Question

We are using Hibernate 3.6.1.FINAL against Oracle 11g.

We are experiencing blocking locks in our database that I believe is caused by an unexpected side-effect of Hibernate.

This is a critical issue we need to resolve, as the database blocking locks quickly snowball and transactions begin timing out, causing a production outage.

I am trying to figure out if there is a certain scenario in which Hibernate can do a behind-the-scenes delete and re-insert of a collection of children, or some way that it could cause blocking locks on a delete/insert.

I’m looking for ideas/suggestions/insight about possible causes, not necessarily an exact answer.

I’ve isolated the entities involved and their mappings. Unfortunately due to the fact that we have a huge legacy system with tons of code, I had to come up with a simplified example:

Parent.hbm.xml

// Parent has a unique ID generated with a Sequence,
// an index column called ORD, and a Version column called UPDATE_COUNTER.
<!-- bi-directional one-to-many association to Child -->
<bag
    name="children"
    lazy="false"
    inverse="true"
    cascade="all-delete-orphan"
    order-by="ord asc">
    <key column="PARENT_ID" />
    <one-to-many class="Child" />
</bag>

Child.hbm.xml

// Child has a unique ID generated with a Sequence, 
// and a Version column called UPDATE_COUNTER.
<!-- bi-directional many-to-one association to Parent -->
<many-to-one
    name="parent"
    class="Parent"
    not-null="true"
>
    <column name="PARENT_ID" />
</many-to-one>

Our DBAs have reported that they see the following:

The Blocked SQL:

insert into CHILD (UPDATE_COUNTER, ORD, PARENT_ID, ID) values (:1 , :2 , :3 , :4)

The Blocking SQL:

delete from CHILD where ID=:1  and UPDATE_COUNTER=:2

Here’s what I’ve tried thus far:

1. Did a full audit of our codebase, established that there is nowhere where we delete and insert a row with the same primary key. All of our INSERT for Child is done with ID set to null, in which case the Sequence will generate the Unique ID.

2. Turned on show_sql. Also, DBAs set up a trigger on Child to capture all DML activity so we could get the bind variables in development.

   I found the following articles saying that Hibernate could do delete all then re-insert for bags and lists:

   Why Hibernate does “Delete all then re-insert”

   Hibernate – set, map, bag & list differences

   However, I determined that this is not occurring. I wrote test code that removes one of the Child objects in the middle (ORD=5 with a collection of 10 children that had ORD=0 through ORD=9) and then saved Parent.

   I saw in the DML activity table that a delete was issued for the removed Child and UPDATE was done for the remaining 9 children that set their ORD to the correct value.

As I mentioned, I’m not looking for an exact answer. I would just be tremendously grateful if anyone who has a lot of experience with Hibernate could lend any insight or suggestions as to additional paths or avenues I could run down.

Thank you very much in advance.

Answer 1

You may want to consider adopting Pessimistic Locking. Once done… Hibernate will always use the locking mechanism of the database; it never lock objects in memory.