kubeai/readme: fix mistake in nri policy config#1115
kubeai/readme: fix mistake in nri policy config#1115poussa merged 1 commit intoopea-project:mainfrom
Conversation
mkbhanda
left a comment
There was a problem hiding this comment.
minor edits and a question
| ## Overview | ||
|
|
||
| [NRI plugins][nri-plugins] provide a way to | ||
| optimize the resource placement of applications in a Kubernetes cluster. They |
There was a problem hiding this comment.
| optimize the resource placement of applications in a Kubernetes cluster. They | |
| optimize the resources allocated to an application in a Kubernetes cluster. They |
There was a problem hiding this comment.
Thx. See #1119. I used wording "node-level resource assignment" there
| ## Installation of Balloons Policy Plugin | ||
|
|
||
| > **NOTE:** To avoid disturbing already running workloads it is recommended to | ||
| > install the NRI plugin to an empty node (do it right after node bootstrap, or |
There was a problem hiding this comment.
| > install the NRI plugin to an empty node (do it right after node bootstrap, or | |
| > install the NRI plugin on an empty node, that is, before launching workloads on the same (do it right after node bootstrap, or |
|
|
||
| > **NOTE:** To avoid disturbing already running workloads it is recommended to | ||
| > install the NRI plugin to an empty node (do it right after node bootstrap, or | ||
| > drain the node before installation). |
There was a problem hiding this comment.
| > drain the node before installation). | |
| > drain the node of work before installation). |
There was a problem hiding this comment.
"draining a node" (kubectl drain <node>) is a well-established term in Kubernetes so I didn't change this
|
|
||
| > **NOTE**: With containerd version earlier than v2.0 you need to enable | ||
| > the NRI support in the containerd configuration file. Instead of manual | ||
| > configuration you can provide `--set nri.runtime.patchConfig=true` to the Helm |
There was a problem hiding this comment.
| > configuration you can provide `--set nri.runtime.patchConfig=true` to the Helm | |
| > configuring/editing the file, you can pass in `--set nri.runtime.patchConfig=true` to the Helm |
There was a problem hiding this comment.
Thx. See #1119. I used slightly different wording
| ## Configuration of Balloons Policy Plugin | ||
|
|
||
| The aim of the balloons policy configuration is to isolate the model (inference | ||
| engine) containers to minimize the impact of containers on each other. |
There was a problem hiding this comment.
| engine) containers to minimize the impact of containers on each other. | |
| engine containers, to minimize noisy neighbor effects, interfering with each other. |
There was a problem hiding this comment.
Thx. See #1119. I used slightly different wording
|
|
||
| # Observability | ||
|
|
||
| With [Prometheus](../helm-charts/monitoring.md) running, install script can enable monitoring of the vLLM inference engine instances. |
There was a problem hiding this comment.
which install script? Are you asking the user to install a script for observability?
There was a problem hiding this comment.
I think this refers to kubeai/install.sh
Fix a mistake in the balloons policy config snippet. We want to match pod labels not container labels. Signed-off-by: Markus Lehtonen <markus.lehtonen@intel.com>
3a11dec to
2867399
Compare
Fix a mistake in the balloons policy config snippet. We want to match pod labels not container labels.