I agree with Jake that supporting values whose serialized size is larger than 2G would be a big change. Geode has a number of places in the implementation that use a byte array to store the serialized object form and the maximum array length in Java is 2G.
On Mon, Jul 10, 2017 at 8:43 AM, Jacob Barrett <jbarr...@pivotal.io> wrote: > Daniel, > > A change like this would be pretty deep and impact things like off heap and > the network protocol. > > You really need to consider the problems with storing such large objects > and ask yourself if you can do something different. Transferring a large > object requires that it be serialized into memory first so a 2GB object > requires 2GB of contiguous memory space for the byte[]. Allocating a 2GB > array will always be done immediately in the old generation and very high > probability result in a stop the world (STW) GC compaction event to > defragment the memory enough to allocate it. Every JVM that touches this > object (clients, primary, secondaries) will be STW GC'ing on nearly every > transmission of this object. If there are many of these objects on the > server then STW compaction events may take longer than the server timeout > and result in servers being ejected from the cluster. If auto > recover/rebalancing is enabled this could cause other servers to be force > to take on these large objects are result in them also entering a STW > compaction event and they too could be ejected. > > I mention this because I have seen this happen with an application where > they were trying to store 1GB objects. It is highly recommend that you > restructure your objects into regions of smaller objects. Since it is > likely your 2GB object isn't a single object but a relationship of many > much smaller objects then this should be very easy to do. > > -Jake > > > On Sat, Jul 8, 2017 at 11:13 PM Daniel Farcovich < > daniel.farcov...@amdocs.com> wrote: > > > Raising this question again. > > > > Daniel > > > > From: Daniel Farcovich > > Sent: Wednesday, July 05, 2017 1:26 PM > > To: 'dev@geode.apache.org' <dev@geode.apache.org> > > Subject: getSizeInBytes() return type > > > > > > Hi, > > We implement getSizeInBytes() in from Sizeable interface. > > We have objects with size bigger than MAXINT, bigger than 2GB. > > What is the impact of refactoring the code to return long instead of int? > > I mean except the technical aspect of changing the return types of the > > calling functions etc. > > Which mechanisms / functionalities will be affected from this change, I > > know rebalancing will be affected for example. > > > > Thanks, > > Daniel Farcovich > > > > > > This message and the information contained herein is proprietary and > > confidential and subject to the Amdocs policy statement, > > > > you may review at https://www.amdocs.com/about/email-disclaimer < > > https://www.amdocs.com/about/email-disclaimer> > > >
