NYC Systems
NYC Systems
  • 12
  • 10 941
Designing an Agent Management Protocol for Observability by Jacob Aronoff
From NYC Systems December 2024.
Links:
* www.linkedin.com/in/jaronoff97/
* github.com/jaronoff97
* bsky.app/profile/jaronoff.com
Stay in the loop with future events: nycsystems.xyz.
Переглядів: 174

Відео

Advancing the frontier of data freshness by Nikhil Benesch
Переглядів 19914 днів тому
From NYC Systems December 2024. Links: * www.linkedin.com/in/nikhilbenesch/ * bsky.app/profile/benesch.bsky.social * github.com/benesch Stay in the loop with future events: nycsystems.xyz.
Predictive Scaling in MongoDB Atlas, an experiment by A. Jesse Jiryu Davis and Matthieu Humeau
Переглядів 2243 місяці тому
From NYC Systems October 2024. Links: * www.linkedin.com/in/ajessejiryudavis/ * www.linkedin.com/in/matthieuhumeau/ Stay in the loop with future events: nycsystems.xyz.
Techniques for Dynamic Tracing by Andrew Werner
Переглядів 6473 місяці тому
From NYC Systems October 2024. Links: * www.linkedin.com/in/andrew-werner-8228a438/ * side-eye.io * Slides: docs.google.com/presentation/d/1syg-c0Prx4wtYe7T-CfLkcRt9QsxMu9p_584DrBDl7U/ Stay in the loop with future events: nycsystems.xyz.
Dear UDFs, I Broke Up With You But Now I'm Ready To Give You a Second Chance by Sam Arch
Переглядів 7135 місяців тому
From NYC Systems August 2024. Links: * db.cs.cmu.edu/ * SamArchDB * www.linkedin.com/in/sam-arch/ Stay in the loop with future events: nycsystems.xyz.
Fast, lazy container loading in modal.com by Jonathon Belotti
Переглядів 8275 місяців тому
From NYC Systems June 2024. Links: * modal.com * jonobelotti_IO * www.linkedin.com/in/jonathonbelotti/ Stay in the loop with future events: nycsystems.xyz.
Elegant Full Text Search by Sujay Jayakar
Переглядів 8226 місяців тому
From NYC Systems June 2024. Links: * www.convex.dev/ * sujayakar314 * www.linkedin.com/in/sujayakar/ Stay in the loop with future events: nycsystems.xyz.
Scalable Analysis on Dynamic Semi-Structured Data by Dan Harris
Переглядів 2896 місяців тому
From NYC Systems June 2024. Links: * github.com/thinkharderdev * www.linkedin.com/in/dsh2va/ * thinkharderdev Stay in the loop with future events: nycsystems.xyz.
Designing & Running Low-latency Systems on Kubernetes by Jimmy Zelinskie
Переглядів 1,3 тис.9 місяців тому
From NYC Systems April 2024. Links: * authzed.com/ * jimmyzelinskie * Slides: tinyurl.com/3e8u7m34 Stay in the loop with future events: nycsystems.xyz.
Correct and low-latency real-time streaming joins by Neil Ramaswamy
Переглядів 1,8 тис.9 місяців тому
From NYC Systems April 2024. Links: * entersudonym Stay in the loop with future events: nycsystems.xyz.
Floating Point Ranges (in Julia) by Stefan Karpinski
Переглядів 40210 місяців тому
From NYC Systems February 2024. Links: * julialang.org/ * juliahub.com/ Stay in the loop with future events: nycsystems.xyz.
You Might Not Need a CRDT: Document Sync in the Wild by Paul Butler
Переглядів 3,6 тис.10 місяців тому
From NYC Systems February 2024. Links: * paulbutler.org/ * jamsocket.com/ * digest.browsertech.com/ Stay in the loop with future events: nycsystems.xyz.

КОМЕНТАРІ

  • @JapanoiseBreakfast
    @JapanoiseBreakfast 18 днів тому

    Really interesting talk. Slides are unfortunately a few seconds ahead - not sure if that's fixable after the video has gone up.

  • @felix.geisendoerfer
    @felix.geisendoerfer 3 місяці тому

    Great presentation, thanks for sharing! Given the title, I was hoping you'd also cover the distributed aspect of tracing, in particular following request flows between (Go) applications. Is that also on your radar, or are you focused on the local mechanics of observing individual processes for now?

    • @awerner32
      @awerner32 3 місяці тому

      They got the title wrong! For whatever reason the video doesn't show the title slide. The title was Techniques for Dynamic Tracing (slides: docs.google.com/presentation/d/1syg-c0Prx4wtYe7T-CfLkcRt9QsxMu9p_584DrBDl7U/ ).

    • @awerner32
      @awerner32 3 місяці тому

      As for the question: most interesting systems these days are distributed, so being able to tie distributed context together is crucial to understand whole system dynamics. Certainly observing and correlating data across processes and machines is on our radar, just wasn't the focus of this talk.

    • @felix.geisendoerfer
      @felix.geisendoerfer 3 місяці тому

      @@awerner32 got ya! No worries, great talk either way. And glad to hear y'all are thinking about the distributed systems angles as well!

    • @NYCSystems
      @NYCSystems 3 місяці тому

      @@felix.geisendoerfer Yes this was our fault, apologies! The title has been edited now.

  • @espeon91
    @espeon91 3 місяці тому

    Informative talk. Does side-eye use Frida for non-Linux use cases?

    • @awerner32
      @awerner32 3 місяці тому

      Yes! Or at least that’s the plan. It mostly works. Finishing hooking up the Linux frida implementation literally today and will move on to arm Macs soon. Tools for developers need to meet developers where they are!

  • @gungun974
    @gungun974 4 місяці тому

    Nice talk. It’s interesting in the end you have recreated what Multiplayer game server need to deal with latency’s of actions like someone is shooting and the other is moving at the same time. The idea of having one source of truth and everything client be deterministic are a great way to optimistically fill the network whole while awaiting the server response. But I think the more clever tricks is still re applied user change locally when the server didn’t respond with an acknowledge of is state ^^

  • @sumitstir
    @sumitstir 4 місяці тому

    With choosing watermark as p99 don't we have possibility of missing 1% events? If yes, is that not an issue?

    • @NeilRamaswamy
      @NeilRamaswamy 4 місяці тому

      Yes, you do have that possibility (see 11:43 for the example of dropping a record because it's "too late"). Whether it's an issue depends on the workload. If you're trying to provide a dashboard to users that's showing them their real-time ad-campaign performance, an approximate result is fine. You don't need it to be perfect. If you're trying to determine how much to pay someone, then you might not want an approximate result. You can use a materialized view for an exact result, but that can cause you to take a hit on latency. And if data truly can be months delayed, you probably need a business process to reflect that a "materialized" row can be constantly changing. For example, if you owe someone $10 for the month of March on April 1st, and then by April 15th you owe them $12, you need some process to send them $2 extra. And if it goes from $10 for the month of March on April 1st, and then by April 15th you owe them $8, you might want some way to reclaim your $2. Do you tack on $2 to the invoice for May? Do you take the $2 hit? You can see how that gets a little tricky. In general, I see people setting a longer-than-needed watermark delay to get one output result, and not have to deal with that trickiness of the previous paragraph. Sometimes missing 1% (or 0.01% of events) is worth not having to deal with the headache of a result that changes several times.

  • @Silentanwa661
    @Silentanwa661 5 місяців тому

    break away from academia and thank yourself for it

  • @AZombieHippie
    @AZombieHippie 5 місяців тому

    Excellent talk and examples

  • @amidamarurookie
    @amidamarurookie 9 місяців тому

    the slides link in the video description is broken. Can you fix it?

    • @NYCSystems
      @NYCSystems 9 місяців тому

      Should be fixed now, thanks for reporting!