From 33b5fc3b0e5f8dd8cf9980441cb3d86aa7e018a1 Mon Sep 17 00:00:00 2001
From: chiyi <cyi.cc@outlook.com>
Date: Thu, 9 Jul 2026 17:06:43 +0800
Subject: [PATCH] fix(runway): retry the /v1/tasks submit on transient errors

A dropped proxy connection (EOF) or 5xx on the task-create failed the whole generation. submitTask now retries transient (ErrTemporaryUpstream) submits up to 3x with backoff, honoring ctx. Applies to both Nano Banana image and Gen-4 video. Validated: 10/10 4K images (19-24MB PNG each) generated across 10 accounts.
---
 backend/internal/provider/runway/image.go |  3 +--
 backend/internal/provider/runway/video.go | 20 +++++++++++++++++++-
 2 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/backend/internal/provider/runway/image.go b/backend/internal/provider/runway/image.go
index dde1c56..32e27db 100644
--- a/backend/internal/provider/runway/image.go
+++ b/backend/internal/provider/runway/image.go
@@ -9,7 +9,6 @@ import (
 	"strings"
 	"time"
 
-	http "github.com/bogdanfinn/fhttp"
 	tlsclient "github.com/bogdanfinn/tls-client"
 	"github.com/google/uuid"
 )
@@ -127,7 +126,7 @@ func (c *Client) createImageTask(ctx context.Context, client tlsclient.HttpClien
 	if len(refImages) > 0 {
 		opts["reference_images"] = refImages
 	}
-	res, err := c.apiJSON(ctx, client, token, teamID, http.MethodPost, "/v1/tasks", map[string]any{
+	res, err := c.submitTask(ctx, client, token, teamID, map[string]any{
 		"taskType":  "gemini_3_1_flash_image",
 		"options":   opts,
 		"asTeamId":  jsonNumberOrString(teamID),
diff --git a/backend/internal/provider/runway/video.go b/backend/internal/provider/runway/video.go
index 463ceb7..53f33d7 100644
--- a/backend/internal/provider/runway/video.go
+++ b/backend/internal/provider/runway/video.go
@@ -213,7 +213,7 @@ func (c *Client) createTask(ctx context.Context, client tlsclient.HttpClient, to
 	if assetGroupID != "" {
 		opts["assetGroupId"] = assetGroupID
 	}
-	res, err := c.apiJSON(ctx, client, token, teamID, http.MethodPost, "/v1/tasks", map[string]any{
+	res, err := c.submitTask(ctx, client, token, teamID, map[string]any{
 		"taskType":  "gen4_turbo",
 		"options":   opts,
 		"asTeamId":  jsonNumberOrString(teamID),
@@ -271,6 +271,24 @@ func (c *Client) pollTask(ctx context.Context, client tlsclient.HttpClient, toke
 	}
 }
 
+// submitTask POSTs a /v1/tasks create, retrying a few times on transient
+// (network / 5xx) failures. A dropped proxy connection ("EOF") on the submit
+// would otherwise fail the whole generation even though a quick retry succeeds.
+func (c *Client) submitTask(ctx context.Context, client tlsclient.HttpClient, token, teamID string, body map[string]any) (map[string]any, error) {
+	var res map[string]any
+	var err error
+	for attempt := 0; attempt < 3; attempt++ {
+		res, err = c.apiJSON(ctx, client, token, teamID, http.MethodPost, "/v1/tasks", body)
+		if err == nil || !errors.Is(err, ErrTemporaryUpstream) {
+			return res, err
+		}
+		if sleepCtx(ctx, time.Duration(attempt+1)*time.Second) != nil {
+			return nil, ctx.Err()
+		}
+	}
+	return res, err
+}
+
 // apiJSON performs an authed JSON request against the Runway API and returns the
 // parsed body, mapping status codes to the shared provider error sentinels.
 func (c *Client) apiJSON(ctx context.Context, client tlsclient.HttpClient, token, teamID, method, path string, body any) (map[string]any, error) {