🌱 crd/markers: Emit cheaper CEL rules for oneOf markers

[HadrienPatte]

Followup to #1212

AtMostOneOf, ExactlyOneOf, and AtLeastOneOf currently emit a CEL rule of the form:

[has(self.a), has(self.b), ...].filter(x, x == true).size() <op> N

This pays for list construction and a comprehension over the list. Two equivalent but cheaper forms work just as well:

• AtMostOneOf/ExactlyOneOf: ternary-sum:

  (has(self.a)?1:0) + (has(self.b)?1:0) + ... <op> 1
  

  Constant-cost scalar arithmetic, no list, no iteration.

• AtLeastOneOf: short-circuit OR:

  has(self.a) || has(self.b) || ...
  

  Already a boolean, no count/compare, short-circuits as soon as any field is set.

User-visible behavior is unchanged: violation messages are emitted from the marker's static Message field, not from the rule expression itself.

I did some benchmarking with github.com/google/cel-go/checker to measure the estimated cost of both the old and the new CEL expressions with a couple number of items:

• estimated cost is from checker.EstimateCost: worst-case, what the apiserver budgets against the per-CRD limit at CRD-creation time
• runtime cost is from interpreter ActualCost: what counts against the per-call budget at admission

Runtime input states:

• none: no field set
• one: only first field set
• all: every field set
• last: only last field set

ExactlyOneOf / AtMostOneOf (filter -> sum)

N	est.min	est.max	rt.none	rt.one	rt.all
2	33 -> 6	57 -> 6 (−89%)	31 -> 6	44 -> 6	44 -> 6
3	38 -> 9	74 -> 9 (−88%)	35 -> 9	48 -> 9	48 -> 9
5	48 -> 15	108 -> 15 (−86%)	43 -> 15	56 -> 15	56 -> 15
10	73 -> 30	193 -> 30 (−84%)	63 -> 30	76 -> 30	76 -> 30

AtLeastOneOf (filter -> ||)

N	est.min	est.max	rt.none	rt.one	rt.last
2	33 -> 2	57 -> 4 (−93%)	31 -> 4	44 -> 2	44 -> 4
3	38 -> 2	74 -> 6 (−92%)	35 -> 6	48 -> 2	48 -> 6
5	48 -> 2	108 -> 10 (−91%)	43 -> 10	56 -> 2	56 -> 10
10	73 -> 2	193 -> 20 (−90%)	63 -> 20	76 -> 2	76 -> 20

Note the rt.one column for AtLeastOneOf (cost 2 across all N): the || form short-circuits as soon as any field is found, so runtime cost is independent of how many fields the marker covers when any earlier one is set.

So these numbers show that:

• both estimated and runtime cost are always lower with the new expressions
• the maximum estimated cost, which as raised in ✨ crd/marker: add AtMostOneOf and ExactlyOneOf constraints #1212 can quickly balloon when nested deep in layers of unbounded arrays is significantly reduced by about 90% in average

[k8s-ci-robot]

Hi @HadrienPatte. Thanks for your PR.

I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work.

Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[JoelSpeed]

Thanks for the analysis of the cost and the optimisation suggestions, LGTM
/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: HadrienPatte, JoelSpeed

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [JoelSpeed]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[k8s-ci-robot]

LGTM label has been added.

Details

Git tree hash: 985e3d39ae4341373ea6b344e4e1c8c0306b5dec

[JoelSpeed]

/ok-to-test