Merge 686a3b72f3 into f96200bcea

terraform/provider/CHANGELOG.md

@@ -67,6 +67,7 @@ longer signal it.
 - **mcp_server, model, key, organization_member**: Requests now accept any 2xx response instead of requiring exactly HTTP 200; the proxy legitimately returns 201 from `POST /v1/mcp/server` (and other create endpoints), so `litellm_mcp_server` create previously succeeded on the proxy but failed in the provider, leaving the resource out of state and risking a duplicate server on retry
 - **unified_access_group**: create now accepts any 2xx response instead of requiring exactly HTTP 200; `POST /v1/unified_access_group` legitimately returns 201, so creation previously succeeded on the proxy but failed in the provider, leaving the group out of state and forcing a `terraform import` to recover on the next apply's 409. Same fix and shape as the one already applied to `mcp_server`/`model`/`key`/`organization_member`; `handleResponse` (shared by several other resources) was the one status-check helper that fix didn't reach. The legacy `litellm_access_group` resource calls the unrelated `/access_group/new` endpoint, which already returns 200, so it was never affected
 - **mcp_server**: A plan right after a clean apply of a minimal litellm_mcp_server no longer shows changes. alias is now computed so the proxy's default (the server name) is kept, spec_version keeps the configured value when the proxy returns none, and an mcp_info block the proxy fills with fields the provider does not model is no longer written to state
+- **tag**: Refresh now reads `model_max_budget` back from `/tag/info`, so a per-model budget changed or cleared outside Terraform shows up as drift and an imported tag no longer plans to set it again, and equivalent JSON (key order, `25` vs `25.0`) plans no change, matching `litellm_budget` and `litellm_key` (#47010). `budget_id` is now computed as well as optional, so a tag created with inline budget fields keeps the id of the budget the proxy made instead of planning `budget_id -> null` on every run
 
 ### Changed
 

terraform/provider/docs/resources/tag.md

@@ -25,7 +25,7 @@ The following arguments are supported:
 * `name` - (Required, Forces new resource) Unique name of the tag. Also used as the resource ID
 * `description` - (Optional) Description of the tag
 * `models` - (Optional) List of model IDs this tag applies to
-* `budget_id` - (Optional) Existing budget ID to associate with this tag. If omitted and budget fields are set, the proxy creates a budget
+* `budget_id` - (Optional) Existing budget ID to associate with this tag. If omitted and budget fields are set, the proxy creates a budget and its ID is stored here
 * `max_budget` - (Optional) Max budget in USD for this tag
 * `soft_budget` - (Optional) Soft budget in USD for this tag
 * `max_parallel_requests` - (Optional) Max concurrent requests allowed for this tag

terraform/provider/litellm/resource_tag.go

@@ -19,13 +19,14 @@ const (
 )
 
 type tagBudgetTable struct {
-	BudgetID            string   `json:"budget_id"`
-	MaxBudget           *float64 `json:"max_budget"`
-	SoftBudget          *float64 `json:"soft_budget"`
-	MaxParallelRequests *int     `json:"max_parallel_requests"`
-	TPMLimit            *int     `json:"tpm_limit"`
-	RPMLimit            *int     `json:"rpm_limit"`
-	BudgetDuration      string   `json:"budget_duration"`
+	BudgetID            string      `json:"budget_id"`
+	MaxBudget           *float64    `json:"max_budget"`
+	SoftBudget          *float64    `json:"soft_budget"`
+	MaxParallelRequests *int        `json:"max_parallel_requests"`
+	TPMLimit            *int        `json:"tpm_limit"`
+	RPMLimit            *int        `json:"rpm_limit"`
+	BudgetDuration      string      `json:"budget_duration"`
+	ModelMaxBudget      interface{} `json:"model_max_budget"`
 }
 
 type tagInfoEntry struct {
@@ -68,7 +69,8 @@ func resourceLiteLLMTag() *schema.Resource {
 			"budget_id": {
 				Type:        schema.TypeString,
 				Optional:    true,
-				Description: "Existing budget ID to associate with this tag.",
+				Computed:    true,
+				Description: "Existing budget ID to associate with this tag. When unset, holds the ID of the budget the proxy creates for inline budget fields.",
 			},
 			"max_budget": {
 				Type:        schema.TypeFloat,
@@ -101,9 +103,10 @@ func resourceLiteLLMTag() *schema.Resource {
 				Description: "Duration for budget reset (e.g. '1h', '1d', '30d').",
 			},
 			"model_max_budget": {
-				Type:        schema.TypeString,
-				Optional:    true,
-				Description: "JSON object string with per-model budget configuration.",
+				Type:             schema.TypeString,
+				Optional:         true,
+				DiffSuppressFunc: budgetSuppressEquivalentJSON,
+				Description:      "JSON object string with per-model budget configuration.",
 			},
 		},
 	}
@@ -233,6 +236,8 @@ func resourceLiteLLMTagRead(d *schema.ResourceData, m interface{}) error {
 			d.Set("rpm_limit", *bt.RPMLimit)
 		}
 		d.Set("budget_duration", GetStringValue(bt.BudgetDuration, d.Get("budget_duration").(string)))
+		modelMaxBudget, _ := budgetModelMaxBudgetString(bt.ModelMaxBudget)
+		d.Set("model_max_budget", modelMaxBudget)
 	}
 
 	log.Printf("[INFO] Successfully read tag with name: %s", d.Id())

terraform/provider/litellm/resource_tag_test.go

@@ -1,6 +1,7 @@
 package litellm
 
 import (
+	"context"
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
@@ -8,6 +9,7 @@ import (
 	"testing"
 
 	"github.com/hashicorp/terraform-plugin-sdk/v2/helper/schema"
+	"github.com/hashicorp/terraform-plugin-sdk/v2/terraform"
 )
 
 func tagInfoBody(name string) string {
@@ -140,6 +142,98 @@ func TestResourceLiteLLMTagRead(t *testing.T) {
 	}
 }
 
+const tagInfoWithModelMaxBudget = `{"prod": {
+	"name": "prod",
+	"litellm_budget_table": {
+		"budget_id": "bud-created-by-proxy",
+		"model_max_budget": {"gpt-4": {"max_budget": 25.0, "budget_duration": "1d"}}
+	}
+}}`
+
+// readTagFromProxy returns the state a Read leaves behind for a tag configured with config.
+func readTagFromProxy(t *testing.T, config map[string]interface{}) *schema.ResourceData {
+	t.Helper()
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(tagInfoWithModelMaxBudget))
+	}))
+	defer srv.Close()
+
+	d := schema.TestResourceDataRaw(t, resourceLiteLLMTag().Schema, config)
+	d.SetId("prod")
+	if err := resourceLiteLLMTagRead(d, NewClient(srv.URL, "test-key", true)); err != nil {
+		t.Fatalf("read failed: %v", err)
+	}
+	return d
+}
+
+func planAttributes(t *testing.T, prior *schema.ResourceData, config map[string]interface{}) map[string]*terraform.ResourceAttrDiff {
+	t.Helper()
+	diff, err := resourceLiteLLMTag().Diff(context.Background(), prior.State(), terraform.NewResourceConfigRaw(config), nil)
+	if err != nil {
+		t.Fatalf("diff failed: %v", err)
+	}
+	if diff == nil {
+		return nil
+	}
+	return diff.Attributes
+}
+
+func TestResourceLiteLLMTagRead_PicksUpModelMaxBudgetChangedOnTheProxy(t *testing.T) {
+	d := readTagFromProxy(t, map[string]interface{}{
+		"name":             "prod",
+		"model_max_budget": `{"gpt-4": {"max_budget": 10, "budget_duration": "1d"}}`,
+	})
+
+	if !budgetSuppressEquivalentJSON("", d.Get("model_max_budget").(string), `{"gpt-4": {"max_budget": 25, "budget_duration": "1d"}}`, nil) {
+		t.Errorf("expected the proxy's per-model budget in state, got %s", d.Get("model_max_budget"))
+	}
+}
+
+func TestResourceLiteLLMTag_ModelMaxBudgetClearedOnTheProxyPlansItBack(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(`{"prod": {"name": "prod", "litellm_budget_table": {"budget_id": "bud-1", "model_max_budget": null}}}`))
+	}))
+	defer srv.Close()
+	config := map[string]interface{}{
+		"name":             "prod",
+		"model_max_budget": `{"gpt-4": {"max_budget": 10, "budget_duration": "1d"}}`,
+	}
+	prior := schema.TestResourceDataRaw(t, resourceLiteLLMTag().Schema, config)
+	prior.SetId("prod")
+	if err := resourceLiteLLMTagRead(prior, NewClient(srv.URL, "test-key", true)); err != nil {
+		t.Fatalf("read failed: %v", err)
+	}
+
+	attrs := planAttributes(t, prior, config)
+	if attrs["model_max_budget"] == nil || attrs["model_max_budget"].Old != "" {
+		t.Errorf("expected the plan to set model_max_budget again after it was cleared on the proxy, got %+v", attrs["model_max_budget"])
+	}
+}
+
+func TestResourceLiteLLMTag_PlanAfterInlineBudgetCreateKeepsTheProxyBudgetID(t *testing.T) {
+	config := map[string]interface{}{
+		"name":             "prod",
+		"model_max_budget": `{"gpt-4": {"max_budget": 25, "budget_duration": "1d"}}`,
+	}
+	prior := readTagFromProxy(t, config)
+
+	if attrs := planAttributes(t, prior, config); attrs["budget_id"] != nil {
+		t.Errorf("expected no budget_id change when the config leaves it unset, got %+v", attrs["budget_id"])
+	}
+}
+
+func TestResourceLiteLLMTag_EquivalentModelMaxBudgetJSONPlansNoChange(t *testing.T) {
+	prior := readTagFromProxy(t, map[string]interface{}{"name": "prod"})
+
+	attrs := planAttributes(t, prior, map[string]interface{}{
+		"name":             "prod",
+		"model_max_budget": `{"gpt-4": {"max_budget": 25.0, "budget_duration": "1d"}}`,
+	})
+	if attrs["model_max_budget"] != nil {
+		t.Errorf("expected key order and number formatting to plan no change, got %+v", attrs["model_max_budget"])
+	}
+}
+
 func TestResourceLiteLLMTagRead_404ClearsID(t *testing.T) {
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.WriteHeader(http.StatusNotFound)