[hibernate-dev] HHH-9857 - Reuse of EntityEntry for bytecode enhanced read-only reference cached entities

[John O'Hara]

On 09/06/15 15:30, Sanne Grinovero wrote:
> On 9 June 2015 at 13:50, John O'Hara <johara at redhat.com> wrote:
>> On 09/06/15 13:14, Sanne Grinovero wrote:
>>> There are lots of setters on EntityEntry, but sharing it would require
>>> at least the implementation to be fully immutable to be threadsafe.
>>>
>>> I see three options for the custom EntityEntry implementation:
>>>    - simply ignore any write method by implementing each method as a no-op
>>>    - throw exceptions on any write method
>>>    - split the EntityEntry interface into a parent interface
>>> "ReadOnlyEntityEntry" which doesn't have any such method
>>>
>>> The first option seems the easy way out but we would not notice any
>>> unintended / illegal usage; I'd prefer the third one but I'm not sure
>>> which impact it would have, seems like a large change that needs
>>> experimenting.
>>>
>>> I just noticed an ImmutableEntityEntry implementation exists now, but
>>> it's not actually immutable? That should be fixed, at very least the
>>> javadoc to explain what that class purpose is?
>> Yes, an ImmutableEntityEntry instance will be created for the EntityEntry in
>> our use case (e.g. when this would be a performance benefit), so we can test
>> for instanceof ImmutableEntityEntry or add no-ops for write operations for
>> this implementation.
>>
>> The object isn't immutable as the state field changes during the lifetime of
>> the object. This question was asked by Steve, i.e. whether it was the Entity
>> that was immutable or the EntityEntry that was immutable, I thought I had
>> replied with my thoughts but I can not find my response to that question.
>> Should clarify this in the javadoc. I think that ImmutbaleEntityEntry should
>> refer to the Entity being immutable.
> Possibly, but then you can't reuse the same instance across multiple Session(s).
> If your goal is to completely avoid allocating new instances of
> ImmutbaleEntityEntry, you have to make it really immutable, or play
> with synchronized and volatiles.. wouldn't we be adding a worse
> problem in that case? I guess we could try and measure, but if we can
> find a way to make it completely immutable that would be easier.
Yes, the goal would be to avoid allocating new instances and reusing 
across sessions. I think synchronization would be v. expensive in this 
area of code. My instinct would be try and make ImmutbaleEntityEntry 
immutable to ensure thread safety, but would need to think about current 
state changes e.g. the status field changing LOADING->READ_ONLY

> Sanne
>
>> Thanks,
>>
>> John
>>
>>> Thanks,
>>> Sanne
>>>
>>> On 9 June 2015 at 13:03, John O'Hara <johara at redhat.com> wrote:
>>>> For our use case, bytecode enhanced reference cached immutable entities,
>>>> our top object for memory allocation is EntityEntry.
>>>>
>>>> We see an EntityEntry object created every time an immutable entity is
>>>> added to a persistence context.
>>>>
>>>> In our use case, where we know the entity is immutable and we already
>>>> have an EntityEntry cached, can we re-use the EntityEntry between
>>>> sessions? This would reduce the allocation rate of EntityEntry in our
>>>> use case by ~50%.
>>>>
>>>>
>>>> --
>>>> 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).
>>>>
>>>> _______________________________________________
>>>> hibernate-dev mailing list
>>>> hibernate-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>>
>>
>> --
>> 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).
>>


-- 
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).