System Design Interview: Design a Webhook Service w/ a Google Engineer

Поділитися
Вставка
  • Опубліковано 16 лис 2024

КОМЕНТАРІ • 23

  • @MrG0olden
    @MrG0olden Місяць тому +3

    Fantastic! I liked the wholistic approach and the explanation that although code-free, is practical

  • @akashchoudhary8162
    @akashchoudhary8162 Місяць тому +6

    Hey, man. Awesome video. In my current company, I've actually worked on a project which uses webhooks. The initial creators of the project had implemented the webhook service in a very simple but inefficient approach that you discussed first. It caused many of the pain points that you have neatly wrapped in this short video. I wish the creators of that project had watched something like this when starting the project.
    Btw, I'd really appreciate it if you could create a similar video for the counterpart of the webhook server, i.e. the external service that sends the webhook events to all the webhook servers. It'd be interesting to know how they handle sending so many events at scale and tracking whether the events were processed correctly or not and having a retry mechanism.

    • @SystemDesignSchool
      @SystemDesignSchool  Місяць тому

      Hey, thanks for the comment! Glad the video is helpful!
      For the counterpart, it'd be also be a queue based system, at a high level
      - Producer generates events like signups, payment etc, triggered by API calls to our server. The events are pushed into a message queue. Event would have destination URL, payload etc, typically pulled from DB.
      - MQ buffers the events
      - Consumers pulls events and send requests to the webhook servers. If a request fails (not 200, due to server downtime incorrect logic), the event is re-queued with exponential backoff (retries sent after progressively longer intervals) and the retry counter is incremented to track attempts.
      MQ and consumers are quite scalable.

  • @coolY2k
    @coolY2k Місяць тому +6

    2:30 API -> Request Handler -> DB has the same problem as API -> Request Handler -> Message Queue -> Consumer -> Database. The only difference is that now, instead of DB, Message Queue Service can fail. It's better, but has the exact same problem.

    • @SystemDesignSchool
      @SystemDesignSchool  Місяць тому +4

      You are right in that we are shifting the responsibility to MQ. However, scaling MQs (essentially sequential logs) are easier than DBs (more complex B-tree or LSM tree). You also get retries with MQ whereas if direct DB writes fail we lose the request.

    • @jitendraasawa6582
      @jitendraasawa6582 Місяць тому

      Request Handler -> Message Queue -> Consumer -> Database.
      To handle the failure at producing messages to MQ, we can write the failed event into a file everyday and append the failed event for that day. This file can be processed by a scheduled job that will retry the failed records. If any of the event fails again due to some reason then the customer can be notified with a list of the same.
      Question 1 is whether we want our customers to know immediately or at a later point of time that their request got failed to process?
      There is a possibility that the event processing can fail at the consumer level. For all the failed events, are we notifying the customer?
      When we get an answer to this, we will get answer to question 1.

  • @iasoto
    @iasoto Місяць тому

    What a beautiful and great explanation! Thanks for your effort to bring us such good video.

  • @databasemadness
    @databasemadness Місяць тому +1

    Amazing video!

  • @sidhantachoudhury1162
    @sidhantachoudhury1162 Місяць тому +1

    The website is bang on. Leet code of system design!

    • @SystemDesignSchool
      @SystemDesignSchool  Місяць тому

      Absolutely! You've nailed it! That’s exactly what we’re aiming for - making system design as accessible as LeetCode. Stay tuned for even more content and updates!

  • @elliotaldersone4969
    @elliotaldersone4969 Місяць тому

    great video. really helpful

  • @thewaymakerofficial
    @thewaymakerofficial 9 днів тому +1

    how to maintain the sequence of webhooks?

    • @SystemDesignSchool
      @SystemDesignSchool  9 днів тому

      do you mean how to handle out-of-order events? basically we shouldn’t expect delivery of these events in this order, and use the API to get the latest state from the source instead of relying on local state. For example, imaging we are implementing a Stripe webhook, if invoice.paid is received before invoice.created, then we'd want to call Stripe's API to get the latest state of the invoice and use that.

    • @thewaymakerofficial
      @thewaymakerofficial 8 днів тому

      @@SystemDesignSchool Got it, thanks!
      Let’s say I want to handle it on my end without involving the developer consuming the webhook. I noticed that the Telegram bot webhook sends messages in a sequential order. If one fails, it holds off on sending the next message until it successfully reaches the specified URL.
      How to achieve this?

  • @stephancodes
    @stephancodes Місяць тому

    How would a simple implementation of the MQ look like? Im currently working on implementing a webhook service that receives the webhook request from a CRM software to an API which was built on Wix, so basically I currently have the Design without the MQ and Consumer.

    • @SystemDesignSchool
      @SystemDesignSchool  Місяць тому +1

      What's the traffic like? If the traffic is not large, it's fine to not have the MQ. The sender (implemented correctly) should do retry and exponential back off.
      The solution with the MQ is more for when you have a high traffic and a strict service-level agreements (SLAs) that you have to process all requests you receive reliably (a scenario you might be asked about in an interview). And you don't have to implement an MQ from scratch; there are many cloud providers e.g. AWS SQS or Kafka etc which you can use out of the box. It's a matter of creating the queue in their console and write the code to push and consume messages.

    • @stephancodes
      @stephancodes Місяць тому

      @@SystemDesignSchool Have never used Kafka or SQS before, will definitely make some projects now! Thanks for your response it was really helpful! with the traffic we have right now, the implementation wouldn't be necessary and we can configure the retries without issue, so Ill keep this one as is, like you recommended.
      Im really liking your channel, keep up the good work

  • @Toastymuffin
    @Toastymuffin 26 днів тому +1

    Nice video, mispronounced idempotent though. ;)

  • @bithon5242
    @bithon5242 Місяць тому +5

    Wheres the google engineer?