better de/serialization for resourcelimits #535
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
pretentious7
force-pushed
the
abhi/better-serde-resource-limits
branch
from
8aed55b to
ea8b923
Compare
| pub cpu_period: Option<DockerCpuPeriod>, | ||
|
|
||
| /// Proportion of period used by container | ||
| pub cpu_period_percent: Option<u8>, | ||
|
|
||
| /// Total cpu time allocated to container | ||
| #[serde_as(as = "Option<serde_with::DurationSeconds<u64>>")] |
There was a problem hiding this comment.
it feels weird having two time units on the JSON representation side. Thoughts on using a floating-point seconds representation for both?
There was a problem hiding this comment.
hmmm, this is how docker takes it, that's how we'd justified it last time.
Not fp, too much stuff close to 0. but we could use nanos or micros consistently, wdyt?
There was a problem hiding this comment.
What's a typical value for DockerCpuPeriod? I thought it would be millis at least, but maybe I misunderstand it.
I am ok straying from docker's way of doing things here, the docker API has a lot of vestigial stuff that I'm happy to paper over where we can.
There was a problem hiding this comment.
100 milliseconds, but it reports in microseconds (https://docs.kernel.org/admin-guide/cgroup-v2.html#cgroup-v2-cpu)
you generally only go under that when you've got like super short lived loops and spinlocks and such.
pretentious7
force-pushed
the
abhi/better-serde-resource-limits
branch
from
ea8b923 to
ba2f2cc
Compare
No description provided.