[management] Check a provider's url and credential before saving it

e2e/agentnetwork/credential_check_live_test.go

@@ -0,0 +1,204 @@
+//go:build e2e
+
+package agentnetwork
+
+import (
+	"context"
+	"net/http"
+	"os"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/client/rest"
+	"github.com/netbirdio/netbird/shared/management/http/api"
+)
+
+// credentialCase is one vendor to try the save-time check against. The key is
+// the real one the suite already sources; corrupting it is what produces the
+// refusal, so the pair of cases differ only in the credential.
+type credentialCase struct {
+	name      string
+	catalogID string
+	upstream  string
+	apiKey    string
+}
+
+// liveCredentialCases mirrors the discovery matrix's env gating so a partial
+// key set still yields partial coverage. Vertex is left out: its credential is
+// a service-account keyfile, and mangling one produces a client-side parse
+// failure rather than the vendor refusal this is about.
+func liveCredentialCases() []credentialCase {
+	var cases []credentialCase
+
+	if k := os.Getenv("OPENAI_TOKEN"); k != "" {
+		cases = append(cases, credentialCase{
+			name: "openai", catalogID: "openai_api",
+			upstream: "https://api.openai.com", apiKey: k,
+		})
+	}
+	if k := os.Getenv("ANTHROPIC_TOKEN"); k != "" {
+		cases = append(cases, credentialCase{
+			name: "anthropic", catalogID: "anthropic_api",
+			upstream: "https://api.anthropic.com", apiKey: k,
+		})
+	}
+	if k := os.Getenv("AWS_BEARER_TOKEN_BEDROCK"); k != "" {
+		region := os.Getenv("AWS_REGION")
+		if region == "" {
+			region = "eu-central-1"
+		}
+		cases = append(cases, credentialCase{
+			name: "bedrock", catalogID: "bedrock_api",
+			upstream: "https://bedrock-runtime." + region + ".amazonaws.com", apiKey: k,
+		})
+	}
+
+	return cases
+}
+
+// TestLiveProviderCredentialCheck drives the save-time check against the real
+// vendors. A unit test can only assert that a mocked refusal is classified;
+// what it cannot show is that these vendors refuse a bad key on their listing
+// endpoint at all, which is the assumption the whole feature rests on.
+//
+// The good-key case matters just as much as the bad one: a check that refused
+// everything would pass a test asserting only the refusal, and would make the
+// product unusable.
+func TestLiveProviderCredentialCheck(t *testing.T) {
+	cases := liveCredentialCases()
+	if len(cases) == 0 {
+		t.Skip("no live provider credentials in the environment; source ~/.llm-keys to run")
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Minute)
+	defer cancel()
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Run("a good credential saves", func(t *testing.T) {
+				prov, err := srv.CreateProvider(ctx, credentialProviderRequest(tc, "e2e-cred-ok-"+tc.name, tc.apiKey))
+				require.NoError(t, err, "the suite's own credential must pass its check")
+				t.Cleanup(func() { _ = srv.DeleteProvider(context.Background(), prov.Id) })
+				require.NotEmpty(t, prov.Id)
+			})
+
+			t.Run("a rejected credential is refused", func(t *testing.T) {
+				_, err := srv.CreateProvider(ctx, credentialProviderRequest(tc, "e2e-cred-bad-"+tc.name, corrupt(tc.apiKey)))
+				require.Error(t, err, "a key the vendor rejects must not save")
+
+				var apiErr *rest.APIError
+				require.ErrorAs(t, err, &apiErr)
+				require.Equal(t, http.StatusUnprocessableEntity, apiErr.StatusCode,
+					"a refused credential is the caller's problem to fix, not a server fault")
+				require.Contains(t, strings.ToLower(apiErr.Message), "rejected the credential",
+					"the message must name the credential rather than the url")
+
+				// The record must be absent, not merely unusable: a provider
+				// saved despite its check is the state this prevents.
+				all, listErr := srv.ListProviders(ctx)
+				require.NoError(t, listErr)
+				for _, p := range all {
+					require.NotEqual(t, "e2e-cred-bad-"+tc.name, p.Name, "a refused provider must not be stored")
+				}
+			})
+		})
+	}
+}
+
+// TestLiveProviderUrlCheck points a real credential at a host that is not the
+// vendor's API. It is the half of the split a wrong key cannot exercise: the
+// operator has to be told the URL is at fault while their key is fine.
+func TestLiveProviderUrlCheck(t *testing.T) {
+	cases := liveCredentialCases()
+	if len(cases) == 0 {
+		t.Skip("no live provider credentials in the environment; source ~/.llm-keys to run")
+	}
+	tc := cases[0]
+
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
+	defer cancel()
+
+	// A name that resolves nowhere. The check has to reach a verdict without
+	// the vendor's help, which is the transport half of the classification.
+	req := credentialProviderRequest(tc, "e2e-cred-badurl", tc.apiKey)
+	req.UpstreamUrl = "https://not-a-real-vendor-host.netbird-e2e.invalid"
+
+	_, err := srv.CreateProvider(ctx, req)
+	require.Error(t, err, "an upstream that does not resolve must not save")
+
+	var apiErr *rest.APIError
+	require.ErrorAs(t, err, &apiErr)
+	require.Equal(t, http.StatusUnprocessableEntity, apiErr.StatusCode)
+	require.Contains(t, strings.ToLower(apiErr.Message), "could not be reached",
+		"the message must name the url rather than the credential")
+}
+
+// TestLiveProviderUpdateKeepsTheWorkingKey is the state the check exists to
+// prevent on the update path: a rejected rotation that has already replaced
+// the credential would take a working provider down.
+func TestLiveProviderUpdateKeepsTheWorkingKey(t *testing.T) {
+	cases := liveCredentialCases()
+	if len(cases) == 0 {
+		t.Skip("no live provider credentials in the environment; source ~/.llm-keys to run")
+	}
+	tc := cases[0]
+
+	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Minute)
+	defer cancel()
+
+	prov, err := srv.CreateProvider(ctx, credentialProviderRequest(tc, "e2e-cred-rotate", tc.apiKey))
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = srv.DeleteProvider(context.Background(), prov.Id) })
+
+	rotation := credentialProviderRequest(tc, "e2e-cred-rotate", corrupt(tc.apiKey))
+	_, err = srv.UpdateProvider(ctx, prov.Id, rotation)
+	require.Error(t, err, "a rotation the vendor rejects must not be stored")
+
+	var apiErr *rest.APIError
+	require.ErrorAs(t, err, &apiErr)
+	require.Equal(t, http.StatusUnprocessableEntity, apiErr.StatusCode)
+
+	// The stored key is never returned by the API, so the proof that it
+	// survived is that an edit which reuses it still passes its check. A
+	// replaced key would fail here exactly as the rotation just did.
+	renamed := credentialProviderRequest(tc, "e2e-cred-rotate-renamed", "")
+	updated, err := srv.UpdateProvider(ctx, prov.Id, renamed)
+	require.NoError(t, err, "the working key must still be the stored one")
+	require.Equal(t, "e2e-cred-rotate-renamed", updated.Name)
+}
+
+// credentialProviderRequest builds a create/update body for a case. An empty apiKey is
+// omitted rather than sent blank, which is how the form asks to keep whatever
+// is already stored.
+func credentialProviderRequest(tc credentialCase, name, apiKey string) api.AgentNetworkProviderRequest {
+	req := api.AgentNetworkProviderRequest{
+		Name:        name,
+		ProviderId:  tc.catalogID,
+		UpstreamUrl: tc.upstream,
+		Enabled:     ptr(true),
+	}
+	if apiKey != "" {
+		req.ApiKey = &apiKey
+	}
+	return req
+}
+
+// corrupt returns a key the vendor will reject while keeping the shape of the
+// original. Replacing the last character rather than appending keeps any
+// length or prefix validation satisfied, so the refusal comes from the vendor
+// checking the secret rather than from it rejecting an obviously malformed
+// one.
+func corrupt(key string) string {
+	if key == "" {
+		return key
+	}
+	last := key[len(key)-1]
+	replacement := byte('A')
+	if last == 'A' {
+		replacement = 'B'
+	}
+	return key[:len(key)-1] + string(replacement)
+}

e2e/agentnetwork/management_test.go

@@ -16,14 +16,20 @@ import (
 
 func ptr[T any](v T) *T { return &v }
 
-// newProvider creates an OpenAI-catalog provider with a dummy key (these tests
-// never call the upstream) and registers cleanup.
+// newProvider creates an OpenAI-catalog provider these tests can hang a policy
+// off, and registers cleanup. Nothing here calls the upstream.
 func newProvider(t *testing.T, ctx context.Context, name string) api.AgentNetworkProvider {
 	t.Helper()
+	// A provider save is credential-checked against the vendor, and every
+	// caller here wants a provider row to hang a policy off rather than a
+	// working upstream. A private address is left unchecked — the proxy would
+	// reach it through the tunnel, management cannot reach it at all — which
+	// keeps this fixture independent of whether the run has vendor keys, and
+	// covers the unchecked-provider-still-saves path while it is at it.
 	prov, err := srv.CreateProvider(ctx, api.AgentNetworkProviderRequest{
 		Name:        name,
 		ProviderId:  "openai_api",
-		UpstreamUrl: "https://api.openai.com",
+		UpstreamUrl: "https://10.255.255.1",
 		ApiKey:      ptr("sk-dummy-e2e-key"),
 	})
 	require.NoError(t, err, "create provider %q", name)

management/internals/modules/agentnetwork/credentialcheck.go

@@ -0,0 +1,98 @@
+package agentnetwork
+
+import (
+	"context"
+	"errors"
+	"net/http"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/management/internals/modules/agentnetwork/modeldiscovery"
+	"github.com/netbirdio/netbird/management/internals/modules/agentnetwork/types"
+	"github.com/netbirdio/netbird/shared/management/status"
+)
+
+// ModelLister is the vendor-facing half of the credential check.
+// modeldiscovery.Client is the only production implementation; it is an
+// interface because the check runs on a write path, so without a seam every
+// test that saves a provider would reach a vendor to do it.
+type ModelLister interface {
+	Fetch(ctx context.Context, req modeldiscovery.Request) ([]modeldiscovery.Model, error)
+}
+
+// checkProviderCredential refuses a record whose upstream or credential the
+// vendor will not accept.
+//
+// It reuses the discovery Fetch rather than a lighter status probe so it
+// exercises the path the model picker takes: a URL answering 200 with a login
+// page fails here instead of producing an empty picker later.
+func (m *managerImpl) checkProviderCredential(ctx context.Context, provider *types.Provider) error {
+	_, err := m.modelDiscovery.Fetch(ctx, modeldiscovery.Request{
+		CatalogID:   provider.ProviderID,
+		UpstreamURL: provider.UpstreamURL,
+		APIKey:      provider.APIKey,
+	})
+	if err == nil {
+		return nil
+	}
+
+	message, blocking := credentialCheckFailure(err)
+	if !blocking {
+		log.WithContext(ctx).Debugf("agent network provider %s not credential-checked: %v", provider.ProviderID, err)
+		return nil
+	}
+
+	// WriteError logs only what we return, and that carries no status code,
+	// so the vendor's number is recorded here or nowhere.
+	log.WithContext(ctx).Infof("agent network provider %s failed its credential check: %v", provider.ProviderID, err)
+
+	return status.Errorf(status.InvalidArgument, "%s", message)
+}
+
+// credentialCheckFailure renders a discovery failure as the sentence the
+// provider form shows, and reports whether it should block the write.
+//
+// The strings survive WriteError lowercasing them, and never echo the
+// operator's URL: paths are case-sensitive, so an echoed URL comes back
+// altered and describes something they did not type.
+func credentialCheckFailure(err error) (message string, blocking bool) {
+	// Not checkable. The record may be perfectly good and we have no way to
+	// ask, so reporting a failure would be a guess.
+	switch {
+	case errors.Is(err, modeldiscovery.ErrNoDiscovery),
+		errors.Is(err, modeldiscovery.ErrNoDiscoveryHost),
+		errors.Is(err, modeldiscovery.ErrPrivateHost):
+		return "", false
+	}
+
+	var vendor *modeldiscovery.VendorStatusError
+	if errors.As(err, &vendor) {
+		switch vendor.Status {
+		case http.StatusUnauthorized, http.StatusForbidden:
+			return "the provider rejected the credential", true
+		case http.StatusNotFound, http.StatusMethodNotAllowed:
+			return "the upstream url did not answer a model listing", true
+		default:
+			// 5xx and 429 included: an outage still leaves the record
+			// unverified, which is what this refuses to save.
+			return "the provider returned an error", true
+		}
+	}
+
+	var unreachable *modeldiscovery.UnreachableError
+	if errors.As(err, &unreachable) {
+		if reason := unreachable.Reason(); reason != "" {
+			return "the upstream url could not be reached: " + reason, true
+		}
+		return "the upstream url could not be reached", true
+	}
+
+	if errors.Is(err, modeldiscovery.ErrUnparseableListing) {
+		return "the upstream url answered, but not with a model listing", true
+	}
+
+	// Ours rather than the vendor's — a request this code built badly, or a
+	// catalog entry that does not match its parser. Still unverified, so it
+	// still blocks.
+	return "the provider could not be checked", true
+}