Add note to starvation docs re: cpus in k8s #3750
Conversation
docs/core/starvation-and-tuning.md
Outdated
| @@ -434,7 +442,7 @@ If this is not possible, the next-best approach is to rely on a circular buffer. | |||
|
|
|||
| > This scenario is specific to Kubernetes, Amazon ECS, and similar resource-controlled containerized deployments. | |||
|
|
|||
| In many Docker clusters, it is relatively standard practice to over-provision CPU credits by some factor (even 100% over-provisioning is quite common). What this effectively means is that the container environment will promise (or at least *allow*) *m* applications access to *n* vCPUs each, despite only *(m * n) / (1 + k)* CPUs being physically present across the underlying hardware cluster. In this equation, *k* is the over-provisioning factor, often referred to as "burst credits". | |||
| In many Docker clusters, it is relatively standard practice to over-provision CPU credits by some factor (even 100% over-provisioning is quite common). What this effectively means is that the container environment will promise (or at least *allow*) *m* applications access to *n* vCPUs each, despite only *(m* n) / (1 + k)*CPUs being physically present across the underlying hardware cluster. In this equation,*k* is the over-provisioning factor, often referred to as "burst credits". | |||
There was a problem hiding this comment.
I don't think we want the spaces removed from *(m * n) / (1 + k)* CPUs, it looks like the formatting takes over the multiplication
There was a problem hiding this comment.
oh drat, yeah I didn't even touch that - it's just the file formatting. I'll fix it
|
Before this gets merged I want to add some wording about overriding the thread count in |
|
drive-by-dropping a link to more material we should incorporate https://discord.com/channels/632277896739946517/839263556754472990/1135955773613146152 |
|
going to close and reopen b/c I think that's what's needed to get the updated CI workflows |
|
nope I appear to have been wrong about that womp womp |
We merged your fix to 3.5.x, but this PR is targeting 3.x. Actually this PR could target 3.5.x. |
|
Woops just changing the target branch is not sufficient 🤦♂️ |
This is a common "gotcha" when running in k8s. Ideally we could automatically detect this environment in the CE runtime itself on initialization and warn the user proactively, but at least this gives people something to ctrl-f for in the doc
This is a common "gotcha" when running in k8s.
Ideally we could automatically detect this environment in the CE runtime itself on initialization and warn the user proactively, but at least this gives people something to ctrl-f for in the doc