OutOfMemory Error in commit

Question

Greetings (Kodo3.4.1, JDK1.4, Oracle9i, Weblogic8.1)
I have a case where an OutOfMemory Error occurs during
PersistenceManager.currentTransaction().commit(). The problem is that,
although commit() throws this Error, the database transaction appears to
have already been committed.
Do you have any advice regarding what I should do in this case? I am
thinking that if OutOfMemory error occurs during commit, then it's possible
that the Kodo caches may be out of sync with the database itself, therefore
I should exit the JVM. The state of the records in the database should be
ok. What do you reckon?
Unfortunately there is no stack trace available, probably due to the Java
shortcoming as detailed in
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4753347
Cheers and thanks
.droo.

Answer 1

Thanks Henning - My thinking was that under most circumstances only the
thread which received the OOM Error would need to die, assuming threads do
not share any objects. This means that it would be ok to recover from an
OOM which happens anywhere, EXCEPT: inside commit() after database has been
updated but possibly while Kodo is updating it's internally shared objects -
I don't know for sure as I don't have the source code and I don't know what
it's doing when it dies. But... You may be right, even the threading
system itself or the container may have cactus shared objects after an OOM.
I will try to get hold of CR282897.
Also thanks Denis, I am concerned with shared DataCache/QueryCaches but also
was wanting to find out from Kodo what else I should be concerned about, as
I don't know what commit is trying to do when it dies.
In any case I have decided that if this OOM happens, I will disable the
current application. Anyone who calls a SessionBean will be to go away
because the server is experiencing difficulties, and it will try to notify
the administrators. Hopefully I don't need to kill the VM because other
applications may also be running in the same app server instance.
Cheers and thanks
.droo.
On 21/9/06 7:49 AM, in article [email protected], "Henning
Andersen" <Henning Andersen> wrote:
Drew,
Out Of Memory Errors normally cannot be recovered from. If the VM is near an
out of memory error situation, the exception can be thrown in any thread, not
just your kodo-comitting-thread. This exception can happen almost anywhere and
nobody codes to handle it properly (for instance, most of java.util does not
handle this properly, eg. TreeMap.put will leave treeset inconsistent if it
cannot allocate an Entry). This means that if an out of memory error occurs
you really should restart the vm.
Only solution is to ensure that you do not get out of memory errors.
I saw a similar problem (but not the same) on kodo 3.4.1, which had to do with
creating many new objects and doing flushes in between, which would cause an
internal list to grow in the DataCacheStoreManager. The fix for this is in
CR282897, which I would recommend trying to get a hold of through support.On 20/9/06 8:51 PM, in article [email protected], "Denis
Sukhoroslov" <[email protected]> wrote:
Drew,
if you concerned on cache sync with DB, you can easly clear it in case of
OOM error:
KodoPersistenceManager pm = ...// get PM
DataCache dCache =
pm.getConfiguration().getDataCacheManager().getDataCache();
if (dCache != null) {
dCache.clear();
Denis.