Ledjon Behluli

Here is something I have been working on: Memory-aware Activation Rebalancing for @msftorleans / @dotnet youtube.com/watch?v=sTFD3p…

@reubenbond As many as there are [plug your number]-level inheritance hierarchies out there 😅

Fast Flexible Vertical CASPaxos is scientifically feasible. How many variations can we tack on before God intervenes?

@KooKiz @guardivelasquez Its typical for EF Core properties, but i'd argue applyinh [AllowNull] on the property is better.

@guardivelasquez TIL. I hate it but I guess it makes sense.

I'm trying GPT-5.4 and scratching my head when seeing the generated code. What reasoning process could lead to explicitly initializing to null a property, _and_ using the null-forgiving operator.

@dodyg Let me know if it was informative

Bookmarked

Microsoft Orleans is still incredible

Had the pleasure to be a guest on the NET Live Show! Myself & @reubenbond did a deep dive into the recent advancements for Placement & Balancing in @msftorleans @dotnet youtube.com/watch?v=87AlzE…

@yevhen Last one blew my mind

Such a rare gem! neal.fun/size-of-life

@dodyg Nice summary 😅

The answer is magic

@condef5 @msftorleans @dotnet Appriciated!

@BehluliLedjon @msftorleans @dotnet you rock!

Blogged: "Just how does a grain migrate?" A deep dive into how the Orleans grain migration process works. ledjonbehluli.com/posts/orleans_… @msftorleans @dotnet #aspnetcore

@TigerBeetleDB has a fantastic support for .NET. Use them in your business stack. docs.tigerbeetle.com/coding/clients…

Microsoft Orleans 9.20 with its Log-structured Grain Storage is fantastic github.com/dotnet/orleans…

@Aaronontheweb Ah escaping the scheduler. Makes sense, thanks for the infos 🙂

@BehluliLedjon PipeTo is really useful if you want to be able to spawn parallel tasks that execute outside the actor's context and then push the end-result back inside of it, but we're kind of abusing it here

Built a durable reminders / "distributed scheduler" system for Akka .NET actors yesterday. Wanted to use it for handling recurring billing and the multitude of deadline-driven edge cases that appear there. Client side invocation looks like this

@Aaronontheweb Interesting!!! I thought that was meant to be like that from the begining, essentially to fabor fire-and-forget style. Seems its more like a relict of old times.

@BehluliLedjon I'm finding a lot of things about our own APIs that I am nonplussed with in this project

@Aaronontheweb That failure may not arrive either. But i see the point/flow, i guess unless the state and reminder scheduling are stored atomically, such guarantees can not be made.

@BehluliLedjon should be - we'll get a failure notification back if it doesn't go through. I need to check to see if that includes the entire reminder payload, because we should be able to just retry scheduling it in one of the common handlers this actor has.

@Aaronontheweb Sorry for dragging on this but i am kind of (mis)using this thread to learn about akka without learning about akka 😅

@Aaronontheweb I dont know what PipeTo(self) means but i would assume the (dotnet) task is put on the (actor) scheduler i.e. kind of "go back in the queue". If not, sorry i am naive about this.

@Aaronontheweb When there is a short period for the reminder and so it happens that the new time to tick comes before the first tick/request hits the actor? Since equests interleave as the actor is reentrant, i would assume it would put some headache on end users to deal with idempotency & co.

@BehluliLedjon Appreciate you taking a look at it - I need to go clean the README up before I start asking our users to take a look at it in case they have the same questions

@Aaronontheweb Got it 👍

@BehluliLedjon > requests sent by the reminders would interleave with possible other requets oh yes, they're interleaved - the actor can process other messages / reminders. A reminder is just a scheduled future message - it doesn't immediately pause or suspend the actor