.sourceforge.net/otl3_intro.htm) which is a wrapper over Oracle's
OCI interface written in C++ the basic
http://otl.sourceforge.net/otl3_ex10.htm is much (~10 times) faster than
DBD::Oracle row based insertion. DBD::Oracle implements DATA_AT_EXEC mode
with OCIBindDynamic for row based insertion (not array) as can be seen
below and this block is repeated for each value.

...
dbd_bind_ph(1): bind :id <== '1774684594' (type 0 (DEFAULT (varchar)))
dbd_rebind_ph() (1): rebinding :id as '17746...' (not-utf8, ftype 1
(VARCHAR), csid 0, csform 0(0), inout 0)
dbd_rebind_ph_char() (1): bind :id <== '17746...' (size 10/12/0, ptype
3(VARCHAR), otype 1 )
Changing maxlen to 12
dbd_rebind_ph_char() (2): bind :id <== '17746...' (size 10/12, otype
1(VARCHAR), indp 0, at_exec 1)
      bind :id as ftype 1 (VARCHAR)
    OCIBindByName(163ad08,1675e58,1614d20,":id",placeh_len=3,value_p=1675a00,value_sz=12,dty=1,indp=1675e78,alenp=0,rcodep=1675e70,maxarr_len=0,curelep=0
(*=0),mode=DATA_AT_EXEC,2)=SUCCESS
    OCIBindDynamic(163a970,1614d20,1675e20,7fdf910dacc0,1675e20,7fdf910daf30)=SUCCESS
    OCIAttrGet(163a970,OCI_HTYPE_BIND,1675e3c,0,31,1614d20)=SUCCESS
dbd_rebind_ph(): bind :id <== '17746...' (in, not-utf8, csid
873->0->873, ftype 1 (VARCHAR), csform 0(0)->0(0), maxlen 12,
maxdata_size 0)
    OCIAttrSet(163a970,OCI_HTYPE_BIND, 7ffda0f7f960,0,Attr=31,1614d20)=SUCCESS
...
[download] <https://perlmonks.org/?displaytype=displaycode;abspart=1;part=1;node_id=11126161>

Is this design intentional? OCI documentation said that OCIBindDynamic can
be useful at working with big data items
https://web.stanford.edu/dept/itss/docs/oracle/10gR2/appdev.102/b14250/oci05bnd.htm#i427753.
C++ code using otlv4 can process blob/clob data and it does not use
OCIBindDynamic at all. So, why DBD::Oracle implemented this way?

Thanks in advance

Thread Next


nntp.perl.org: Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at ask@perl.org | Group listing | About