[hibernate-dev] HHH-9701 - Develop "immutable EntityEntry" impl
John O'Hara
johara at redhat.com
Fri Apr 17 15:04:38 EDT 2015
Thanks for reviewing, I was unsure of the best way to test if the Entity
was referenced cached (or could be cached), I look at
org.hibernate.persister.entity.EntityPersister#canUseReferenceCacheEntries
instead.
Would it matter if the entity was Immutable and not cached? In that case
we would not be nulling the ref to the EntityEntry, but the Entity could
not be shared between sessions. Would the Entity object (and EntityEntry
object) be GC'd after a session was closed if the entity was not cached?
Is there the potential for a mem leak there?
I will also change the Exception name and cache the EntityEntryFactory
when the EntityPersister is constructed
On 17/04/15 19:55, Steve Ebersole wrote:
> I commented on your pull request.
>
> My main concern is how you determine "canClearEntityEntryReference".
> That needs to change. See my comment.
>
>
>
> On Mon, Apr 13, 2015 at 5:37 AM, John O'Hara <johara at redhat.com
> <mailto:johara at redhat.com>> wrote:
>
> I think there are two factors to consider, when the entity is
> marked as Immutable, and when the EntityEntry can be shared
> between sessions.
>
> Our use case concerns the ImmutableEntityEntry being cached and
> shared between sessions, so we need to check if the Entity is
> reference cached in the 2lc.
>
> Therefore I think the answer to the question below is;
>
> * I would consider the the trigger for using
> ImmutableEntityEntry when the Entity is marked as Immutable;
> * I would consider the trigger for not nulling the EntityEntry
> in the EntityEntryContext when the EntityEntry is an
> ImmutableEntityEntry and the Entity is reference cached in the 2lc
>
>
> There might be further optimizations for ImmutableEntityEntries
> that are not cached and shared between sessions that I am not
> aware of.
>
> It seems the most logical place to determine whether to create a
> MutableEntityEntry or an ImmutableEntityEntry is when a call is
> made to AbstractEntityPersister getEntityEntryFactory() and we can
> test to see if the Entity is Immutable.
>
> I was unsure how of the best method to determine if the Enity
> would be cached in the 2lc, so tested at the point EntityEntry is
> nulled in EntityEntryContext. removeEntityEntry() or
> EntityEntryContext.clear()
>
>
>
> On 10/04/15 17:46, Steve Ebersole wrote:
>> To post my question again...
>>
>> It really depends on what y'all consider the trigger for using
>> ImmutableEntityEntry. When would the EntityPersister use the
>> EntityEntryFactory producing ImmutableEntityEntry instances?
>> a) when the entity is marked immutable?
>> b) when the entity is marked immutable *and* we need to cache it
>> by reference?
>> c) some other condition?
>>
>> Because the answer to this ^^ dictates how we need to check
>> whether we can clear this reference
>>
>>
>> On Fri, Apr 10, 2015 at 11:45 AM, Steve Ebersole
>> <steve at hibernate.org <mailto:steve at hibernate.org>> wrote:
>>
>> Well the performance is exactly what I had in mind when I
>> asked you about the exact "trigger for using
>> ImmutableEntityEntry" in your use-case. But you never
>> replied to that.
>>
>>
>>
>> On Fri, Apr 10, 2015 at 9:26 AM, John O'Hara
>> <johara at redhat.com <mailto:johara at redhat.com>> wrote:
>>
>> I have updated my current branch:
>> https://github.com/johnaoahra80/hibernate-orm/tree/HHH-9701
>>
>> * Extracted the deserialization of EntityEntry in
>> EntityEntryContext into a separate method:
>> https://github.com/johnaoahra80/hibernate-orm/commit/d9f5dc76e05a111d37dbdac64c6aa1bd485a1ba9
>> * Created a LockModeException for invalid lock modes
>> for an ImmutableEntity:
>> https://github.com/johnaoahra80/hibernate-orm/commit/99e308ae267f9bfc30547784087955d7dc1e73e0
>>
>> Do we need a new Exception type here, or could we just
>> use a HibernateException with a message about invalid
>> lock type?
>>
>> * Added a check when EntityEntryContext.clear() or
>> EntityEntry. removeEntityEntry() is called to see if
>> the EntityEntry is ImmutableEntityEntry AND the
>> Entity is referenced cached in the 2lc:
>> https://github.com/johnaoahra80/hibernate-orm/commit/310a2eb5b8988189806a2e688090a7ea16ae5474
>>
>> Not sure of the performance implications on this, I am
>> planning a run on WF9 to test and regressions in our use
>> case.
>>
>> * Extracted an AbstractEntityEntry superclass that
>> MutableEntityEntry and ImmutableEntityEntry both
>> inherit from:
>> https://github.com/johnaoahra80/hibernate-orm/commit/df50d344441ed4cc8f3c3c8df90edc106dd52adb
>>
>> John
>>
>>
>>
>>
>> On 07/04/15 20:20, Steve Ebersole wrote:
>>> TBH, I have no idea what happens to comments on a Pull
>>> Request when you squash and force push.
>>>
>>> I'd just leave the multiple commits. We can squash them
>>> later.
>>>
>>> On Tue, Apr 7, 2015 at 1:59 PM, John O'Hara
>>> <johara at redhat.com <mailto:johara at redhat.com>> wrote:
>>>
>>> Steve,
>>>
>>> I have made changes based on the github comments and
>>> your comments below
>>> (https://github.com/johnaoahra80/hibernate-orm/commits/HHH-9701).
>>>
>>> Do you want me to squash the commits down to one?
>>> Not sure how this would effect the comments you have
>>> already made on GH
>>>
>>> Thanks
>>>
>>> John
>>>
>>>
>>> On 06/04/15 04:51, Steve Ebersole wrote:
>>>>
>>>>
>>>> On Thu, Apr 2, 2015 at 4:45 AM, John O'Hara
>>>> <johara at redhat.com <mailto:johara at redhat.com>> wrote:
>>>>
>>>> Steve,
>>>>
>>>> I have pushed a proposal for HHH-9701 to:
>>>> https://github.com/johnaoahra80/hibernate-orm/tree/HHH-9701
>>>>
>>>> There are a couple of areas that I would
>>>> appreciate feedback;
>>>>
>>>> 1) Serialization/Deserialization -
>>>> EntityEntries implementations can be serialized
>>>> and each implementation provide their own
>>>> serialization method. I have modified the
>>>> serialization of EntityEntry in
>>>> EntityEntryContext to write the Implementation
>>>> class to the OutputStream so the correct class
>>>> can be used to deserialize the object stream.
>>>> Is the exception handling sufficient here, or
>>>> do we need more robust handling of
>>>> deserialization exceptions? : see
>>>> https://github.com/johnaoahra80/hibernate-orm/commit/ec9b1fa3b97131ff1e65a1cc30ff6e4f2d2c8a28#diff-b55e1e51b30abe8d3c280bb22aeb6a44R380
>>>>
>>>>
>>>> I added some comments to that section. Also,
>>>> overall I would extract the deserialization bit
>>>> into a separate method (deserializeEntityEntry).
>>>>
>>>>
>>>> 2) In our (perf team) use case, we want to be
>>>> able to share the ImmutableEntityEntry between
>>>> sessions when they are referenced cached in the
>>>> 2lc. I have modified EntityEntryContext to not
>>>> null managedEntity.$$_hibernate_setEntityEntry
>>>> if the EntityEntry is an instance of
>>>> ImmutableEntityEntry. Do we need to add an
>>>> extra checks here, to ensure that the entity is
>>>> Reference Cached? I am not sure how we would
>>>> test that case? : see
>>>> https://github.com/johnaoahra80/hibernate-orm/commit/ec9b1fa3b97131ff1e65a1cc30ff6e4f2d2c8a28#diff-b55e1e51b30abe8d3c280bb22aeb6a44L281
>>>>
>>>>
>>>> It really depends on what y'all consider the
>>>> trigger for using ImmutableEntityEntry. When would
>>>> the EntityPersister use the EntityEntryFactory
>>>> producing ImmutableEntityEntry instances?
>>>> a) when the entity is marked immutable?
>>>> b) when the entity is marked immutable *and* we
>>>> need to cache it by reference?
>>>> c) some other condition?
>>>>
>>>> I agree that we should only not clear that
>>>> reference when the entity is enabled for
>>>> cache-by-reference. How that plays into this
>>>> depends on the answer to the above question.
>>>>
>>>> If (a), then I think that yes it makes sense to add
>>>> a check to only clear the ManagedEntity's
>>>> EntityEntry reference if using cache-by-reference.
>>>>
>>>> If (b), then the EntiytPersister is only using the
>>>> EntityEntryFactory producing ImmutableEntityEntry
>>>> instances when both are true. So the fact that an
>>>> entry is an instance of ImmutableEntityEntry
>>>> indicates that we need to not clear it from the
>>>> ManagedEntity.
>>>>
>>>>
>>>>
>>>> 3) Lock Mode: Steve you mentioned about not
>>>> doing locking for Immutable entities. Where is
>>>> the locking implemented? Would it be sufficient
>>>> to simply set the LockMode on the
>>>> ImmutableEntityEntry to NONE/READ_ONLY when
>>>> setLockMode is called?
>>>>
>>>>
>>>>
>>>> Locking is implemented in many places.
>>>>
>>>> What I had in mind, in terms of implementation for
>>>> EntityEntry, is somewhat influenced by the choice
>>>> between ignore versus exception in cases where
>>>> something is not supported. Basically I had thought
>>>> to throw an exception in
>>>> ImmutableEntityEntry#setLockMode or to simply
>>>> ignore the call altogether. This is not a great
>>>> solution.
>>>>
>>>> It is hard for me to justify ignoring the lock
>>>> request in all cases. What does everyone else think?
>>>>
>>>>
>>>>
>>>
>>>
>>> --
>>> John O'Hara
>>> johara at redhat.com <mailto:johara at redhat.com>
>>>
>>> JBoss, by Red Hat
>>> Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United Kingdom.
>>> Registered in UK and Wales under Company Registration No. 3798903 Directors: Michael Cunningham (USA), Charlie Peters (USA), Matt Parsons (USA) and Michael O'Neill (Ireland).
>>>
>>>
>>
>>
>> --
>> John O'Hara
>> johara at redhat.com <mailto:johara at redhat.com>
>>
>> JBoss, by Red Hat
>> Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United Kingdom.
>> Registered in UK and Wales under Company Registration No. 3798903 Directors: Michael Cunningham (USA), Charlie Peters (USA), Matt Parsons (USA) and Michael O'Neill (Ireland).
>>
>>
>>
>
>
> --
> John O'Hara
> johara at redhat.com <mailto:johara at redhat.com>
>
> JBoss, by Red Hat
> Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United Kingdom.
> Registered in UK and Wales under Company Registration No. 3798903 Directors: Michael Cunningham (USA), Charlie Peters (USA), Matt Parsons (USA) and Michael O'Neill (Ireland).
>
>
--
John O'Hara
johara at redhat.com
JBoss, by Red Hat
Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United Kingdom.
Registered in UK and Wales under Company Registration No. 3798903 Directors: Michael Cunningham (USA), Charlie Peters (USA), Matt Parsons (USA) and Michael O'Neill (Ireland).
More information about the hibernate-dev
mailing list