go-sip/deploys/preprod/probe.go

package main

import (
	"context"
	"crypto/tls"
	"crypto/x509"
	"encoding/json"
	"errors"
	"flag"
	"fmt"
	"net"
	"net/http"
	"os"
	"syscall"
	"time"

	"git.ipao.vip/rogee/go-sip/internal/config"
	"git.ipao.vip/rogee/go-sip/internal/configread"
	"git.ipao.vip/rogee/go-sip/internal/rpc"
	amqp "github.com/rabbitmq/amqp091-go"
	"github.com/santhosh-tekuri/jsonschema/v6"
)

type input struct {
	SaaSURL        string `json:"saas_url"`
	DispatcherID   string `json:"dispatcher_id"`
	Secret         string `json:"secret"`
	MQHost         string `json:"mq_host"`
	MQPort         int    `json:"mq_port"`
	MQUser         string `json:"mq_user"`
	MQPassword     string `json:"mq_password"`
	MQVhost        string `json:"mq_vhost"`
	ConnectionName string `json:"connection_name"`
	HoldMQMS       int    `json:"hold_mq_ms"`
}

type report struct {
	Success            bool     `json:"success"`
	Phase              string   `json:"phase"`
	HTTPStatus         int      `json:"http_status"`
	ErrorClass         string   `json:"error_class,omitempty"`
	TransportFailure   string   `json:"transport_failure,omitempty"`
	SchemaLocations    []string `json:"schema_locations,omitempty"`
	Trunks             int      `json:"trunks"`
	Providers          int      `json:"providers"`
	Tasks              int      `json:"tasks"`
	ValidatedTasks     int      `json:"validated_tasks"`
	MQConnected        bool     `json:"mq_connected"`
	MQChannelClosed    bool     `json:"mq_channel_closed"`
	MQConnectionClosed bool     `json:"mq_connection_closed"`
}

type transport struct{ report *report }

func (t transport) RoundTrip(req *http.Request) (*http.Response, error) {
	response, err := http.DefaultTransport.RoundTrip(req)
	if response != nil {
		t.report.HTTPStatus = response.StatusCode
	}
	if err != nil {
		t.report.TransportFailure = transportFailure(err)
	}
	return response, err
}
func transportFailure(err error) string {
	var verification *tls.CertificateVerificationError
	var authority x509.UnknownAuthorityError
	var hostname x509.HostnameError
	var invalid x509.CertificateInvalidError
	var roots x509.SystemRootsError
	if errors.As(err, &verification) || errors.As(err, &authority) || errors.As(err, &hostname) || errors.As(err, &invalid) || errors.As(err, &roots) {
		return "tls_failed"
	}
	var network net.Error
	if errors.Is(err, syscall.ECONNREFUSED) || (errors.As(err, &network) && network.Timeout()) {
		return "unavailable"
	}
	return "transport_failed"
}
func rejected(r *report, err error) {
	r.Success = false
	r.ErrorClass = fmt.Sprintf("%T", err)
	var validation *jsonschema.ValidationError
	if errors.As(err, &validation) {
		var visit func(*jsonschema.ValidationError)
		visit = func(e *jsonschema.ValidationError) {
			if len(e.Causes) == 0 {
				r.SchemaLocations = append(r.SchemaLocations, fmt.Sprint(e.InstanceLocation))
			}
			for _, cause := range e.Causes {
				visit(cause)
			}
		}
		visit(validation)
	}
}
func check(in input) (r report) {
	ctx, cancel := context.WithTimeout(context.Background(), 90*time.Second)
	defer cancel()
	r.Phase = "mq_connect"
	uri := amqp.URI{Scheme: "amqp", Host: in.MQHost, Port: in.MQPort, Username: in.MQUser, Password: in.MQPassword, Vhost: in.MQVhost}
	conn, err := amqp.DialConfig(uri.String(), amqp.Config{Heartbeat: 5 * time.Second, Locale: "en_US", Properties: amqp.Table{"connection_name": in.ConnectionName}, Dial: func(network, addr string) (net.Conn, error) {
		c, err := (&net.Dialer{Timeout: 8 * time.Second}).DialContext(ctx, network, addr)
		if err != nil {
			return nil, err
		}
		if err := c.SetDeadline(time.Now().Add(15 * time.Second)); err != nil {
			return nil, errors.Join(err, c.Close())
		}
		return c, nil
	}})
	if err != nil {
		rejected(&r, err)
		return
	}
	r.MQConnected = true
	if in.HoldMQMS > 0 && in.HoldMQMS <= 10000 {
		time.Sleep(time.Duration(in.HoldMQMS) * time.Millisecond)
	}
	r.Phase = "mq_channel"
	channel, err := conn.Channel()
	if err != nil {
		rejected(&r, errors.Join(err, conn.Close()))
		return
	}
	if err := channel.Close(); err != nil {
		rejected(&r, errors.Join(err, conn.Close()))
		return
	}
	r.MQChannelClosed = true
	if err := conn.Close(); err != nil {
		rejected(&r, err)
		return
	}
	r.MQConnectionClosed = true
	r.Phase = "saas_client"
	client, err := configread.NewClient(in.SaaSURL, in.DispatcherID, in.Secret, &http.Client{Timeout: 15 * time.Second, Transport: transport{&r}})
	if err != nil {
		rejected(&r, err)
		return
	}
	r.Phase = "sip"
	sip, err := client.ReadSIP(ctx)
	if err != nil {
		rejected(&r, err)
		return
	}
	var trunks []json.RawMessage
	if err := json.Unmarshal(sip.Trunks, &trunks); err != nil {
		rejected(&r, err)
		return
	}
	r.Trunks = len(trunks)
	r.Phase = "providers"
	providers, err := client.ReadProviders(ctx)
	if err != nil {
		rejected(&r, err)
		return
	}
	r.Providers = len(providers)
	var tasks []configread.DiscoveredTask
	cursor := ""
	seen := map[string]bool{}
	for {
		r.Phase = "tasks"
		page, err := client.ReadTasks(ctx, cursor)
		if err != nil {
			rejected(&r, err)
			return
		}
		for _, task := range page.Tasks {
			if seen[task.TaskID] {
				rejected(&r, errors.New("duplicate discovery task"))
				return
			}
			seen[task.TaskID] = true
			tasks = append(tasks, task)
		}
		if page.Cursor == "" {
			break
		}
		cursor = page.Cursor
	}
	r.Tasks = len(tasks)
	for _, task := range tasks {
		r.Phase = "task_and_quota"
		if _, err := client.ReadTask(ctx, task.TaskID, task.TenantID, sip, providers); err != nil {
			rejected(&r, err)
			return
		}
		r.ValidatedTasks++
	}
	r.Phase = "complete"
	r.Success = true
	return
}

// environmentCheck opens only explicit local configuration and certificate
// files. It never opens SQLite, network connections, or cloud requests.
func environmentCheck(role string) (r report) {
	var ca, cert, key string
	r.Phase = role + "_environment"
	if role == "agent" {
		s, err := config.LoadAgentEnvironment("nonprod-real")
		if err != nil {
			rejected(&r, err)
			return
		}
		ca, cert, key = s.CAFile, s.CertFile, s.KeyFile
	} else if role == "dispatcher" {
		s, err := config.LoadDispatcherRuntimeEnvironment("nonprod-real")
		if err != nil {
			rejected(&r, err)
			return
		}
		r.Phase = "agent_inventory"
		if _, err := config.LoadMockAgentEndpoint(s.AgentEndpointsFile); err != nil {
			rejected(&r, err)
			return
		}
		r.Phase = "oss_configuration"
		if _, err := config.LoadNonprodRealOSSConfig(s.OSSConfigFile, s.DispatcherID); err != nil {
			rejected(&r, err)
			return
		}
		ca, cert, key = s.CAFile, s.CertFile, s.KeyFile
	} else {
		rejected(&r, errors.New("unknown environment role"))
		return
	}
	r.Phase = role + "_tls"
	caBytes, err := os.ReadFile(ca)
	if err != nil {
		rejected(&r, err)
		return
	}
	certBytes, err := os.ReadFile(cert)
	if err != nil {
		rejected(&r, err)
		return
	}
	keyBytes, err := os.ReadFile(key)
	if err != nil {
		rejected(&r, err)
		return
	}
	if _, err := rpc.NewServerTLSConfig(caBytes, certBytes, keyBytes); err != nil {
		rejected(&r, err)
		return
	}
	r.Phase = "environment_complete"
	r.Success = true
	return
}
func main() {
	role := flag.String("environment", "", "pure agent or dispatcher deployment configuration inspection")
	flag.Parse()
	if *role != "" {
		r := environmentCheck(*role)
		if err := json.NewEncoder(os.Stdout).Encode(r); err != nil {
			os.Exit(2)
		}
		if !r.Success {
			os.Exit(1)
		}
		return
	}
	var in input
	d := json.NewDecoder(os.Stdin)
	d.DisallowUnknownFields()
	if err := d.Decode(&in); err != nil {
		_ = json.NewEncoder(os.Stdout).Encode(report{Phase: "input", ErrorClass: fmt.Sprintf("%T", err)})
		os.Exit(1)
	}
	r := check(in)
	if err := json.NewEncoder(os.Stdout).Encode(r); err != nil {
		os.Exit(2)
	}
	if !r.Success {
		os.Exit(1)
	}
}