Feature: NATS Backplane
Added a NATS Backplane
I added a new project which uses NATS as a backplane for cache events. It's relatively straightforward
Reduced allocations for backplane messages
I also noticed some easy improvements that could be made to drastically reduce allocations for Backplane message, so I did a few things.
-
BackplaneMessageis now areadonly structinstead of a class which means it can be allocated on the stack -
BackplaneMessagenow has aTryParsestatic method which tries to deserialize aReadOnlySpan<byte>into aBackplaneMessageand now provides an instance methodWriteTowhich serializes theBackplaneMessageto anIBufferWriter<byte>with no allocations.
Made Backplane Tests easily extensible
I made the L1BackplaneTests and the L1L2BackplaneTests abstract classes which can easily be inherited and implement methods to provide the IFusionCacheBackplane and the IDistributedCache so tests for new backplane and Distributed Caches can easily be implemented.
This could be further improved by making these test classes perhaps take the Backplane and Distributed cache as parameters for parameterized tests. I might look at doing it like that to make it even easier.
Hi @stebet , that is awesome, thanks!
I'll take a look at the changes in details in the nex few days, will ping back here.
Hi @stebet , sorry for the delay, even though part of the blame for it is on your country... such an incredible place to visit 🙂
Anyway I've been finally able to resolve the conflict (by updating the SLN file to the new SLNX formt) and updated the branch.
I also added a couple of // TODO comments about a couple of things I quickly spotted: I still have to take a look at the rest, but it's a start.
Thanks!
BackplaneMessage is now a readonly struct instead of a class which means it can be allocated on the stack
Isn't it better to split this PR into 2 PRs. One for BackplaneMessage refactoring and another one for NATS backplane?
BackplaneMessage is now a readonly struct instead of a class which means it can be allocated on the stack
Isn't it better to split this PR into 2 PRs. One for BackplaneMessage refactoring and another one for NATS backplane?
Yep that would be better. I might amend this PR to skip the BackplaneMessage changes and put that in a separate PR later.
Thanks for the feedback