The scenario that created this question:
We have a package that is a dependency of another package, sometimes making changes to the "parent" package causes the dependent package to become invalid, but sometimes it doesn't.
It has caught us by surprise before.
It would be very useful to simply understand what causes invalidation so I could predict/plan for it.
Changing anything object that a package relies upon (e.g. tables, views, triggers, other packages) will automatically mark the package as invalid. As tuinstoel notes above, Oracle is smart enough to recompile the package when it is first used.
If you are concerned about this, every time you make schema changes (e.g. tables, views, triggers, procedures), run a DBMS_UTILITY.compile_schema (or have your DBA do it). This will force compile all the packages and let you know where, or if, there are errors before you find them the hard way.
Or you can query the following table to see what dependencies you have
select *
from dba_dependencies
where name = 'YOUR_PACKAGE'
and referenced_owner = 'ANYUSER' --- Comment this out if you are looking for yourself
and owner = USER --- Or can be set to any user
This will show all dependencies. For your objects query user_dependencies.
I agree with Thomas Jones-Low however there are a couple more issues to do with long sessions and recompilation.
If you reference a package in a session and that package (or a dependant package) gets recompiled during the same session then you'll get oracle error "ORA-06508: PL/SQL: could not find program unit being called"
Once you've referenced the package in a session you generally can't change the package without invalidating it for that session. This is a particular problem for development environments where packages change frequently but also a problem for production environments where you want to do a small patch without taking the whole environment down. Note that this error will occur even when there are no errors in the changed packages.
BTW, If I'm completely wrong about the situation... apologies in advance
Caught by surprise?
Not sure what the implications of that are...
Did something break in production?
What EXACTLY happened?
The reason I ask is because understanding every possible change's ramifications is much harder than dealing with the outcome. Why the invalidation become an issue? My guess is because you got an "Existing state of Package has been discarded" error in your application. Is that the REAL issue?
Again I suspect that it is and if so, let's just deal with that instead of the list of changes which as I put in a comment is version specific. (11g tracks dependency down to the column of a table instead of the table as a whole for example).
This may not seem like an important error to you if you're not using package state. If you were this would be an important error and you wouldn't have been surprised, so I'm guessing you're not.
Since you are not this error is ok to ignore. Since you can safely ignore it, you can code your client app to ignore this error and retry your call, because, as others have pointed out Oracle will recompile your package for you. This is a worthwhile exercise. Because rather than knowing every possible thing you need to worry about when you make a change, and then in the emergency fix you forget one of those, your app will just handle it and move on, without worry.
In addition to Thomas Jones-Low's answer, if you only modify the package BODY, a dependent object might not be marked as invalid.
However, as soon as you modify the package specification, that is bound to happen.
If try to execute an invalid Oracle package, Oracle will try to compile it. Only when it remains invalid after compiling Oracle will throw an exception.
Related
I have some questions regarding the registering/updating of 3rd party plugins that were previously loaded via a managed solution by a 3rd party.
The issue we are having is that they(3rd party) sent us a plugin update and a new plugin outside of the Managed Solution and had us register it manually though the registration tool. Then, next time we tried to import a later version of their solution, the Managed solution import failed. We eventually realized that there where duplicate rows in the pluginassembly and pluginassemblytype table that had the same Pluginassemblyid and plugintypeid respectively with different solutionids.
These solutionids were "Active" which I assume came from the manual registration and "IPM Global" which is our 3rd Party Managed Solution. The only way we were successful in getting the solution to import was to change the overwrite time
on the table(s) to 0 and then delete the "Active" pluginassembly and plugintype records.
Is there any other way to accomplish this same thing that is supported?
BTW. We did try to unregister the plugins before trying this, but there were too many dependencies in our workflows.
Wow, this is a thorny problem. Since you mention updating the tables directly, I'll assume that the system is on-prem.
Registering a plugin that exists in a managed solution outside of that managed solution is something I've never done, and while I have directly updated the plugin registration table, it is certainly something to minimize.
As unpleasant as it sounds, to get back to a good state in a supported way you may be looking at having to:
Backup the SQL database
Backup all the data from any managed solution entities.
Undo all dependencies on the managed solution (i.e. edit all the workflows so they no longer depend on the managed solution). To ease the pain of this piece you might want to experiment with exporting the affected workflows via an unmanaged solution. Then you could delete them rather than trying to weed out the dependencies. Then after you have the managed solution back in the system, you could theoretically import the unmanaged workflow solution to restore the workflow. But, admittedly this working depends on workflows finding the plugin assemblies they depend on by name rather than Id, which I'm not sure is the case - so like I said, experiment.
Unregister the "out-of-band" plugin
Uninstall the managed solution
Install a clean copy of the managed solution, INCLUDING the previously problematic plugin.
Restore/reconfigure the workflows
Restore the managed entities data
It's a lot... so much in fact that I would consider opening a Microsoft support ticket to see if they can provide any alternative methods to correct the situation.
In this situation I personally might also consider unsupported methods like using SQL to copy the tables of any managed entities before deleting the managed solution and then using SQL to copy the data back after the managed solution is fixed. Of course I (almost) never recommend using SQL in an unsupported way, so explore that option at your own risk (and with copious backups).
First, try to avoid direct DB updates in system tables whenever possible. You never know when it will hit you (next solution import, next CRM upgrade, moving to cloud, etc).
I assume that yours vendor solution contains entities and attributes and not only assemblies with SDK message processing steps. Thus you can't just simply remove that managed solution cause there will be data loss. Also I assume there are no workflow activities in their assemblies.
Ask them for solution with properly registered assemblies and SDK message processing steps. Then go into your organization with plugin registration tool (https://msdn.microsoft.com/en-us/library/gg309580.aspx) and unregister their assemblies. Then just import their latest solution. It should be able to import their assemblies with whatever is inside them.
It's good idea to restore copy of prod organization and playthrough whole process in safe environment first.
Whenever I Import I find lots of errors that since the database hasn't been worked with this tool, we did not have to fight case sensitive issues, nor views with unreferenced tables and a lot of other stuff that didn't worry us when we were on DevEnv.
Beacuse of this, I resolved to try working somewhat disconnected as a transitional state. but without loosing the TFS code repo or issue tracker.
Is it possible to work on sqlprojects and be able to validate and deploy code without having to import the database into the proj?
If it is, how?
#Edit:
Ok, so I've been asked to maintain the self DB reference inside the queries from within SP's, since DTSs execute some of those SP's, and it should evaluate the references without error.
(I'll be posting more stuff.)
The scenario that created this question:
We have a package that is a dependency of another package, sometimes making changes to the "parent" package causes the dependent package to become invalid, but sometimes it doesn't.
It has caught us by surprise before.
It would be very useful to simply understand what causes invalidation so I could predict/plan for it.
Changing anything object that a package relies upon (e.g. tables, views, triggers, other packages) will automatically mark the package as invalid. As tuinstoel notes above, Oracle is smart enough to recompile the package when it is first used.
If you are concerned about this, every time you make schema changes (e.g. tables, views, triggers, procedures), run a DBMS_UTILITY.compile_schema (or have your DBA do it). This will force compile all the packages and let you know where, or if, there are errors before you find them the hard way.
Or you can query the following table to see what dependencies you have
select *
from dba_dependencies
where name = 'YOUR_PACKAGE'
and referenced_owner = 'ANYUSER' --- Comment this out if you are looking for yourself
and owner = USER --- Or can be set to any user
This will show all dependencies. For your objects query user_dependencies.
I agree with Thomas Jones-Low however there are a couple more issues to do with long sessions and recompilation.
If you reference a package in a session and that package (or a dependant package) gets recompiled during the same session then you'll get oracle error "ORA-06508: PL/SQL: could not find program unit being called"
Once you've referenced the package in a session you generally can't change the package without invalidating it for that session. This is a particular problem for development environments where packages change frequently but also a problem for production environments where you want to do a small patch without taking the whole environment down. Note that this error will occur even when there are no errors in the changed packages.
BTW, If I'm completely wrong about the situation... apologies in advance
Caught by surprise?
Not sure what the implications of that are...
Did something break in production?
What EXACTLY happened?
The reason I ask is because understanding every possible change's ramifications is much harder than dealing with the outcome. Why the invalidation become an issue? My guess is because you got an "Existing state of Package has been discarded" error in your application. Is that the REAL issue?
Again I suspect that it is and if so, let's just deal with that instead of the list of changes which as I put in a comment is version specific. (11g tracks dependency down to the column of a table instead of the table as a whole for example).
This may not seem like an important error to you if you're not using package state. If you were this would be an important error and you wouldn't have been surprised, so I'm guessing you're not.
Since you are not this error is ok to ignore. Since you can safely ignore it, you can code your client app to ignore this error and retry your call, because, as others have pointed out Oracle will recompile your package for you. This is a worthwhile exercise. Because rather than knowing every possible thing you need to worry about when you make a change, and then in the emergency fix you forget one of those, your app will just handle it and move on, without worry.
In addition to Thomas Jones-Low's answer, if you only modify the package BODY, a dependent object might not be marked as invalid.
However, as soon as you modify the package specification, that is bound to happen.
If try to execute an invalid Oracle package, Oracle will try to compile it. Only when it remains invalid after compiling Oracle will throw an exception.
Following is my oracle version
Oracle Database 11g Enterprise Edition Release 11.2.0.2.0 - 64bit Production
I was trying to alter a table to add a column but since it took long time, i cancelled it.
After sometime, i started seeing errors in all packages and views that referenced the unit.
Error message is
PLS-907: cannot load library unit (table name)
Googling on this error does not give much information. In one place, i saw that below command will help but it seems to me as very risky command
alter system flush shared_pool
Can anyone help me with the reason and solution to this?
Thanks,
Jeevan
it happens because of DB links are dropped and recreated for connections so just recompiling the package will resolve this and some times it automatically recompile it .
While changing the length of a varchar2 may resolve a problem in one procedure, for example, it can cause other objects to go invalid. Typically the solution for resolving the invalid object is just recompiling. Before recompiling - parameters, variables, etc need to be checked to ensure they accommodate for the change.
I'm fairly new with EF and got myself rather confused about what's going on with my solution.
I'm in the situation where my code appears to be working however the changes aren't being written to the database that I would expect.
I'm using Web Developer 2010 and SQL 2008, code first approach but choosing to make my own changes in the database and manually ensure my classes match correctly.
Things seemed ok until I came across an error where the db hash wasn't what entity framework was expecting, so I looked at modelBuilder.Conventions.Remove(); which wasn't available - it seems that's not around in the later versions. So, I figured if the later versions doesn't do this check, I'll go ahead and remove my reference to EF 4 and put the 4.3 dll in it's place. I think I also ended up deleting and renaming my database. That didn't work, so I followed up on something I found on Scott Gu's blog about naming your dbContext the same as the database, which seemed to help.
However I'm now getting the most bizarre scenario. My code is running, the data is saving, but it's not saving to my database. In fact I'm running a profile trace on the db and it doesn't even seem to be trying to connect. I can change my connection string to something invalid too, but my code will happily run, storing the data #somewhere#
Any ideas what's going on? Might it be using a local database or cache that I'm unaware of? Should I just start my small project again and pretend this never happened? That'd be the professional approach, right?
I would suggest to use Database first approach, if you have your Database set, or want to have maximum control over database.