Normalize inference endpoint URLs

This commit is contained in:
mmckeen-nv
2026-04-29 13:01:59 -07:00
parent 10f5a0b6aa
commit e55dc7b9e0
5 changed files with 63 additions and 7 deletions
+8 -1
View File
@@ -8,6 +8,7 @@ const deleteRoutePath = path.join(root, 'app/api/sandbox/delete/route.ts')
const hostCommandsPath = path.join(root, 'app/lib/hostCommands.ts')
const telemetryRoutePath = path.join(root, 'app/api/telemetry/real/route.ts')
const inferenceRoutePath = path.join(root, 'app/api/inference/route.ts')
const inferenceEndpointUrlPath = path.join(root, 'app/lib/inferenceEndpointUrl.ts')
const ollamaModelsRoutePath = path.join(root, 'app/api/ollama/models/route.ts')
const sandboxInferenceRoutePath = path.join(root, 'app/api/sandbox/[sandboxId]/inference/route.ts')
const sandboxInferenceApplyRoutePath = path.join(root, 'app/api/sandbox/[sandboxId]/inference/apply/route.ts')
@@ -28,6 +29,7 @@ const deleteRouteSource = await readFile(deleteRoutePath, 'utf8')
const hostCommandsSource = await readFile(hostCommandsPath, 'utf8')
const telemetryRouteSource = await readFile(telemetryRoutePath, 'utf8')
const inferenceRouteSource = await readFile(inferenceRoutePath, 'utf8')
const inferenceEndpointUrlSource = await readFile(inferenceEndpointUrlPath, 'utf8')
const ollamaModelsRouteSource = await readFile(ollamaModelsRoutePath, 'utf8')
const sandboxInferenceRouteSource = await readFile(sandboxInferenceRoutePath, 'utf8')
const sandboxInferenceApplyRouteSource = await readFile(sandboxInferenceApplyRoutePath, 'utf8')
@@ -91,7 +93,11 @@ assert.match(inferenceRouteSource, /\["provider", "delete", name\]/, 'inference
assert.match(inferenceRouteSource, /\["inference", "set", "--provider", name, "--model", model\]/, 'inference route must set the active OpenShell inference route')
assert.match(inferenceRouteSource, /isVllmEndpoint\(name, baseUrl\)/, 'inference route must skip verification for vLLM-compatible endpoints')
assert.match(inferenceRouteSource, /normalizeProviderType\(requestedType, name, baseUrl\)/, 'inference route must force vLLM endpoints through the OpenAI-compatible provider type')
assert.match(inferenceRouteSource, /OPENAI_BASE_URL=\$\{baseUrl\}/, 'inference route must pass endpoint URL as provider config')
assert.match(inferenceRouteSource, /normalizeInferenceBaseUrlForGateway\(\s*baseUrl,/, 'inference route must normalize endpoint URLs for the OpenShell gateway pod')
assert.match(inferenceRouteSource, /OPENAI_BASE_URL=\$\{normalizedBaseUrl\}/, 'inference route must pass the normalized endpoint URL as provider config')
assert.match(createRouteSource, /normalizeInferenceBaseUrlForGateway\(endpointUrl, "openai"\)/, 'sandbox create must normalize env-derived OpenAI-compatible endpoints before NemoClaw onboarding')
assert.match(inferenceEndpointUrlSource, /host\.docker\.internal/, 'endpoint normalization must rewrite host-local URLs to a gateway-reachable host alias')
assert.match(inferenceEndpointUrlSource, /\/chat\/completions/, 'endpoint normalization must strip concrete OpenAI endpoint suffixes to the base URL')
assert.match(inferencePanelSource, /INFERENCE ENDPOINTS/, 'settings UI must expose inference endpoint configuration')
assert.doesNotMatch(inferencePanelSource, /NVIDIA Preset|useNvidiaInferencePreset|aws\/anthropic\/claude-opus-4-5/, 'settings UI must not expose NVIDIA-corp-specific inference presets')
assert.match(inferencePanelSource, /useVllmPreset/, 'settings UI must expose a vLLM endpoint preset')
@@ -119,6 +125,7 @@ assert.match(sandboxInferencePanelSource, /Add Route/, 'selected sandbox UI must
assert.match(sandboxInferenceApplyLibSource, /\/sandbox\/\.openclaw\/openclaw\.json/, 'live apply must patch OpenClaw config inside the running sandbox')
assert.match(sandboxInferenceApplyLibSource, /sha256sum \/sandbox\/\.openclaw\/openclaw\.json > \/sandbox\/\.openclaw\/\.config-hash/, 'live apply must refresh the immutable OpenClaw config hash')
assert.match(sandboxInferenceApplyLibSource, /\["inference", "set", "--no-verify", "--provider", primary\.provider, "--model", primary\.model\]/, 'live apply must point OpenShell inference at the primary route')
assert.match(sandboxInferenceApplyLibSource, /name: route\.model/, 'live apply must keep model display names raw while primary refs remain provider-qualified')
assert.doesNotMatch(sandboxRestartRouteSource, /delete", "pod"|kubectl delete/, 'restart route must not delete the backing sandbox pod')
assert.match(sandboxRestartRouteSource, /restartMode: "openclaw-runtime"/, 'restart route must restart only the in-sandbox OpenClaw runtime')
assert.match(sandboxRestartRouteSource, /waitForSandboxReady\(sandboxName, 15000, 1000\)/, 'restart route must verify OpenShell reports the sandbox ready before runtime restart')