llm-d Router: The intelligent entry point for inference requests
C-Rank Grade: S (Elite) - 71/100
External PR Merge Rate: 69%
Response Time: 14h
First Timer Success: 62%
llm-d/llm-d-router has a recorded first-timer success rate of 61.8%. Repositories ranked S typically provide actionable feedback during code reviews and actively nurture new community contributors.
Maintainers in llm-d/llm-d-router respond to incoming external pull requests in approximately 14.1 hours on average. Keeping PRs focused on single tasks and ensuring tests pass helps maintainers review faster.
The C-Rank™ system evaluates GitHub projects on a 0–100 scale using real data: PR merge rates, review turnaround time, active maintainer presence, and first-time contributor success. A score of 71.1 places llm-d/llm-d-router in the S tier.
The external contributor pull request merge rate for llm-d/llm-d-router is 68.7%, based on public PR activity from non-core contributors.
llm-d/llm-d-router currently has 5 active issue(s) tagged with beginner-friendly labels like "good first issue", "beginner", or "up-for-grabs".
Balanced & professional review environment in llm-d/llm-d-router. Maintains strict focus on technical quality and test standards.
Average Response Latency
Tracks hours until a maintainer leaves a review, comment, or PR response.
External Acceptance Rate
Percentage of community pull requests successfully merged into main.
First PR Conversion
Rate at which developers submitting their first repository PR succeed.
Show contributors that your repository actively reviews and merges external pull requests.
[](https://getmerged.abhishekco.de/llm-d/llm-d-router?utm_source=github&utm_medium=badge)
Historical health & responsiveness evolution
External PR acceptance
First PR success rate
50% receive response in
90% receive response in
Maintainer review activity heatmap & prime window prediction
Calculating local time...
Calculating local time...
Detected: UTC
Maintainers are most responsive around 14:00 UTC (14:00 UTC in your timezone). Recommended submission window is Tuesday - Thursday from 14:00 to 18:00 UTC (14:00 - 18:00 UTC local).
Maintainer Strategy: Maintainer Sweet Spot: PRs under 150 lines merge in ~9 hours (86% approval). Large PRs (>500 lines) take ~68 hours (31% approval). Keep your diff compact for rapid review.
What happened: The TTFT load gate keeps all endpoints when the best sticky endpoint's estimated TTFT exceeds the best non-sticky endpoint's by more than maxTTFTPenaltyMs (pkg/epp/framework/plugins/scheduling/filter/prefixcacheaffinity/plugin.go:205-215). When an endpoint lacks the attribute backing the configured ttftSource, endpointTTFT substitutes a default: math.MaxFloat64 on the latencyPredictor path, 0 ms on the prefillThroughput path (plugin.go:266-286). A gate decision can then compare an observed estimate against a substituted one. The two paths fail in opposite directions: ttftSource missing signal on substituted TTFT gate effect latencyPredictor every sticky endpoint MaxFloat64 fires against any non-sticky endpoint with a prediction; affinity lost latencyPredictor every non-sticky endpoint MaxFloat64 never fires; stickiness holds under arbitrary overload prefillThroughput any sticky endpoint 0 ms best sticky TTFT pinned to 0; never fires prefillThroughput any non
Feature Area p2p-source-producer plugin Problem Statement In multi-pod LLM serving, recomputing KV cache for long prompts (e.g., system prompts, multi-turn chat history, agentic workflows) incurs high GPU prefill latency and degrades Time-To-First-Token (TTFT). To mitigate this, P2P KV Cache Sharing allows worker pods to offload KV blocks to host CPU memory (via OffloadingConnector) and transfer them directly between peer pods over high-speed interconnects (NIXL / NVLink / RDMA) instead of recomputing them from scratch. Currently, P2P KV Cache Sharing triggers a P2P KV pull when a candidate peer holds at least minCachedTokenDelta more cached tokens than the scheduled pod. This static threshold has several critical limitations: Load-blind: While source selection became queue-aware in #2032, the decision to pull or recompute is still a fixed scalar delta with zero awareness of the destination pod's load ($d$). A static crossover measured on an idle pod (e.g., 13.6K tokens for GLM-5.2)
What would you like to be added: A way to enable an EPP feature gate (for example flowControl) on top of an existing plugin configuration without copying the entire EndpointPickerConfig document. This issue proposes (a), a --feature-gates flag, as the primary direction, and (b), structured plugin config in the chart, as a complementary chart-side option. (a) A --feature-gates EPP flag Kubelet-style: --feature-gates=flowControl=true. loadFeatureConfig (pkg/epp/config/loader/configloader.go) parses kubelet-style name=bool entries and defaults bare names to true. The flag can feed the same parser. Later featureGates entries override earlier ones. Appending flag-supplied entries to rawConfig.FeatureGates in Runner.parseConfigurationPhaseOne (cmd/epp/runner/runner.go), after the config file is read, makes flag entries win over file entries. The flag definition belongs in pkg/epp/server/options.go, alongside --config-file. No chart change is needed for Helm delivery: routerlib renders arbit
/kind feature /label observability Description The router emits OpenTelemetry spans and structured zap (logr) logs, but the two are never correlated: no code path injects trace_id/span_id into the logger, and the zap setup has no automatic correlation hook. When debugging a request, there is no way to jump from a log line to the corresponding trace, or to find the logs belonging to a given trace. Current behavior In the EPP entry point (pkg/epp/handlers/server.go), the logger is enriched with the request ID and stored via log.IntoContext before the gateway.request span is started; after tracer.Start, the logger in the context is never updated, so all downstream logs carry only the request ID. In the sidecar entry point (pkg/sidecar/proxy/chat_completions.go), the forward_request span only sets span attributes and the proxy logger is unaware of the span context. The global zap configuration (pkg/common/observability/logging/logger.go) is a plain JSON encoder with no otel correlation, so
See example: https://github.com/llm-d/llm-d-router/actions/runs/27055507318/job/79859240934?pr=1458
Rate what actually matters after opening a pull request here.
Rate your first-hand PR experience (review speed, maintainer responsiveness, and onboarding ease) to help other contributors.
When evaluating whether to contribute to llm-d/llm-d-router, response velocity and maintainer engagement are crucial. GetMerged continuously tracks pull request trajectories, first-comment latency, and code review rounds to help developers avoid submitting pull requests to backlogged repositories.
Currently, maintainers of llm-d/llm-d-router acknowledge new external contributions in approximately 14 hours. Out of all submitted pull requests from non-core authors in the last 180-day window, 68.7% were successfully merged into the primary branch.
llm-d/llm-d-router has a recorded first-timer success rate of 61.8%. Repositories ranked Elite typically provide actionable feedback during code reviews and actively nurture new community contributors.
The initial maintainer response time averages ~14 hours. Keeping PRs scoped to single concerns and ensuring CI checks succeed will optimize review turnaround.
The C-Rank™ index scores repositories on a 0 to 100 scale using an objective formula: external PR merge rates, initial response speed, active maintainer count, and first-time contributor retention. A score of 71.1 places llm-d/llm-d-router in the Elite tier.
The external pull request merge rate is 68.7%. GetMerged isolates non-core community contributions so external developers get an accurate benchmark of PR acceptance probability.
Yes, llm-d/llm-d-router currently has 5 active issue(s) tagged with beginner-friendly labels. You can inspect these directly from the repository issues tab.
All metrics displayed for llm-d/llm-d-router are automatically retrieved via the public GitHub API and recalculated daily. Insider pull requests submitted by repository owners or organization members are excluded from merge rate calculations to preserve objective external contributor statistics.