Files
spark-mcp/internal/mcp/tools/spark_submit.go
T
tao.chenandClaude f374ebfdb4 uploads,tools,config,main: 将 upload_file 返回路径改为服务器端绝对路径并透传给 spark_submit
之前 upload_file 把文件写到 DataDir/uploads/<filename>,返回相对路径
uploads/hello.txt。MCP 服务器与 agent 不在同一台机器,agent 无法构造服务
器本地路径,因此 spark_submit 无法定位脚本。

改为由新的 internal/uploads 包 mint 一个 32 字符十六进制 file_id,文件落盘为
DataDir/uploads/<file_id>,元数据写入 <file_id>.meta.json(原始文件名只存在
sidecar 里)。upload_file 返回 {file_id, path, name, size, sha256},其中 path
是绝对路径。spark_submit 的 description 明确要求 LLM 直接把 upload_file 返回
的 path 放进 args,不要自己构造路径。

为什么只在描述里约束而不在代码里拒绝非 mint 的绝对路径:集群本地已有路径
(如 /opt/spark/examples/pi.py)是合法的 spark-submit 参数,代码不能替
LLM 拒绝。

测试锁定:
- internal/uploads: Save 往返、AbsPath/Validate 非法路径、Sweep 过期/未过期/
  孤立 sidecar
- internal/mcp/tools: upload_file 新响应字段、spark_submit 透传 mint 路径与
  集群本地路径

刻意未做:S3/HDFS 上传、给 spark_submit 新增 file_id 参数、在代码层面拒绝
集群本地绝对路径。

破坏性变更:upload_file 响应从相对 path 改为绝对 path,并新增 file_id/name/
sha256 字段。该工具尚无外部调用者。

Co-Authored-By: Claude <noreply@anthropic.com>
2026-07-13 19:13:14 +08:00

122 lines
4.0 KiB
Go

package tools
import (
"context"
"fmt"
"github.com/mark3labs/mcp-go/mcp"
"spark-mcp-go/internal/executor"
"spark-mcp-go/internal/uploads"
)
const SparkSubmitName = "spark_submit"
// NewSparkSubmitTool returns the schema for the spark_submit MCP Tool.
func NewSparkSubmitTool() mcp.Tool {
return mcp.NewTool(SparkSubmitName,
mcp.WithDescription("Submit a Spark application to a configured cluster using the cluster's local spark-submit binary. The `args` array is appended to the cluster's default_submit_args; pass one argv element per CLI flag. If the script to run was just uploaded via upload_file, put the `path` field from that response as one of the args — do NOT construct your own path. Cluster-local paths (e.g. /opt/spark/examples/pi.py) are also accepted. Returns {app_id, exit_code, stdout_tail, stderr_tail, duration_ms}. Binary is invoked as a child process — no shell, no command injection."),
mcp.WithString("cluster_id",
mcp.Required(),
mcp.Description("ID of the configured cluster (from list_clusters)"),
),
// mcp-go v0.56.0 uses PropertyOption for array item schema.
mcp.WithArray("args",
mcp.Description("Spark-submit CLI args (after default_submit_args). Each element becomes one argv entry — no shell interpretation."),
mcp.Items(map[string]any{"type": "string"}),
),
)
}
// SparkSubmitHandler runs spark-submit against the requested cluster.
func (d *Deps) SparkSubmitHandler(ctx context.Context, req mcp.CallToolRequest) (*mcp.CallToolResult, error) {
clusterID, err := req.RequireString("cluster_id")
if err != nil {
return errResult("spark_submit: " + err.Error()), nil
}
args := req.GetArguments()
var userArgs []string
if rawArgs, ok := args["args"]; ok && rawArgs != nil {
arr, ok := rawArgs.([]any)
if !ok {
return errResult("spark_submit: args is not an array"), nil
}
for i, item := range arr {
s, ok := item.(string)
if !ok {
return errResult(fmt.Sprintf("spark_submit: args[%d] is not a string", i)), nil
}
userArgs = append(userArgs, s)
}
}
callLog := startToolCall(ctx, d.Logger, SparkSubmitName, map[string]any{
"cluster_id": clusterID,
"args": userArgs,
})
defer callLog.End()
cl, err := d.ClusterRepo.Get(ctx, clusterID)
if err != nil {
callLog.WithError(err)
return errResult("spark_submit: cluster " + clusterID + ": " + err.Error()), nil
}
fullArgs := append([]string{}, cl.DefaultSubmitArgs...)
fullArgs = append(fullArgs, translateArgs(userArgs, d.UploadStore)...)
result, err := executor.Run(ctx, executor.SparkSubmitOpts{
Binary: cl.SparkSubmitExecuteBin,
Args: fullArgs,
Timeout: d.SparkSubmitTimeout,
})
if err != nil {
callLog.WithError(err)
// Even on error, result carries ExitCode/Stderr for the LLM.
if result != nil {
callLog.WithResult(map[string]any{
"app_id": result.AppID,
"exit_code": result.ExitCode,
"duration_ms": result.DurationMS,
})
return textResult(encodeJSON(map[string]any{
"app_id": result.AppID,
"exit_code": result.ExitCode,
"stdout_tail": result.StdoutTail,
"stderr_tail": result.StderrTail,
"duration_ms": result.DurationMS,
"error": err.Error(),
})), nil
}
return errResult("spark_submit: " + err.Error()), nil
}
callLog.WithResult(map[string]any{
"app_id": result.AppID,
"exit_code": result.ExitCode,
"duration_ms": result.DurationMS,
})
return textResult(encodeJSON(result)), nil
}
// translateArgs canonicalizes server-minted upload paths and leaves
// cluster-local paths untouched. It is a code-level guardrail, not a
// validator: non-minted absolute paths are intentionally accepted.
func translateArgs(userArgs []string, store *uploads.Store) []string {
if store == nil {
return userArgs
}
out := make([]string, len(userArgs))
for i, arg := range userArgs {
if fileID, err := store.Validate(arg); err == nil {
if canonical, err := store.AbsPath(fileID); err == nil {
out[i] = canonical
continue
}
}
out[i] = arg
}
return out
}