Hi, On Wed, Oct 12, 2011 at 11:34 AM, Thomas Mueller <[email protected]> wrote: > Please note the data store can be shared by multiple *processes* (multiple > distinct repositories and multiple cluster nodes).
I think we should drop support for multiple distinct repositories using the same data store. For multiple cluster nodes we should be able to come up with a mechanism by which the nodes can coordinate on the garbage collection task. > Keeping a separate log is problematic, as accessing the separate log would > need to be synchronized somehow. Keeping the list in memory is even more > problematic (not to mention memory usage). Yep, it's obviously a tradeoff. The upside would be that with something like that we'd be able to avoid the troublesome need to update the last access timestamp whenever a record is accessed. Ideally the datastore would only be written to when a new record is added or when the garbage collector decides to remove an unused record. BR, Jukka Zitting
