Updated on 2024-11-12 GMT+08:00

Custom Runtime

Scenarios

A runtime runs the code of a function, reads the handler name from an environment variable, and reads invocation events from the runtime APIs of FunctionGraph. The runtime passes event data to the function handler and returns the response from the handler to FunctionGraph.

FunctionGraph supports custom runtimes. You can use an executable file named bootstrap to include a runtime in your function deployment package. The runtime runs the function's handler method when the function is invoked.

Your runtime runs in the FunctionGraph execution environment. It can be a shell script or a binary executable file that is compiled in Linux.

After programming, simply package your code into a ZIP file (Java, Node.js, Python, and Go) or JAR file (Java), and upload the file to FunctionGraph for execution. When creating a ZIP file, place the handler file under the root directory to ensure that your code can run normally after being decompressed.

If you edit code in Go, zip the compiled file, and ensure that the name of the dynamic library file is consistent with the plug-in name of the handler. For example, if the name of the dynamic library file is testplugin.so, set the handler name to testplugin.Handler.

Compiling Description

If you compile files on Windows and execute them on Linux, you need to configure the following parameters:
CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build main.go

Runtime File bootstrap

If there is a file named bootstrap in your function deployment package, FunctionGraph executes that file. If the bootstrap file is not found or not executable, your function will return an error when invoked.

The runtime code is responsible for completing initialization tasks. It processes invocation events in a loop until it is terminated.

The initialization tasks run once for each instance of the function to prepare the environment for handling invocations.

Runtime APIs

FunctionGraph provides HTTP runtime APIs to receive function invocation events and returns response data in the execution environment.

  • Obtaining Invocation Event

    Method – Get

    Path – http://$RUNTIME_API_ADDR/v1/runtime/invocation/request

    This API is used to retrieve an invocation event. The response body contains the event data. The following table describes additional data about the invocation contained in the response header.

    Table 1 Response header information

    Parameter

    Description

    X-Cff-Request-Id

    Request ID.

    X-CFF-Access-Key

    AK of the account. An agency must be configured for the function if this variable is used.

    X-CFF-Auth-Token

    Token of the account. An agency must be configured for the function if this variable is used.

    X-CFF-Invoke-Type

    Invocation type of the function.

    X-CFF-Secret-Key

    SK of the account. An agency must be configured for the function if this variable is used.

    X-CFF-Security-Token

    Security token of the account. An agency must be configured for the function if this variable is used.

  • Invocation Response

    Method – POST

    Path – http://$RUNTIME_API_ADDR/v1/runtime/invocation/response/$REQUEST_ID

    This API is used to send a successful invocation response to FunctionGraph. After the runtime invokes the function handler, it publishes the response from the function to the invocation response path.

  • Invocation Error

    Method – POST

    Path – http://$RUNTIME_API_ADDR/v1/runtime/invocation/error/$REQUEST_ID

    $REQUEST_ID is the value of variable X-Cff-Request-Id in the header of an event retrieval response. For more information, see Table 1.

    $RUNTIME_API_ADDR is a system environment variable. For more information, see Table 2.

    This API is used to send an error invocation response to FunctionGraph. After the runtime invokes the function handler, it publishes the response from the function to the invocation response path.

Runtime Environment Variables

You can use both custom and runtime environment variables in function code. The following table lists the runtime environment variables that are used in the FunctionGraph execution environment.

Table 2 Environment variables

Key

Description

RUNTIME_PROJECT_ID

Project ID

RUNTIME_FUNC_NAME

Function name

RUNTIME_FUNC_VERSION

Function version

RUNTIME_PACKAGE

App to which the function belongs

RUNTIME_HANDLER

Function handler

RUNTIME_TIMEOUT

Function timeout duration

RUNTIME_USERDATA

Value passed through an environment variable

RUNTIME_CPU

Number of allocated CPU cores

RUNTIME_MEMORY

Allocated memory

RUNTIME_CODE_ROOT

Directory that stores the function code

RUNTIME_API_ADDR

Host IP address and port of a custom runtime API

The value of a custom environment variable can be retrieved in the same way as the value of a FunctionGraph environment variable.

Example

This example contains one file called bootstrap. The file is implemented in Bash.

The runtime loads the function script from the deployment package by using two variables.

The bootstrap file is as follows:

#!/bin/sh
set -o pipefail
#Processing requests loop
while true
do
HEADERS="$(mktemp)"
  # Get an event
  EVENT_DATA=$(curl -sS -LD "$HEADERS" -X GET "http://$RUNTIME_API_ADDR/v1/runtime/invocation/request")
  # Get request id from response header
  REQUEST_ID=$(grep -Fi x-cff-request-id "$HEADERS" | tr -d '[:space:]' | cut -d: -f2)
  if [ -z "$REQUEST_ID" ]; then
    continue
  fi
  # Process request data
  RESPONSE="Echoing request: hello world!"
  # Put response
  curl -X POST "http://$RUNTIME_API_ADDR/v1/runtime/invocation/response/$REQUEST_ID" -d "$RESPONSE"
done

After loading the script, the runtime processes invocation events in a loop until it is terminated. It uses the API to retrieve invocation events from FunctionGraph, passes the events to the handler, and then sends responses back to FunctionGraph.

To obtain the request ID, the runtime saves the API response header in a temporary file, and then reads the request ID from the x-cff-request-id header field. The runtime processes the retrieved event data and sends a response back to FunctionGraph.

The following is an example of source code in Go. It can be executed only after compilation.

package main

import (
    "bytes"
    "encoding/json"
    "fmt"
    "io"
    "io/ioutil"
    "log"
    "net"
    "net/http"
    "os"
    "strings"
    "time"
)

var (
    getRequestUrl           = os.ExpandEnv("http://${RUNTIME_API_ADDR}/v1/runtime/invocation/request")
    putResponseUrl          = os.ExpandEnv("http://${RUNTIME_API_ADDR}/v1/runtime/invocation/response/{REQUEST_ID}")
    putErrorResponseUrl     = os.ExpandEnv("http://${RUNTIME_API_ADDR}/v1/runtime/invocation/error/{REQUEST_ID}")
    requestIdInvalidError   = fmt.Errorf("request id invalid")
    noRequestAvailableError = fmt.Errorf("no request available")
    putResponseFailedError  = fmt.Errorf("put response failed")
    functionPackage         = os.Getenv("RUNTIME_PACKAGE")
    functionName            = os.Getenv("RUNTIME_FUNC_NAME")
    functionVersion         = os.Getenv("RUNTIME_FUNC_VERSION")

    client = http.Client{
        Transport: &http.Transport{
            DialContext: (&net.Dialer{
                Timeout: 3 * time.Second,
            }).DialContext,
        },
    }
)

func main() {
    // main loop for processing requests.
    for {
        requestId, header, payload, err := getRequest()
        if err != nil {
            time.Sleep(50 * time.Millisecond)
            continue
        }

        result, err := processRequestEvent(requestId, header, payload)
        err = putResponse(requestId, result, err)
        if err != nil {
            log.Printf("put response failed, err: %s.", err.Error())
        }
    }
}

// event processing function
func processRequestEvent(requestId string, header http.Header, evtBytes []byte) ([]byte, error) {
    log.Printf("processing request '%s'.", requestId)
    result := fmt.Sprintf("function: %s:%s:%s, request id: %s, headers: %+v, payload: %s", functionPackage, functionName,
        functionVersion, requestId, header, string(evtBytes))
    
    var event FunctionEvent
    err := json.Unmarshal(evtBytes, &event)
    if err != nil {
        return (&ErrorMessage{ErrorType: "invalid event", ErrorMessage: "invalid json formated event"}).toJsonBytes(), err
    }

    return (&APIGFormatResult{StatusCode: 200, Body: result}).toJsonBytes(), nil
}

func getRequest() (string, http.Header, []byte, error) {
    resp, err := client.Get(getRequestUrl)
    if err != nil {
        log.Printf("get request error, err: %s.", err.Error())
        return "", nil, nil, err
    }
    defer resp.Body.Close()

    // get request id from response header
    requestId := resp.Header.Get("X-CFF-Request-Id")
    if requestId == "" {
        log.Printf("request id not found.")
        return "", nil, nil, requestIdInvalidError
    }

    payload, err := ioutil.ReadAll(resp.Body)
    if err != nil {
        log.Printf("read request body error, err: %s.", err.Error())
        return "", nil, nil, err
    }

    if resp.StatusCode != 200 {
        log.Printf("get request failed, status: %d, message: %s.", resp.StatusCode, string(payload))
        return "", nil, nil, noRequestAvailableError
    }

    log.Printf("get request ok.")
    return requestId, resp.Header, payload, nil
}

func putResponse(requestId string, payload []byte, err error) error {
    var body io.Reader
    if payload != nil && len(payload) > 0 {
        body = bytes.NewBuffer(payload)
    }

    url := ""
    if err == nil {
        url = strings.Replace(putResponseUrl, "{REQUEST_ID}", requestId, -1)
    } else {
        url = strings.Replace(putErrorResponseUrl, "{REQUEST_ID}", requestId, -1)
    }

    resp, err := client.Post(strings.Replace(url, "{REQUEST_ID}", requestId, -1), "", body)
    if err != nil {
        log.Printf("put response error, err: %s.", err.Error())
        return err
    }
    defer resp.Body.Close()

    responsePayload, err := ioutil.ReadAll(resp.Body)
    if err != nil {
        log.Printf("read request body error, err: %s.", err.Error())
        return err
    }

    if resp.StatusCode != 200 {
        log.Printf("put response failed, status: %d, message: %s.", resp.StatusCode, string(responsePayload))
        return putResponseFailedError
    }

    return nil
}

type FunctionEvent struct {
    Type string `json:"type"`
    Name string `json:"name"`
}

type APIGFormatResult struct {
    StatusCode      int               `json:"statusCode"`
    IsBase64Encoded bool              `json:"isBase64Encoded"`
    Headers         map[string]string `json:"headers,omitempty"`
    Body            string            `json:"body,omitempty"`
}

func (result *APIGFormatResult) toJsonBytes() []byte {
    data, err := json.MarshalIndent(result, "", "  ")
    if err != nil {
        return nil
    }

    return data
}

type ErrorMessage struct {
    ErrorType     string `json:"errorType"`
    ErrorMessage  string `json:"errorMessage"`
}

func (errMsg *ErrorMessage) toJsonBytes() []byte {
    data, err := json.MarshalIndent(errMsg, "", "  ")
    if err != nil {
        return nil
    }

    return data
}

Table 3 describes the environment variables used in the preceding code.

Table 3 Environment variables

Environment Variable

Description

RUNTIME_FUNC_NAME

Function name

RUNTIME_FUNC_VERSION

Function version

RUNTIME_PACKAGE

App to which the function belongs

Streaming Response

Custom runtime supports response of large packets whose size ranges from 6 MB to 200 MB in data streams.

  1. Log in to the FunctionGraph console, and click the name of the custom runtime function to go to the details page.
  2. Choose Configuration > Advanced Settings and enable Streaming Response.