You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
***Target audience: developers and operators of topology updaters (NodeResourceTopology producers)***
161
+
162
+
In addition to logically partitioning a cluster like explained above, the topology-aware scheduler needs to know key node-specific configuration settings like Topology manager policy and scope.
163
+
This data is expected to be provided as top-level `Attributes` of the NodeResourceTopology objects:
164
+
165
+
NodeResourceTopology producers should add top-level `Attributes` in the following format
166
+
- For `Name` and `Value` of attributes, words should be `snakeCase`
167
+
- The `Name` of each attribute should be **the same of the corresponding kubelet configuration option**.
- they should be expanded in key-value pairs, using the `String()` representation
174
+
- each key-value pair should be preceded by the `topologyManagerOption` prefix
175
+
- every other provision described above applies
176
+
- example: the `prefer-closest-numa-nodes` option becomes `topologyManagerOptionPreferClosestNumaNodes`, accepting exactly one of either `true` and `false`.
177
+
- **RATIONALE**: this representation wants to guarantee all the Attribute Names are unique (no aliasing). It must be noted this is a stricter requirement with respect to the Attribute representation
178
+
in NRT objects, and this requirement could be lifted in the future (an upgrade path will be provided).
0 commit comments