Audit changes: differentiate between user change and app change

I learned an important lesson over the last few days.

You all probably know this already, but as you may also know I am generally not reticent to expose my relative ignorance.

So I follow a standard of adding four audit columns to my tables, populated by triggers, which keep track of who inserted/updated the row, and when:


I certainly did this for the qdb_users table, which is the users table for the PL/SQL Challenge and Oracle Dev Gym.

So far, so good.

But recently a player complained that she was not receiving emails with results of her quizzes. I checked and found that the preference was turned off. Had she modified her user profile lately or had my code done something to her row?

It was pretty much impossible to tell, because we keep track of the user's last visit to the site - which means the app itself updates the qdb_users.changed_by/on columns every time a user comes to the site. This would overwrite whatever the user's last changed_on value was.

Yuch.

I was using a "low level" audit column to also keep track of user-level behavior, with the result being a loss of information. 

So I added a new column (changed_on_by_user), which is updated only when the user executes an action that updates his or her profile - all controlled through my PL/SQL API:



Lesson learned: don't mix system information (row-level audit information) and application/user information. Keep them separate, making it much easier (possible!) to track activity within your application!

Comments

  1. I find these four columns often but I doubt the usefulness. You can only see the last change and you don't know what has been changed.

    We use two different approaches: for entities where you need the audit for security reasons and/or only some out of many attributes will be updated we have [entity]_hist table that saves which user changed which attribute (together with old/new value and other facts). For entities where we regularily need to query a past state, the whole row is stored in an [entit]_hist table or in the table itself with valid from/to columns.

    This might need a lot more storage but you can be sure that you know who changed what and when.

    Marcus

    ReplyDelete
  2. Marcus - yes, absolutely right. This is just about the most minimal auditing you can do, and does not reveal what was changed. Seems like this approach could be nicely templated and then generated for a given table. Care to write such a table for the oddgen project? oddgen.org

    ReplyDelete
  3. Hello All,

    In our "advanced era", why not using the FLASHBACK ARCHIVE feature ?

    As by Murphy's laws ...
    auditing is a feature that you either invest a lot of time
    in implementing it and nobody will ever need its results,
    or you don't implement it and then everybody suddenly needs it.

    With the FLASHBACK ARCHIVE feature, it is only a matter of
    space consumption. In 12c it also can use compression for more
    optimal space usage.
    So, I think that this feature is very convenient.

    Cheers & Best Regards,
    Iudith

    ReplyDelete
  4. Thanks, Iudith. Yes, I certainly should have mentioned the flashback archive feature! I wouldn't want to only propose this, since as you point out there is an issue of space consumption, and therefore likely limits on the amount of audit information that will be preserved. Plus, this is something that can be done by developers without having to work it through with their DBAs.

    ReplyDelete

Post a Comment

Popular posts from this blog

Table Functions, Part 1: Introduction and Exploration

Recommendations for unit testing PL/SQL programs

The future of Oracle PL/SQL: some thoughts on Sten Vesterli's thoughts