ORA-04068 on 11g RAC does not resolve automatically !

Question

Did you ever get the combination of these three ORAs?
ORA-04068: Existing state of package has been discarded
ORA-04065: Not executed, altered or dropped stored procedure <PROCEDURE_NAME>
ORA-06508: PL/SQL: Could not find program unit being called: <PROCEDURE_NAME>
*1st I checked for invalid objects in the database, there are none!*
*2nd Then I checked the metalink note 1058873.1 that describes a similar error:*
In hecheck output, we see that the objects affected are listed as "Dependency$ p_timestamp > > mismatch for VALID objects".
Also, the following script can be used for finding the mismatches:
set pagesize 10000
column d_name format a20
column p_name format a20
select do.obj# d_obj,do.name d_name, do.type# d_type,
po.obj# p_obj,po.name p_name,
to_char(p_timestamp,'DD-MON-YYYY HH24:MI:SS') "P_Timestamp",
to_char(po.stime ,'DD-MON-YYYY HH24:MI:SS') "STIME",
decode(sign(po.stime-p_timestamp),0,'SAME','*DIFFER*') X
from sys.obj$ do, sys.dependency$ d, sys.obj$ po
where P_OBJ#=po.obj#(+)
and D_OBJ#=do.obj#
and do.status=1 /*dependent is valid*/
and po.status=1 /*parent is valid*/
and po.stime!=p_timestamp /*parent timestamp not match*/
order by 2,1;However, the query returned 0 rows.
*3rd Then I read about similar errors on http://oraclequirks.blogspot.com/2007/03/ora-04061-existing-state-of-package-has.html*
The article describes the state of package being invalidated when one user calls a procedure in a package that has a global variable and changes the value of the global variable and then when a developer compiles that package, another user will get the error ORA-04061: existing state of package has been invalidated the first time he calls the procedure, but the second time it will work without returning errors.
However, my error is persistant and Oracle can't resolve it automatically.
Here's how the error occurs:
- I have a PACKAGE that has a global variable and procedures
- User1 connects to Instance1 and calls a procedure inside the PACKAGE. The procedure sets the value of the global variable by calling a function outside the package.
For example: L_GLOBAL_VAR := OUTSIDE_FUNCTION(30);
- The developer compiles the PACKAGE
- User1 can continue working normaly - he receives the 1st error by Oracle informing him that the value of the global variable has changed (normal behaviour), but the second time he calls the procedure inside the PACKAGE, no errors are reported.
- User2 connects to the second instance of the RAC (Instance2)
- When User2 tries to call the same procedure, he constantly gets ORA errors (ORA-04068, ORA-04065, ORA-06508), like this:
ORA-04068: Existing state of package has been discarded
ORA-04065: Not executed, altered or dropped stored procedure "OUTSIDE_FUNCTION"
ORA-06508: PL/SQL: Could not find program unit being called: "OUTSIDE_FUNCTION"
ORA-06512: at "PACKAGE2", line 1920
ORA-06512: at "PACKAGE2", line 265The PACKAGE2 has a procedure which is being called at line 265, and that procedure calls the function OUTSIDE_FUNCTION at line 1920.
Symptoms:
- User2 cannot correct the error by repeatingly calling the PACKAGE procedure.
- The developer cannot resolve the issue by recompiling the PACKAGE.
- The developer can only temporarily resolve the issue by recompiling the OUTSIDE_FUNCTION. After doing it, the User2 does not receive errors. However, if the User2 reconnects, the error returns.
- The solution is to recompile the PACKAGE2. After doing it, the User2 does not receive errors.
However, I'm totally confused by this Oracle behaviour.
I told the developers that they shouldn't compile packages while there are users working on the database (a possible solution), but as a developer I know that is very very hard on a 24/7 production database, especially if there are some urgent bugs to resolve etc.
We were working on a single instance Oracle 9i database before upgrading to Oracle 11g RAC (11.2.0.1.0), and we never had similar problems.
I would be very grateful for any suggestions regarding this issue.
Thanks,
Daniel

Answer 1

Hi P. Forstmann,
Thanks for your answer. I've read about the edition based redefinition but if I understand it correctly, that's not exactly what I'm looking for.
For example, if a connected user reports a bug that prevents her from doing her work, the developers should resolve that bug ASAP and recompile the objects in question, which should affect the connected users.