Skip to content

Latest commit

 

History

History
248 lines (185 loc) · 11.4 KB

proposal.md

File metadata and controls

248 lines (185 loc) · 11.4 KB

LLM Instance Gateway

Proposal Status

Draft

Table of Contents

Summary

This proposal presents 2 new CRD objects to express the needs of the LLM Instance Gateway. BackendPool and LLMUseCase (names up for debate). The BackendPool is the logical grouping of compute, owned by the Inference Platform Admin persona. While the LLMUseCase defines the serving objectives of a specific model or LoRA adapter, and is owned by the LLM Use Case Owner.

NOTE: Some routing terms are defined in the glossary file, to more deeply describe how we will handle behaviors like priority and fairness

Goals

  • Drive concensus on direction of LLM Instance Gateway Solution
  • Documentation of API decisions for posterity

Non-Goals

  • Hash out every implementation detail
  • Be a formal KEP

Proposal

Personas

Before diving into the details of the API, decriptions of the personas will help shape the thought process of the API design.

Inference Platform Admin

The Inference Platform Admin creates and manages the infrastructure necessary to run LLM workloads. Including handling Ops for:

  • Hardware
  • Model Server
  • Base Model
  • Resource Allocation for Workloads
  • Gateway configuration
  • etc

LLM Use Case Owner

An LLM Use Case Owner persona owns and manages 1 or many Generative AI Workloads (LLM focused currently). This includes:

  • Defining SLO
  • Managing fine-tunes
    • LoRA Adapters
    • System Prompts
    • Prompt Cache
    • etc.
  • Managing rollout of adapters

Axioms

The API design is based on these axioms:

  • Pools of shared compute should be discrete for scheduling to properly work
  • Pod-level scheduling should not be handled by a high-level gateway
  • Simple use cases should be simple to define (or are implicitly defined via reasonable defaults)
  • This solution should be composable with other Gateway solutions and flexible to fit customer needs
  • The MVP will heavily assume requests are done using the OpenAI spec, but open to extension in the future
  • The Gateway should route in a way that does not generate a queue of requests at the model server level

The PoC was focused on lower-level scheduling. And the API follows that similar logic, which lead to the proposal of the BackendPool.

BackendPool

The BackendPool at its core is a logical grouping of compute, expressed in the form of Pods (typically model servers), akin to a K8s Service. The BackendPool would deploy its own routing, and offer administrative configuration to the Platform Admin.

It is expected for the BackendPool to:

  • Enforce fair consumption of resources across competing use cases
  • Efficiently route requests across shared compute (as displayed by the PoC)

It is not expected for the BackendPool to:

  • Enforce any common set of adapters or base models are available on the Pods
  • Manage Deployments of Pods within the Pool
  • Manage Pod lifecycle of pods within the pool

Additionally, any Pod that seeks to join a BackendPool would need to support a protocol, defined by LLM Instance Gateway, to ensure the Pool has adequate information to intelligently route requests.

LLMUseCase

An LLMUseCase allows the UseCaseOwner to define:

  • Which LoRA adapter(s) to consume
    • LLMUseCase allows for traffic splitting between adapters in the same backendpool to allow for new LoRA adapter versions to be easily rolled out
  • SLO objectives for the UseCase
  • The Pools this UseCase is relevant to

Spec

LLMUseCase

// LLMUseCaseSet represents a set of LLM use cases that are multiplexed onto one or more backend pools.
// This is generally owned by the "LLM Use Case Owner" persona, which can be teams in an organization.
type LLMUseCaseSet struct {
        metav1.ObjectMeta
        metav1.TypeMeta

        Spec LLMUseCaseSetSpec
}

type LLMUseCaseSetSpec struct {
        // Defines the use cases in the set.
        // UseCases can be in 2 priority classes, CRITICAL and NONCRITICAL. 
        // Priority class is implicit, and by specifying an Objective,
        // places the UseCase in the CRITICAL priority class.
        UseCases   []LLMUseCase
        // Reference to the backend pools that the use cases registers to.
        PoolRef []corev1.ObjectReference
}

// LLMUseCase defines the policies for routing the traffic of a use case, this includes performance objectives 
// and traffic splitting between different versions of the model.
type LLMUseCase struct {
        // The name of the model as the users set in the "model" parameter in the requests.
        // The model name should be unique among the use cases that reference the same backend pool.
        // This is the parameter that will be used to match the request with. In the future, we may
        // allow to match on other request parameters. The other approach to support matching on 
        // on other request parameters is to use a different ModelName f HTTPFilter 
        ModelName string
        // Optional
        // Use cases with an objective have higher priority than use cases without.
        // IMPORTANT: By specifying an objective, this places the UseCase in a higher priority class than UseCases without a defined priority class. 
        // In the face of resource-scarcity. Higher priority requests will be preserved, and lower priority class requests will be rejected.
        Objective *Objective
        // Optional.
	// Allow multiple versions of a model for traffic splitting. 
	// If not specified, the target model name is defaulted to the modelName parameter.
        TargetModels []common.TargetModel
}



// TargetModel represents a deployed model or a LoRA adapter.
type TargetModel struct {
        // The name of the adapter as expected by the ModelServer.
        TargetModelName string
        // Weight is used to determine the percentage of traffic that should be 
        // sent to this target model when multiple versions of the model are specified.
        Weight int
}

// Objective captures the latency SLO of a LLM use case.
// In MVP, meeting the SLO is on a best effort basis.
// Future: Extend the API for different behaviors of meeting the SLO.
// The gateway will perform best-effort load balancing, and work with other components (e.g., autoscaler) to meet the
// objectives. 
type Objective struct {
        // The AverageLatencyPerOutputToken is calculated as the e2e request latency divided by output token 
        // length. Note that this is different from what is known as TPOT (time per output token) which only 
        // takes decode time into account.
        // The P95 is calculated over a fixed time window defined at the operator level.
        DesiredAveragePerOutputTokenLatencyAtP95OverMultipleRequests
    *time.Duration
}

Diagrams

Much of this is better explained visually:

Below is a detailed view of the BackendPool

BackendPool

This diagram lightly follows the example request for a model interestingName. The flow can be described as:

  • The request comes in to our routing solution(Ext-Proc)
  • ExtProc looks up the UseCases affiliated with this pool examplePool
  • interestingName is currently undergoing a change of LoRA adapters from creativeNameGen-v3 (20% traffic split) to veryCreativeNameGen (80% traffic split)
  • veryCreativeNameGen is selected as the LoRA adapter, and replaces interestingName in the body of the request (mutated by ext-proc)
  • the request is then efficiently scheduled onto one of the valid Pods
  • metrics are sent back to the BEP, aggregated and re-emitted via sidecar (following the metric standardization)

How Multiple BackendPools might integrate together:

K8s Gateway with BackendPools

Here we see that we can have:

  • Multiple Routes pointing to the same pool
  • Routes splitting traffic across multiple pools

The functionality of the Kubernetes Gateway is unchanged with this proposal, allowing seamless integration with the BackendPool.

Alternatives

Key Decisions

Our alternatives hinge on some key decisions:

  • Allowing HTTPRoute to treat the BackendPool as the backendRef
    • Whereas the alternatives might have the LLMUseCase as the backend ref
  • Creating a separate layer of abstraction, instead of extending HTTPRoute
    • Explained in more detail in the LLMRoute section

LLMUseCase as a backend ref

We toyed with the idea of allowing an LLMUsecase be the target of an HTTPRouteRules backend ref. However, doing so would require the Kubernetes Gateway to be able to interpret body level parameters (assuming OpenAI protocol continues to require the model param in the body), and require that the HTTPRoute also specify the backend the UseCase is intended to run on. Since we our primary proposal already specifies the backend, packing this functionality would require substantial work on the Kubernetes Gateway, while not providing much flexibility.

LLMRoute

Our original idea was to define all UseCase config at the Kubernetes Gateway layer, and have no BackendPool. This is inherently challenging, as LLMRoute would become a superset of HTTPRoute, or the Gateway would become bespoke, and work only for the LLMRoute use case.

FAQ

  • Why 2 layers of weighting? (HttpRoute & UseCase)
    • Feasibly done - No extension of HttpRoute. Just works, as BackendPool operates like a service.
    • Complexity is only expressed during transition states (model version upgrade)
    • Keeps Pools self contained - multiple K8s gateways can direct traffic to the same pool without needing to re-express Pool-level behavior
  • What is a BEP attempting to define?
    • BackendPool groups resources that should be shared over the UseCases that are affiliated with the pool
    • Best practice would also suggest keeping the same base model for all ModelServers in the pool, but that is not enforced
  • Can a UseCase reference multiple BEPs?
  • How is this deployed?
  • Are all controllers necessary for this solution going to be provided by Instance Gateway(this repo)?
    • Yes

Open Questions

  • Reasonable defaults (how do we behave in the absence of user-specified values in optional fields)
    • Should use cases be required? Or can a customer simply create a pool, and direct requests to the pool, and expect even fairness/priority across the different LoRA adapters that are requested?
      • If so? How should we handle the mix between explicit and implicit use cases? Are implicit usecases just default everything? (and inherently lower prio).
      • NOTE: Current thinking is this is yes we should allow non-use case defined requests, but is a security risk if on by default. So pools should opt-in
  • Configuration control
    • How many routing decisions should we make on behalf of the user vs allow for configuration?
      • Do we decide that SLO adherence is stricter than Fairness adherence? Do we allow for configuration of such tooling? (would be expressed in the BackendPool API)