功能：完善云管助手与GPU故障恢复核对

platform/pkg/proxmox/client.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"net/url"
 	"os"
@@ -123,16 +124,48 @@ func (c *Client) Action(ctx context.Context, node string, vmid int, action strin
 }
 func (c *Client) Task(ctx context.Context, node, upid string) (TaskStatus, error) {
 	var status TaskStatus
-	err := c.request(ctx, "GET", "/nodes/"+url.PathEscape(node)+"/tasks/"+url.PathEscape(upid)+"/status", nil, &status)
+	err := c.readWithRetry(ctx, "/nodes/"+url.PathEscape(node)+"/tasks/"+url.PathEscape(upid)+"/status", &status)
 	return status, err
 }
 func (c *Client) Power(ctx context.Context, node string, vmid int) (string, error) {
 	var status struct {
 		Status string `json:"status"`
 	}
-	err := c.request(ctx, "GET", fmt.Sprintf("/nodes/%s/qemu/%d/status/current", url.PathEscape(node), vmid), nil, &status)
+	err := c.readWithRetry(ctx, fmt.Sprintf("/nodes/%s/qemu/%d/status/current", url.PathEscape(node), vmid), &status)
 	return status.Status, err
 }
+
+// Retry only observations. A lost POST response may already have changed the
+// VM, so commands must never use this path.
+func (c *Client) readWithRetry(ctx context.Context, path string, out any) error {
+	var err error
+	for attempt := 0; attempt < 3; attempt++ {
+		if err = ctx.Err(); err != nil {
+			return err
+		}
+		err = c.request(ctx, "GET", path, nil, out)
+		if err == nil || !retryableRead(err) || attempt == 2 {
+			return err
+		}
+		timer := time.NewTimer(time.Duration(attempt+1) * 250 * time.Millisecond)
+		select {
+		case <-ctx.Done():
+			timer.Stop()
+			return ctx.Err()
+		case <-timer.C:
+		}
+	}
+	return err
+}
+
+func retryableRead(err error) bool {
+	var response *HTTPError
+	if errors.As(err, &response) {
+		return response.Status == 408 || response.Status == 429 || response.Status == 500 || response.Status == 502 || response.Status == 503 || response.Status == 504
+	}
+	var network net.Error
+	return errors.As(err, &network) && (network.Timeout() || network.Temporary()) || errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF)
+}
 func (c *Client) Wait(ctx context.Context, node, upid string, interval time.Duration) error {
 	ticker := time.NewTicker(interval)
 	defer ticker.Stop()