Quelles sont les causes et les solutions pour les erreurs de table muter?

Question

Je comprends les erreurs de table muter sont causées par un défaut de conception ou de la requête problématique.

Une vieille requête a été récemment mise en production qui jette une erreur de table muter. Notre DBA a résolu le problème, mais nous ne savons pas comment.

Quelles sont les causes exactement les erreurs de table muter et comment serait notre DBA a résolu le problème?

Answer 1

The most likely cause of a mutating table error is the misuse of triggers. Here is a typical example:

1. you insert a row in table A
2. a trigger on table A (for each row) executes a query on table A, for example to compute a summary column
3. Oracle throws an ORA-04091: table A is mutating, trigger/function may not see it

This is an expected and normal behaviour, Oracle wants to protect you from yourself since Oracle guarantees:

• (i) that each statement is atomic (i.e will either fail or succeed completely)
• (ii) that each statement sees a consistent view of the data

Most likely when you write this kind of trigger you would expect the query (2) to see the row inserted on (1). This would be in contradiction with both points above since the update is not finished yet (there could be more rows to be inserted).

Oracle could return the result consistent with a point in time just before the beginning of the statement but from most of the examples I have seen that try to implement this logic, people see a multi-row statement as a serie of successive steps and expect the statement [2] to see the changes made by the previous steps. Oracle can not return the expected result and therefore throws the error.

For further reading: "mutating table" on Ask Tom.

If as I suspect the cause of the mutating table error is a trigger, one way to avoid the error is to move the logic out of the trigger into procedures.

Answer 2

A mutating table occurs when a statement causes a trigger to fire and that trigger references the table that caused the trigger. The best way to avoid such problems is to not use triggers, but I suspect the DBA didn’t take the time to do that. He could have done one of the following:

• Changed the trigger to an after trigger.
• Changed it from a row level trigger to a statement level trigger.
• Convert to a Compound Trigger.
• Modified the structure of the triggers to use a combination of row and statement level triggers.
• Made the trigger autonomous with a commit in it.