Birbla

Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

Sign in
  • Hacker News
  • I need to know how to animate things like this for internal documentation.
    by stroebs
  • The author re-implemented a bunch of k8s logic in a typescript library just for these animations: https://github.com/ngrok/webernetes
    by claytonjy
  • This does not state anything new, but explains it so much well than the kubernetes documentation.
    by sidcool
  • Sometimes the k8s docs remind me of google's documentation
    by srichard16
  • Sam has a real way with words when it comes to educational content.

    Though he did find a legit Kubernetes bug while writing the post, so technically there was at least one new thing :)

    by dev_cprice
  • SRE here, Strong disagree with do not fail readiness and liveness checks on upstream dependencies failing. There are several reason to do so and unless you have extreme start up time, what's the problem with restarting?

    Maybe DNS has changed on you but you are stuck with bad local cache because you poorly respect TTLs (Looking at you Java), reseting the process will clear that cache away.

    Maybe TCP connections are in stuck weird state, resetting the process generally helps with that.

    Maybe someone gave you bad ENV VARs and you cannot connect to database, by refusing to progress the rollout, no outage generated.

    So yea, if you are not ready to do work including critical upstream dependencies, don't lie to system and say you are.

    by stackskipton
  • cascading failures on upstream services. then you get 20 different services failing instead of the single one.
    by javier2
  • My favourite: misconfigured Linkerd setup that causes CA certs to rotate every month :) Definitely worth restarting on that
    by figmert
  • The better solution is to not have too many critical upstream services :)
    by connicpu
  • > what's the problem with restarting?

    Exponential backoff can delay recovery up to kubelet’s maxContainerRestartPeriod (default 5m).

    by cmckn
  • I think the best approach is to have an endpoint which returns the current state without probing the database when you check the endpoint.

    Like you have a threat which does all your checks for connectivity write it into memory and the api endpoint checks the last values written.

    Then you can respond to different upstream issues differently: Your DB says 'permission denied' due to broken auth data? Your Pod is not ready.

    Your backend understand what a db maintenance mode is and should still return "Maintenance", your pod stays ready.

    It can be more nuanced.

    But the problem with restarting is, as he stated in the blog: What if now EVERY pod restarts in parallel.

    by Flamkuchlo
  • 1. was already mentioned in sibling - cascade failures

    2. you'll have massive number of restarts for various flake reasons and missing things that got papered over with restarts until you hit 1 and everything is broken. another popular version of this is "just restart when memory leaks too much"

    by dilyevsky
  • Thundering herd / cascading outages. You take out a large enough portion of your fleet, and the remaining load overloads your remaining nodes one by one as they restart, so you can never have enough healthy nodes.
    by arccy
  • Unlike sibling commenters who just read about "thundering herd" problems on the Internet, as someone who spent significant time in SRE roles, I agree with you as a matter of what the default approach should be. If you have a small cluster and no more than a handful of services... what thundering herd problem is there supposed to be, exactly, with so few "cattle" in the "herd"? Meanwhile, there are serious benefits, as you describe.

    Large clusters with dozens of services and traces that go several services deep, with each service owned by a different team, are a whole 'nother ballgame, especially when overall production uptime is owned by an SRE team and not by the developer teams who wrote each of those services. And even in this scenario, you're not necessarily wrong; the risk attached to the cascading failure is domain-specific and may be acceptable.

    by solatic