1
0
Fork 0
mirror of https://github.com/postmannen/ctrl.git synced 2024-12-14 12:37:31 +00:00
ctrl/server.go

540 lines
17 KiB
Go
Raw Normal View History

// Notes:
2021-02-01 10:13:38 +00:00
package steward
2021-01-25 14:23:00 +00:00
import (
"context"
2021-01-25 14:23:00 +00:00
"fmt"
"log"
2021-03-30 08:37:16 +00:00
"net"
2021-08-23 10:47:33 +00:00
"net/http"
"os"
2021-05-12 07:50:03 +00:00
"path/filepath"
2021-01-25 14:23:00 +00:00
"time"
"github.com/nats-io/nats.go"
"github.com/prometheus/client_golang/prometheus"
2021-01-25 14:23:00 +00:00
)
2021-02-26 06:55:28 +00:00
type processName string
// Will return a process name made up of subjectName+processKind
2021-02-26 06:55:28 +00:00
func processNameGet(sn subjectName, pk processKind) processName {
pn := fmt.Sprintf("%s_%s", sn, pk)
return processName(pn)
}
// server is the structure that will hold the state about spawned
// processes on a local instance.
2021-01-27 13:02:57 +00:00
type server struct {
// The main background context
ctx context.Context
// The CancelFunc for the main context
2021-08-11 10:23:37 +00:00
cancel context.CancelFunc
// Configuration options used for running the server
configuration *Configuration
// The nats connection to the broker
2021-01-27 13:02:57 +00:00
natsConn *nats.Conn
2021-07-02 06:38:44 +00:00
// net listener for communicating via the steward socket
2021-08-09 12:41:31 +00:00
StewardSocket net.Listener
// processes holds all the information about running processes
processes *processes
// The name of the node
nodeName string
2022-04-01 05:09:55 +00:00
// toRingBufferCh are the channel where new messages in a bulk
// format (slice) are put into the system.
2021-11-09 12:18:58 +00:00
//
// In general the ringbuffer will read this
// channel, unfold each slice, and put single messages on the buffer.
2022-04-01 05:09:55 +00:00
toRingBufferCh chan []subjectAndMessage
// directSAMSCh
directSAMSCh chan []subjectAndMessage
2021-02-24 09:58:02 +00:00
// errorKernel is doing all the error handling like what to do if
// an error occurs.
errorKernel *errorKernel
2021-09-15 06:39:34 +00:00
// Ring buffer
ringBuffer *ringBuffer
2021-02-18 11:29:14 +00:00
// metric exporter
metrics *metrics
// Version of package
version string
2022-01-08 03:19:51 +00:00
// tui client
tui *tui
// processInitial is the initial process that all other processes are tied to.
processInitial process
2022-04-21 11:21:36 +00:00
// nodeAuth holds all the signatures, the public keys and other components
// related to authentication on an individual node.
nodeAuth *nodeAuth
// helloRegister is a register of all the nodes that have sent hello messages
// to the central server
helloRegister *helloRegister
// holds the logic for the central auth services
centralAuth *centralAuth
2021-01-27 13:02:57 +00:00
}
// newServer will prepare and return a server type
2022-02-07 03:23:13 +00:00
func NewServer(configuration *Configuration, version string) (*server, error) {
2021-08-11 10:23:37 +00:00
// Set up the main background context.
ctx, cancel := context.WithCancel(context.Background())
2022-02-07 03:23:13 +00:00
metrics := newMetrics(configuration.PromHostAndPort)
// Start the error kernel that will do all the error handling
// that is not done within a process.
errorKernel := newErrorKernel(ctx, metrics)
2021-04-19 19:06:37 +00:00
var opt nats.Option
2021-08-16 11:01:12 +00:00
2022-02-07 03:23:13 +00:00
if configuration.RootCAPath != "" {
opt = nats.RootCAs(configuration.RootCAPath)
2021-04-19 19:06:37 +00:00
}
2022-02-07 03:23:13 +00:00
if configuration.NkeySeedFile != "" {
2021-05-20 10:27:25 +00:00
var err error
2021-08-11 10:23:37 +00:00
2022-02-07 03:23:13 +00:00
opt, err = nats.NkeyOptionFromSeed(configuration.NkeySeedFile)
2021-05-20 10:27:25 +00:00
if err != nil {
cancel()
2021-05-20 10:27:25 +00:00
return nil, fmt.Errorf("error: failed to read nkey seed file: %v", err)
}
}
var conn *nats.Conn
2021-08-16 11:01:12 +00:00
// Connect to the nats server, and retry until succesful.
for {
var err error
// Setting MaxReconnects to -1 which equals unlimited.
2022-02-07 03:23:13 +00:00
conn, err = nats.Connect(configuration.BrokerAddress,
2021-12-17 16:28:57 +00:00
opt,
2022-06-21 08:40:32 +00:00
//nats.FlusherTimeout(time.Second*10),
2021-12-17 16:28:57 +00:00
nats.MaxReconnects(-1),
2022-02-07 03:23:13 +00:00
nats.ReconnectJitter(time.Duration(configuration.NatsReconnectJitter)*time.Millisecond, time.Duration(configuration.NatsReconnectJitterTLS)*time.Second),
nats.Timeout(time.Second*time.Duration(configuration.NatsConnOptTimeout)),
2021-12-17 16:28:57 +00:00
)
// If no servers where available, we loop and retry until succesful.
if err != nil {
2022-02-07 03:23:13 +00:00
log.Printf("error: could not connect, waiting %v seconds, and retrying: %v\n", configuration.NatsConnectRetryInterval, err)
time.Sleep(time.Duration(time.Second * time.Duration(configuration.NatsConnectRetryInterval)))
continue
}
2021-02-01 10:13:38 +00:00
break
}
2021-08-11 10:23:37 +00:00
2021-12-16 10:01:01 +00:00
log.Printf(" * conn.Opts.ReconnectJitterTLS: %v\n", conn.Opts.ReconnectJitterTLS)
log.Printf(" * conn.Opts.ReconnectJitter: %v\n", conn.Opts.ReconnectJitter)
2022-01-07 06:36:19 +00:00
var stewardSocket net.Listener
var err error
2022-12-21 06:17:47 +00:00
// Check if tmp folder for socket exists, if not create it
if _, err := os.Stat(configuration.SocketFolder); os.IsNotExist(err) {
err := os.MkdirAll(configuration.SocketFolder, 0700)
if err != nil {
cancel()
return nil, fmt.Errorf("error: failed to create socket folder directory %v: %v", configuration.SocketFolder, err)
}
}
// Open the steward socket file, and start the listener if enabled.
2022-02-07 03:23:13 +00:00
if configuration.EnableSocket {
stewardSocket, err = createSocket(configuration.SocketFolder, "steward.sock")
2022-01-07 06:36:19 +00:00
if err != nil {
cancel()
return nil, err
}
2021-03-30 08:37:16 +00:00
}
2022-01-08 03:19:51 +00:00
// Create the tui client structure if enabled.
var tuiClient *tui
2022-02-07 03:23:13 +00:00
if configuration.EnableTUI {
tuiClient, err = newTui(Node(configuration.NodeName))
2022-01-08 03:19:51 +00:00
if err != nil {
cancel()
return nil, err
}
}
//var nodeAuth *nodeAuth
//if configuration.EnableSignatureCheck {
2022-04-21 11:21:36 +00:00
nodeAuth := newNodeAuth(configuration, errorKernel)
2022-02-11 08:04:14 +00:00
// fmt.Printf(" * DEBUG: newServer: signatures contains: %+v\n", signatures)
//}
//var centralAuth *centralAuth
//if configuration.IsCentralAuth {
centralAuth := newCentralAuth(configuration, errorKernel)
//}
2022-04-01 05:09:55 +00:00
s := server{
ctx: ctx,
cancel: cancel,
configuration: configuration,
nodeName: configuration.NodeName,
natsConn: conn,
StewardSocket: stewardSocket,
toRingBufferCh: make(chan []subjectAndMessage),
directSAMSCh: make(chan []subjectAndMessage),
2022-04-01 05:09:55 +00:00
metrics: metrics,
version: version,
tui: tuiClient,
errorKernel: errorKernel,
2022-04-21 11:21:36 +00:00
nodeAuth: nodeAuth,
2022-04-01 05:09:55 +00:00
helloRegister: newHelloRegister(),
centralAuth: centralAuth,
}
2022-04-01 05:09:55 +00:00
s.processes = newProcesses(ctx, &s)
// Create the default data folder for where subscribers should
2021-03-25 11:50:58 +00:00
// write it's data, check if data folder exist, and create it if needed.
2022-02-07 03:23:13 +00:00
if _, err := os.Stat(configuration.SubscribersDataFolder); os.IsNotExist(err) {
if configuration.SubscribersDataFolder == "" {
return nil, fmt.Errorf("error: subscribersDataFolder value is empty, you need to provide the config or the flag value at startup %v: %v", configuration.SubscribersDataFolder, err)
}
2022-02-07 03:23:13 +00:00
err := os.Mkdir(configuration.SubscribersDataFolder, 0700)
if err != nil {
2022-02-07 03:23:13 +00:00
return nil, fmt.Errorf("error: failed to create data folder directory %v: %v", configuration.SubscribersDataFolder, err)
}
er := fmt.Errorf("info: creating subscribers data folder at %v", configuration.SubscribersDataFolder)
s.errorKernel.logConsoleOnlyIfDebug(er, s.configuration)
}
2022-04-01 05:09:55 +00:00
return &s, nil
2021-02-05 06:25:12 +00:00
}
// helloRegister is a register of all the nodes that have sent hello messages.
type helloRegister struct {
}
func newHelloRegister() *helloRegister {
h := helloRegister{}
return &h
}
// create socket will create a socket file, and return the net.Listener to
// communicate with that socket.
func createSocket(socketFolder string, socketFileName string) (net.Listener, error) {
// Just as an extra check we eventually delete any existing Steward socket files if found.
socketFilepath := filepath.Join(socketFolder, socketFileName)
if _, err := os.Stat(socketFilepath); !os.IsNotExist(err) {
err = os.Remove(socketFilepath)
if err != nil {
er := fmt.Errorf("error: could not delete sock file: %v", err)
return nil, er
}
}
// Open the socket.
nl, err := net.Listen("unix", socketFilepath)
if err != nil {
er := fmt.Errorf("error: failed to open socket: %v", err)
return nil, er
}
return nl, nil
}
2021-02-24 09:58:02 +00:00
// Start will spawn up all the predefined subscriber processes.
2021-02-10 06:25:44 +00:00
// Spawning of publisher processes is done on the fly by checking
2021-02-24 09:58:02 +00:00
// if there is publisher process for a given message subject, and
2021-08-16 11:01:12 +00:00
// if it does not exist it will spawn one.
func (s *server) Start() {
2021-09-23 03:46:25 +00:00
log.Printf("Starting steward, version=%+v\n", s.version)
s.metrics.promVersion.With(prometheus.Labels{"version": string(s.version)})
2021-08-04 08:37:24 +00:00
go func() {
2022-04-01 05:09:55 +00:00
err := s.errorKernel.start(s.toRingBufferCh)
2021-08-04 08:37:24 +00:00
if err != nil {
log.Printf("%v\n", err)
}
}()
2021-02-18 11:29:14 +00:00
// Start collecting the metrics
2021-08-03 11:57:29 +00:00
go func() {
2021-08-04 08:37:24 +00:00
err := s.metrics.start()
2021-08-03 11:57:29 +00:00
if err != nil {
log.Printf("%v\n", err)
os.Exit(1)
}
}()
2021-02-18 11:29:14 +00:00
// Start the checking the input socket for new messages from operator.
2022-01-07 06:36:19 +00:00
if s.configuration.EnableSocket {
go s.readSocket()
}
2021-09-10 03:26:16 +00:00
// Check if we should start the tcp listener for new messages from operator.
if s.configuration.TCPListener != "" {
go s.readTCPListener()
}
2021-09-10 03:26:16 +00:00
// Check if we should start the http listener for new messages from operator.
if s.configuration.HTTPListener != "" {
go s.readHttpListener()
}
// Start up the predefined subscribers.
//
// Since all the logic to handle processes are tied to the process
// struct, we need to create an initial process to start the rest.
2021-08-11 10:23:37 +00:00
//
// NB: The context of the initial process are set in processes.Start.
2021-08-09 07:18:30 +00:00
sub := newSubject(REQInitial, s.nodeName)
2022-04-01 05:09:55 +00:00
s.processInitial = newProcess(context.TODO(), s, sub, "", nil)
// Start all wanted subscriber processes.
s.processes.Start(s.processInitial)
2021-08-09 07:18:30 +00:00
time.Sleep(time.Second * 1)
s.processes.printProcessesMap()
2021-08-23 10:47:33 +00:00
// Start exposing the the data folder via HTTP if flag is set.
if s.configuration.ExposeDataFolder != "" {
log.Printf("info: Starting expose of data folder via HTTP\n")
go s.exposeDataFolder(s.ctx)
}
2022-01-08 03:19:51 +00:00
if s.configuration.EnableTUI {
go func() {
2022-04-01 05:09:55 +00:00
err := s.tui.Start(s.ctx, s.toRingBufferCh)
2022-01-08 03:19:51 +00:00
if err != nil {
log.Printf("%v\n", err)
os.Exit(1)
}
}()
}
2021-08-09 07:18:30 +00:00
// Start the processing of new messages from an input channel.
2021-09-15 06:39:34 +00:00
// NB: We might need to create a sub context for the ringbuffer here
// so we can cancel this context last, and not use the server.
2021-08-25 06:50:24 +00:00
s.routeMessagesToProcess("./incomingBuffer.db")
2021-02-10 06:25:44 +00:00
// Start reading the channel for injecting direct messages that should
// not be sent via the message broker.
s.directSAMSChRead()
2022-01-26 08:23:02 +00:00
// Check and enable read the messages specified in the startup folder.
s.readStartupFolder()
2021-08-09 07:18:30 +00:00
}
// directSAMSChRead for injecting messages directly in to the local system
// without sending them via the message broker.
func (s *server) directSAMSChRead() {
go func() {
for {
select {
case <-s.ctx.Done():
log.Printf("info: stopped the directSAMSCh reader\n\n")
return
case sams := <-s.directSAMSCh:
2022-11-01 08:54:35 +00:00
// fmt.Printf(" * DEBUG: directSAMSChRead: <- sams = %v\n", sams)
// Range over all the sams, find the process, check if the method exists, and
// handle the message by starting the correct method handler.
for i := range sams {
processName := processNameGet(sams[i].Subject.name(), processKindSubscriber)
s.processes.active.mu.Lock()
p := s.processes.active.procNames[processName]
s.processes.active.mu.Unlock()
mh, ok := p.methodsAvailable.CheckIfExists(sams[i].Message.Method)
if !ok {
er := fmt.Errorf("error: subscriberHandler: method type not available: %v", p.subject.Event)
p.errorKernel.errSend(p, sams[i].Message, er)
continue
}
p.handler = mh.handler
go executeHandler(p, sams[i].Message, s.nodeName)
}
}
}
}()
}
2021-08-09 07:18:30 +00:00
// Will stop all processes started during startup.
func (s *server) Stop() {
// Stop the started pub/sub message processes.
2021-08-11 10:23:37 +00:00
s.processes.Stop()
2021-08-09 07:18:30 +00:00
log.Printf("info: stopped all subscribers\n")
2021-07-02 06:38:44 +00:00
2021-08-09 07:18:30 +00:00
// Stop the errorKernel.
s.errorKernel.stop()
log.Printf("info: stopped the errorKernel\n")
2021-07-02 11:26:30 +00:00
2021-08-09 07:18:30 +00:00
// Stop the main context.
2021-08-11 10:23:37 +00:00
s.cancel()
2021-08-09 07:18:30 +00:00
log.Printf("info: stopped the main context\n")
2021-08-09 12:41:31 +00:00
2022-01-05 18:53:26 +00:00
// Delete the steward socket file when the program exits.
2021-08-09 12:41:31 +00:00
socketFilepath := filepath.Join(s.configuration.SocketFolder, "steward.sock")
if _, err := os.Stat(socketFilepath); !os.IsNotExist(err) {
err = os.Remove(socketFilepath)
if err != nil {
er := fmt.Errorf("error: could not delete sock file: %v", err)
log.Printf("%v\n", er)
}
}
}
2022-06-17 22:15:18 +00:00
// samDBValueAndDelivered Contains the sam value as it is used in the
// state DB, and also a delivered function to be called when this message
// is picked up, so we can control if messages gets stale at some point.
type samDBValueAndDelivered struct {
2021-07-02 16:32:01 +00:00
samDBValue samDBValue
delivered func()
2021-07-02 16:32:01 +00:00
}
2021-08-25 06:50:24 +00:00
// routeMessagesToProcess takes a database name it's input argument.
// The database will be used as the persistent k/v store for the work
// queue which is implemented as a ring buffer.
// The ringBufferInCh are where we get new messages to publish.
2021-03-09 06:43:55 +00:00
// Incomming messages will be routed to the correct subject process, where
// the handling of each nats subject is handled within it's own separate
// worker process.
// It will also handle the process of spawning more worker processes
// for publisher subjects if it does not exist.
2021-08-25 06:50:24 +00:00
func (s *server) routeMessagesToProcess(dbFileName string) {
2021-03-09 06:43:55 +00:00
// Prepare and start a new ring buffer
2021-12-21 06:21:12 +00:00
var bufferSize int = s.configuration.RingBufferSize
const samValueBucket string = "samValueBucket"
const indexValueBucket string = "indexValueBucket"
2022-04-01 05:09:55 +00:00
s.ringBuffer = newringBuffer(s.ctx, s.metrics, s.configuration, bufferSize, dbFileName, Node(s.nodeName), s.toRingBufferCh, samValueBucket, indexValueBucket, s.errorKernel, s.processInitial)
2021-08-26 05:02:36 +00:00
ringBufferInCh := make(chan subjectAndMessage)
ringBufferOutCh := make(chan samDBValueAndDelivered)
2021-03-09 06:43:55 +00:00
// start the ringbuffer.
s.ringBuffer.start(s.ctx, ringBufferInCh, ringBufferOutCh)
2021-03-09 06:43:55 +00:00
// Start reading new fresh messages received on the incomming message
// pipe/file requested, and fill them into the buffer.
2022-01-20 06:50:58 +00:00
// Since the new messages comming into the system is a []subjectAndMessage
// we loop here, unfold the slice, and put single subjectAndMessages's on
// the channel to the ringbuffer.
2021-03-09 06:43:55 +00:00
go func() {
2022-04-01 05:09:55 +00:00
for sams := range s.toRingBufferCh {
2021-08-25 08:16:55 +00:00
for _, sam := range sams {
ringBufferInCh <- sam
2021-03-09 06:43:55 +00:00
}
}
close(ringBufferInCh)
2021-03-09 06:43:55 +00:00
}()
// Process the messages that are in the ring buffer. Check and
// send if there are a specific subject for it, and if no subject
// exist throw an error.
2022-01-27 06:19:04 +00:00
var event Event
eventAvailable := event.CheckEventAvailable()
2021-03-09 06:43:55 +00:00
var method Method
methodsAvailable := method.GetMethodsAvailable()
go func() {
2021-11-09 12:18:58 +00:00
for samDBVal := range ringBufferOutCh {
go func(samDBVal samDBValueAndDelivered) {
// Signal back to the ringbuffer that message have been picked up.
samDBVal.delivered()
sam := samDBVal.samDBValue.Data
// Check if the format of the message is correct.
if _, ok := methodsAvailable.CheckIfExists(sam.Message.Method); !ok {
er := fmt.Errorf("error: routeMessagesToProcess: the method do not exist, message dropped: %v", sam.Message.Method)
s.errorKernel.errSend(s.processInitial, sam.Message, er)
return
}
if !eventAvailable.CheckIfExists(sam.Subject.Event, sam.Subject) {
er := fmt.Errorf("error: routeMessagesToProcess: the event type do not exist, message dropped: %v", sam.Message.Method)
s.errorKernel.errSend(s.processInitial, sam.Message, er)
return
}
2021-03-09 06:43:55 +00:00
for {
// Looping here so we are able to redo the sending
// of the last message if a process for the specified subject
// is not present. The process will then be created, and
// the code will loop back here.
m := sam.Message
2021-11-09 13:01:42 +00:00
subjName := sam.Subject.name()
pn := processNameGet(subjName, processKindPublisher)
2021-09-21 21:29:42 +00:00
// Check if there is a map of type map[int]process registered
// for the processName, and if it exists then return it.
s.processes.active.mu.Lock()
proc, ok := s.processes.active.procNames[pn]
s.processes.active.mu.Unlock()
// If found a map above, range it, and are there already a process
// for that subject, put the message on that processes incomming
// message channel.
if ok {
// We have found the process to route the message to, deliver it.
proc.subject.messageCh <- m
break
} else {
// If a publisher process do not exist for the given subject, create it.
// log.Printf("info: processNewMessages: did not find that specific subject, starting new process for subject: %v\n", subjName)
sub := newSubject(sam.Subject.Method, sam.Subject.ToNode)
var proc process
switch {
case m.IsSubPublishedMsg:
proc = newSubProcess(s.ctx, s, sub, processKindPublisher, nil)
default:
proc = newProcess(s.ctx, s, sub, processKindPublisher, nil)
}
proc.spawnWorker()
er := fmt.Errorf("info: processNewMessages: new process started, subject: %v, processID: %v", subjName, proc.processID)
s.errorKernel.logConsoleOnlyIfDebug(er, s.configuration)
// Now when the process is spawned we continue,
// and send the message to that new process.
continue
}
2021-09-21 21:29:42 +00:00
}
}(samDBVal)
2021-03-09 06:43:55 +00:00
}
}()
}
2021-08-23 10:47:33 +00:00
func (s *server) exposeDataFolder(ctx context.Context) {
2021-08-23 15:05:56 +00:00
fileHandler := func(w http.ResponseWriter, r *http.Request) {
// w.Header().Set("Content-Type", "text/html")
http.FileServer(http.Dir(s.configuration.SubscribersDataFolder)).ServeHTTP(w, r)
}
2021-08-23 10:47:33 +00:00
//create a file server, and serve the files found in ./
2021-08-23 15:05:56 +00:00
//fd := http.FileServer(http.Dir(s.configuration.SubscribersDataFolder))
http.HandleFunc("/", fileHandler)
2021-08-23 10:47:33 +00:00
// we create a net.Listen type to use later with the http.Serve function.
nl, err := net.Listen("tcp", s.configuration.ExposeDataFolder)
if err != nil {
log.Println("error: starting net.Listen: ", err)
}
// start the web server with http.Serve instead of the usual http.ListenAndServe
err = http.Serve(nl, nil)
if err != nil {
log.Printf("Error: failed to start web server: %v\n", err)
}
os.Exit(1)
}