Decrease size of flow execution storage to improve performance and ability to use ClientContinuationFlowExecutionStorage in urls [SWF-36] about spring-webflow HOT 2 OPEN

Chris Wood commented

I've come across this issue before, and come up with a good (albeit slightly hacky) solution. Much of the bulk of the java serialization data is made up of class descriptors. Since we're only ever saving the data between requests, there's no chance that the class will change.

My solution is to firstly write a list of 'known classes' to an ObjectOutputStream, then reflect on the object that you're writing and find all the class types that will get written to the stream and write the classes, then flush the stream and take note of the stream length, and then finally write the object and close the stream.

Into a second output buffer with gzip compression write all the names of the classes that were found on introspection as UTF-8, followed by the trailing end of the OOS buffer from above.

On reconstitution prepare a new OOS buffer, write the known classes, and then read the list of found classes from the beginning of the data and write these, then flush the stream and append the rest of the data.

This works quite well, compressing a ~2000 char serialized state (~1000 when gzipped) down to only 300 chars. It works because all the subsequent references to a class already serialized gets written as a back reference in the stream, and so emits only 8 bytes rather than a much larger number.

Another thing I've done is to (optionaly) checksum the resulting data. I applied this checksum to the result of the above process, realy that is not the best time to do the checksum, it would be better to checksum the first OOS and then write the checksum on that because that would protect against the class definition changing.

I'll attach the relivant classes.

from spring-webflow.

Chris Wood commented

NB: I'd advise against putting this data in the request URL. Different browsers handle different maximum lengths, and you never know when the data might just get a bit too long

from spring-webflow.