schema/defs-linux: change weight type to uint16

[0x0916]

Signed-off-by: 0x0916 w@laoqinren.net

[dqminh]

LGTM

[stevvooe]

This change is probably not backwards compatible and against best practices. Do you have any documentation citing this as a uint16?

cc @tonistiigi @crosbymichael

[0x0916]

@stevvooe Please refer to https://github.com/opencontainers/runtime-spec/blob/master/config-linux.md#block-io

also, the kernel document: https://www.kernel.org/doc/Documentation/cgroup-v1/blkio-controller.txt

- blkio.weight
	- Specifies per cgroup weight. This is default weight of the group
	  on all the devices until and unless overridden by per device rule.
	  (See blkio.weight_device).
	  Currently allowed range of weights is from 10 to 1000.

[stevvooe]

@0x0916 How does that counter any of the concerns? The kernel API doesn't mention that this is restricted to a uint16. The specification currently calls for an integer and this requires a uint16. If others were expecting this to take something outside of a uint16, it would be considered backwards incompatible.

It is unlikely that this will break anyone, but I'd urge caution.

[dqminh]

@stevvooe this should not break anyone, it also matches the type in spec-go which has been uint16 since forever. I think this is more like fixing a typo.

[stevvooe]

@dqminh Didn't we recently learn our lesson about liberal use of unsigned integers? The only use case of unsigned integers is when they map a field to a specific bit width. Unsigned integers should not be used to validate that something is positive or within a certain range of values. That is the job of validation.

[wking]

To help this PR along, here are the current types for weight in the current master (3417098):

• uint16 in config-linux.md under blockIO.
• uint16 in config-linux.md under weightDevice.
• *uint16 in config.go under LinuxBlockIO.
• *uint16 in config.go under LinuxWeightDevice.
• integer in defs-linux.json via blockIO in config-linux.json.
• integer in defs-linux.json via weightDevice in config-linux.json and blockIODeviceWeight in defs-linux.json.

The weight entry in defs-linux.json (which is what this PR is touching as of 574182a) is also consumed by the leafWeight entries:

$ git describe
v1.0.0-39-g3417098
$ git grep -B2 definitions/weight schema/
schema/config-linux.json-                            "weight": {
schema/config-linux.json-                                "id": "https://opencontainers.org/schema/bundle/linux/resources/blockIO/weight",
schema/config-linux.json:                                "$ref": "defs-linux.json#/definitions/weight"
--
schema/config-linux.json-                            "leafWeight": {
schema/config-linux.json-                                "id": "https://opencontainers.org/schema/bundle/linux/resources/blockIO/leafWeight",
schema/config-linux.json:                                "$ref": "defs-linux.json#/definitions/weight"
--
schema/defs-linux.json-                    "properties": {
schema/defs-linux.json-                        "weight": {
schema/defs-linux.json:                            "$ref": "#/definitions/weight"
schema/defs-linux.json-                        },
schema/defs-linux.json-                        "leafWeight": {
schema/defs-linux.json:                            "$ref": "#/definitions/weight"

And leafWeight is also a unit16 in master:

$ git grep 'leafWeight.*OPTIONAL' config-linux.md
config-linux.md:* **`leafWeight`** *(uint16, OPTIONAL)* …
config-linux.md:    * **`leafWeight`** *(uint16, OPTIONAL)* …
$ git grep 'leafWeight' specs-go/
specs-go/config.go:     LeafWeight *uint16 `json:"leafWeight,omitempty"`
specs-go/config.go:     LeafWeight *uint16 `json:"leafWeight,omitempty"`

So the only thing in that whole chain that is not uint16 (modulo pointer) is the defs-linux.json entry being fixed in 574182a.

@stevvooe brings up #876, and I have no problem with runtime-spec maintainers deciding that weight and/or leafWeight should really have been int64 (or whatever). But whatever they have chosen as the type in the normative Markdown spec should be reflected in the informative, provided as-is JSON Schema and Go types provided alongside the spec. 574182a is making things consistent assuming the runtime-spec maintainers want to stick with the Markdown uint16. If they decide to switch to a different type, we'd need a separate PR. But just leaving things as they stand in master leaves the JSON Schema out of sync with the normative Markdown for no purpose.

[vbatts]

house keeping: as this is in just the jsonschema and matches what the docs say, i'm cool with it

LGTM

[tianon]

LGTM

This also matches the Go implementation, which is convincing IMO.