Hi all.
I have a very peculiar problem that I’m not remotely capable of solving
(or
so it seems…): my application uses Oracle 10g database and, for safety
and
manageability reasons, has to use an user that is not the owner of the
production schema - and this policy is set in stone here, no way to
change
it (not that I would want it anyway, since I agree with it and we’ve
never
had any kind of trouble before).
To make things easier, the DBA created a logon trigger to the app’s user
(let’s suppose the name of the app user is “my_app_user”) that does
something like this:
alter session set current_schema = ‘app_production_schema’
So, when the application logs on to the database using ‘my_app_user’,
that
trigger makes the tables, views and sequences of ‘app_production_schema’
readily available to ‘my_app_user’ without the need to use stuff like
set_table_name = ‘app_production_schema.my_table’ in every model.
Now, here’s the catch: it’s been working like a charm with MRI and the
activerecord-oracle-adapter-1.0.0.9250 - no problem at all, as everyone
would expect. But, when I try to run the application in JRuby 1.2.0 +
activerecord-jdbc-adapter-0.9.1, I get messages like “Table ORGAOS does
not
exist” for all tables and views of the application!
And things get worse - here’s my situation: I’ve had just “fallen in
love”
with Ruby on Rails when I was designated to “convert” this application
(already in production for quite some time) to JRuby, since my
organisation
(I work for government, by the way), for now, can’t handle managing more
than one kind of app server or deployment environment because of the
tiny
size of the team running the everyday management duties related to
server
machines, physical and logical network config, and app servers.
I had to learn some RoR, then JRuby (and the “ecosystem” around it) and
finaly managed to make everything work in development environment (keep
in
mind I don’t know jack of Java language and didn’t know anything about
Tomcat and the like when the project started).
Problem is, the homologation (and permission to use in future projects)
of
the whole (J)Ruby on Rails platform is pending on the success of this
particular “conversion” project, since this court doesn’t want to keep
the
Apache + mod_rails + Mongrel + Webistrano setup - meaning the app has to
run, at the least, in Tomcat using JNDI. But I suppose it is not
possible if
I can’t see any tables on the production schema when I run JRuby +
Mongrel… And the deadline to deliver the application with some load
and
stress tests is this March 25 already.
There are already other government organisations in my country very
interested in the results of this project, since they want to experiment
with JRuby on Rails but they would like to not waste the same
(inordinate)
amount of time I did to get around the pitfalls of this endeavour.
When this problem first appeared, I was using JRuby 1.1.6,
activerecord-jdbc-adapter-0.8.2 and Ruby on Rails 1.2.3. Just to be sure
I
was not being the victim of obsolete stuff, I gave myself the trouble of
converting this app to RoR 2.1.2 so I could plug JRuby 1.2.0RC1 and the
“trunk” of activerecord-jdbc-adapter from GitHub (version 0.9.1 had not
yet
been released…). And the problem persisted.
After that, I did some runs with ar-jdbc-adapter 0.8.2, 0.9 and 0.9.1,
all
with the same results. Then, I removed the plugins of the application
(classic_pagination, acts_as_list and the latest version of
acts_as_paranoid) just to be sure they weren’t getting in the way (I’ve
had
some bad experience with redhillonrails-core plugin provoking similar
bugs)
and did the same three runs - getting the same results. Since my machine
is
hideously slow (can’t even use and IDE), it took a hell of a lot of time
just to do that. Finally, after looking at the stack trace and at the
code
involved, I finally gave up - I don’t know jack of Java (I did look at
JdbcAdapterInternalService.java
and its columns_internal method), my knowledge about the guts of the
Oracle
Database Engine is non-existent and my knowledge of the internals of the
RoR
framework is kinda small yet. And I don’t have the mental stamina or
time to
keep going at it anymore since I still must get other projects rolling
at
the same time.
Right now, I’m near desperate and kinda sorry for bothering you with
such a
long post, but I didn’t know what to do. I fear a complete fiasco
because of
this bug, and people around the whole country did use this application
already to great success and satisfaction (running in MRI and Rails
1.2.6,
of course) and I fear thousands of users will lose this application -
that
will have to be converted to Java, after only God knows how much time,
since
it is quite big for a RoR app (and will get even bigger in Java, I’m
sure ;p
). And, while my official deadline is March 25, I have to plan and
execute
load and stress tests and, since it is the first time I do that (do you
notice a sad trend here?), I think I would need a solution by March 23.
Me thinks it’s about time I present you a couple stack traces (without
any
plugins in the application already):
first exception: Table ORGAOS does not exist (ORGAOS is a view in fact)
D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1149:in
columns' D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1157:incolumns_hash’
D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1375:in
find_one' D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1366:infind_from_ids’
D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:541:in
find' app/controllers/adm_controller.rb:28:incge’
:1:in `initialize’
second exception: Table QUESTIONARIOS does not exist (QUESTIONARIOS is
a normal table)
D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1149:in
columns' D:/RDev/interpreters/jruby/releases/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-2.1.2/lib/active_record/base.rb:1162:incolumn_names’
(DELEGATION):2:in column_names' app/controllers/questionarios_controller.rb:30:inlist’
:1:in `initialize’
Note: I tried to use set_table_name “<schema_name>.orgaos” and
set_table_name “<schema_name>.questionarios” and got the exact same
results. Again, that problem doesn’t occur with MRI +
activerecord-oracle-adapter.
I think it is somewhat disturbing the fact that
activerecord-jdbc-adapter doesn’t appear in any of those stack-traces,
although I could be very wrong. And both of them end in the method
‘columns’ of ActiveRecord::Base.
A very small clue that I got while examining
activerecord-oracle-adapter whas this: it treats the names of synonyms
in a much different way than it treats the name of other objects. Take
a look at this code:
def columns(table_name, name = nil) #:nodoc:
(owner, table_name) = connection.describe(table_name)
…lots of other code…
end
def describe(name)
@desc ||= @@env.alloc(OCIDescribe)
@desc.attrSet(OCI_ATTR_DESC_PUBLIC, -1) if VERSION >= ‘0.1.14’
@desc.describeAny(@svc, name.to_s, OCI_PTYPE_UNK) rescue raise
%Q{“DESC #{name}” failed; does it exist?}
info = @desc.attrGet(OCI_ATTR_PARAM)
case info.attrGet(OCI_ATTR_PTYPE)
when OCI_PTYPE_TABLE, OCI_PTYPE_VIEW
owner = info.attrGet(OCI_ATTR_OBJ_SCHEMA)
table_name = info.attrGet(OCI_ATTR_OBJ_NAME)
[owner, table_name]
when OCI_PTYPE_SYN
schema = info.attrGet(OCI_ATTR_SCHEMA_NAME)
name = info.attrGet(OCI_ATTR_NAME)
describe(schema + '.' + name)
else raise %Q{“DESC #{name}” failed; not a table or view.}
end
end
Could it be possible that, behind the stage, the Oracle Database
Engine does the “magic” of alter session set current_schema =
“prod_schema” automatically creating a lot of synonyms for my user? If
that is the case (and I’m wild guessing here, I don’t know jack of the
Oracle engine), could the problem be that ar-jdbc doesn’t
retrieve/treat the names differently like ar-oracle-adapter? Or
ar-jdbc reads meta-data in a place different form where ar-oracle does
- and the app user doesn’t have the necessary permissions?
Last, there is a very minor bug and a suggestion to be made - first,
the minor bug: I thought the following lines weren’t needed in
environment.rb if I’m using Rails 2.1.2:
if RUBY_PLATFORM =~ /java/
require 'rubygems'
gem 'activerecord-jdbc-adapter', '0.9.1'
require 'jdbc_adapter'
end
But I have to keep them, or I get the “uninitialized constant
ActiveRecord::VERSION (NameError)” exception - stack trace follows:
/jruby-1.2.0/lib/ruby/gems/1.8/gems/activesupport-2.1.2/lib/active_support/dependencies.rb:279:in
load_missing_constant': uninitialized constant ActiveRecord::VERSION (NameError) from /jruby-1.2.0/lib/ruby/gems/1.8/gems/activesupport-2.1.2/lib/active_support/dependencies.rb:468:inconst_missing’
from
/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-jdbc-adapter-0.9.1/lib/active_record/connection_adapters/jdbc_adapter.rb:14
from
/jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-jdbc-adapter-0.9.1/lib/active_record/connection_adapters/jdbc_adapter.rb:31:in
require' from /jruby-1.2.0/bin/../lib/ruby/site_ruby/1.8/rubygems/custom_require.rb:31:inrequire’
from
/jruby-1.2.0/lib/ruby/gems/1.8/gems/activesupport-2.1.2/lib/active_support/dependencies.rb:510:in
require' from /jruby-1.2.0/lib/ruby/gems/1.8/gems/activesupport-2.1.2/lib/active_support/dependencies.rb:355:innew_constants_in’
from
/jruby-1.2.0/lib/ruby/gems/1.8/gems/activesupport-2.1.2/lib/active_support/dependencies.rb:510:in
require' from /jruby-1.2.0/lib/ruby/gems/1.8/gems/activerecord-jdbc-adapter-0.9.1/lib/active_record/connection_adapters/oracle_adapter.rb:1 ... 36 levels... from /jruby-1.2.0/lib/ruby/gems/1.8/gems/rails-2.1.2/lib/commands/server.rb:31:inrequire’
from
/jruby-1.2.0/bin/…/lib/ruby/site_ruby/1.8/rubygems/custom_require.rb:31:in
`require’
from script\server:3
And, finally, the suggestion (feature request): could we have an easy,
quick “schema” parameter in database.yml? Something like:
homologation:
adapter: oracle
host:
port: <usually 1521>
database:
schema:
username:
password:
Since I come from Delphi, I really can’t get over the fact that I
can’t just configure my working schema with whatever user in a quick &
easy way like I used to (I used the Direct Oracle Access components
and I could config user_name, pass, connection string - or
TNSNAMES.ORA alias - and working schema on my OracleSession object).
We, Oracle users, are not to blame if Oracle uses a database address (
a SID) and a schema address on top of that… =)
Sorry for the terribly long post. I really appreciate the attention
and, if I need to give any extra details or do any further
investigation, I’ll gladly do it - and I’ll be lurking my Inbox this
night…
-Renato.