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

456 lines
14 KiB
Go
Raw Normal View History

2021-02-24 09:58:02 +00:00
package steward
import (
"context"
2021-02-24 09:58:02 +00:00
"fmt"
"log"
"sync"
"time"
"github.com/prometheus/client_golang/prometheus"
2021-02-24 09:58:02 +00:00
)
// processes holds all the information about running processes
type processes struct {
2021-08-11 10:23:37 +00:00
// The main context for subscriber processes.
ctx context.Context
// cancel func to send cancel signal to the subscriber processes context.
cancel context.CancelFunc
// The active spawned processes
2021-10-08 10:07:10 +00:00
active procsMap
// mutex to lock the map
2021-10-08 20:39:46 +00:00
// mu sync.RWMutex
// The last processID created
lastProcessID int
2021-08-18 10:16:21 +00:00
// The instance global prometheus registry.
metrics *metrics
2021-08-12 07:21:56 +00:00
// Waitgroup to keep track of all the processes started
wg sync.WaitGroup
}
// newProcesses will prepare and return a *processes which
// is map containing all the currently running processes.
2021-08-18 10:16:21 +00:00
func newProcesses(ctx context.Context, metrics *metrics) *processes {
p := processes{
2021-10-08 10:07:10 +00:00
active: *newProcsMap(),
}
2021-08-16 11:01:12 +00:00
// Prepare the parent context for the subscribers.
2021-08-11 10:23:37 +00:00
ctx, cancel := context.WithCancel(ctx)
2021-10-08 11:42:19 +00:00
// // Start the processes map.
// go func() {
// p.active.run(ctx)
// }()
2021-10-08 10:07:10 +00:00
2021-08-11 10:23:37 +00:00
p.ctx = ctx
p.cancel = cancel
2021-08-18 10:16:21 +00:00
p.metrics = metrics
return &p
}
// ----------------------
type keyValue struct {
2021-10-08 10:07:10 +00:00
k processName
v map[int]process
ok bool
}
type kvCh chan keyValue
type getValue struct {
2021-10-08 10:07:10 +00:00
k processName
kvCh kvCh
}
type procsMap struct {
2021-10-08 11:42:19 +00:00
m map[processName]map[int]process
mu sync.Mutex
}
func newProcsMap() *procsMap {
cM := procsMap{
2021-10-08 11:42:19 +00:00
m: make(map[processName]map[int]process),
}
return &cM
}
2021-10-08 11:42:19 +00:00
//func (c *procsMap) run(ctx context.Context) {
// for {
// select {
// case kvCh := <-c.mInCh:
// kv := <-kvCh
// c.m[kv.k] = kv.v
//
// case gv := <-c.mGetCh:
// v, ok := c.m[gv.k]
// gv.kvCh <- keyValue{gv.k, v, ok}
//
// case kvCh := <-c.mDelCh:
// kv := <-kvCh
// delete(c.m, kv.k)
//
// case gaCh := <-c.mGetAllCh:
// kvSlice := []keyValue{}
//
// for k, v := range c.m {
// kv := keyValue{k: k, v: v}
// kvSlice = append(kvSlice, kv)
// }
//
// gaCh <- kvSlice
//
// case <-ctx.Done():
// log.Printf("info: processes active.Run: got ctx.Done\n")
// return
// }
// }
//}
func (c *procsMap) put(kv keyValue) {
2021-10-08 11:42:19 +00:00
c.mu.Lock()
c.m[kv.k] = kv.v
c.mu.Unlock()
}
2021-10-08 10:07:10 +00:00
func (c *procsMap) get(key processName) keyValue {
2021-10-08 11:42:19 +00:00
var kv keyValue
2021-10-08 11:42:19 +00:00
c.mu.Lock()
kv.v, kv.ok = c.m[key]
c.mu.Unlock()
2021-10-08 11:42:19 +00:00
return kv
}
func (c *procsMap) del(kv keyValue) {
2021-10-08 11:42:19 +00:00
c.mu.Lock()
delete(c.m, kv.k)
c.mu.Unlock()
}
func (c *procsMap) getAll() []keyValue {
2021-10-08 11:42:19 +00:00
var kvs []keyValue
c.mu.Lock()
for k, v := range c.m {
kvs = append(kvs, keyValue{k: k, v: v})
}
c.mu.Unlock()
2021-10-08 11:42:19 +00:00
return kvs
}
// ----------------------
// Start all the subscriber processes.
// Takes an initial process as it's input. All processes
// will be tied to this single process's context.
func (p *processes) Start(proc process) {
2021-08-11 10:23:37 +00:00
// Set the context for the initial process.
proc.ctx = p.ctx
// --- Subscriber services that can be started via flags
// Allways start the listeners for Op commands
{
log.Printf("Starting REQOpCommand subscriber: %#v\n", proc.node)
sub := newSubject(REQOpCommand, string(proc.node))
proc := newProcess(proc.ctx, p.metrics, proc.natsConn, p, proc.toRingbufferCh, proc.configuration, sub, proc.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(proc.processes, proc.natsConn)
}
{
log.Printf("Starting REQOpProcessList subscriber: %#v\n", proc.node)
sub := newSubject(REQOpProcessList, string(proc.node))
proc := newProcess(proc.ctx, p.metrics, proc.natsConn, p, proc.toRingbufferCh, proc.configuration, sub, proc.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(proc.processes, proc.natsConn)
}
{
log.Printf("Starting REQOpProcessStart subscriber: %#v\n", proc.node)
sub := newSubject(REQOpProcessStart, string(proc.node))
proc := newProcess(proc.ctx, p.metrics, proc.natsConn, p, proc.toRingbufferCh, proc.configuration, sub, proc.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(proc.processes, proc.natsConn)
}
2021-09-20 09:53:17 +00:00
{
log.Printf("Starting REQOpProcessStop subscriber: %#v\n", proc.node)
sub := newSubject(REQOpProcessStop, string(proc.node))
proc := newProcess(proc.ctx, p.metrics, proc.natsConn, p, proc.toRingbufferCh, proc.configuration, sub, proc.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(proc.processes, proc.natsConn)
}
2021-02-24 09:58:02 +00:00
// Start a subscriber for textLogging messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQToFileAppend {
proc.startup.subREQToFileAppend(proc)
2021-02-24 09:58:02 +00:00
}
2021-04-06 17:42:03 +00:00
// Start a subscriber for text to file messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQToFile {
proc.startup.subREQToFile(proc)
2021-04-06 17:42:03 +00:00
}
2021-04-05 06:37:24 +00:00
// Start a subscriber for Hello messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQHello {
proc.startup.subREQHello(proc)
2021-02-24 09:58:02 +00:00
}
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQErrorLog {
2021-04-06 05:56:49 +00:00
// Start a subscriber for REQErrorLog messages
proc.startup.subREQErrorLog(proc)
}
2021-04-06 04:08:26 +00:00
// Start a subscriber for Ping Request messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQPing {
proc.startup.subREQPing(proc)
}
2021-04-06 04:08:26 +00:00
// Start a subscriber for REQPong messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQPong {
proc.startup.subREQPong(proc)
}
2021-04-04 09:19:17 +00:00
// Start a subscriber for REQCliCommand messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQCliCommand {
proc.startup.subREQCliCommand(proc)
}
// Start a subscriber for CLICommandReply messages
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQToConsole {
proc.startup.subREQToConsole(proc)
}
if proc.configuration.StartPubREQHello != 0 {
proc.startup.pubREQHello(proc)
}
2021-04-06 17:42:03 +00:00
// Start a subscriber for Http Get Requests
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQHttpGet {
proc.startup.subREQHttpGet(proc)
2021-04-06 17:42:03 +00:00
}
2021-04-13 09:28:52 +00:00
2021-09-08 16:56:23 +00:00
if proc.configuration.StartSubREQTailFile {
proc.startup.subREQTailFile(proc)
2021-04-13 09:28:52 +00:00
}
if proc.configuration.StartSubREQCliCommandCont {
proc.startup.subREQCliCommandCont(proc)
}
proc.startup.subREQToSocket(proc)
2021-02-24 09:58:02 +00:00
}
2021-08-11 10:23:37 +00:00
// Stop all subscriber processes.
func (p *processes) Stop() {
log.Printf("info: canceling all subscriber processes...\n")
2021-08-11 10:23:37 +00:00
p.cancel()
p.wg.Wait()
log.Printf("info: done canceling all subscriber processes.\n")
2021-08-11 10:23:37 +00:00
}
// ---------------------------------------------------------------------------------------
// Startup holds all the startup methods for subscribers.
2021-08-18 10:16:21 +00:00
type startup struct {
metrics *metrics
}
func newStartup(metrics *metrics) *startup {
s := startup{metrics: metrics}
return &s
}
func (s startup) subREQHttpGet(p process) {
log.Printf("Starting Http Get subscriber: %#v\n", p.node)
sub := newSubject(REQHttpGet, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) pubREQHello(p process) {
log.Printf("Starting Hello Publisher: %#v\n", p.node)
sub := newSubject(REQHello, p.configuration.CentralNodeName)
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindPublisher, nil)
// Define the procFunc to be used for the process.
proc.procFunc = procFunc(
func(ctx context.Context) error {
ticker := time.NewTicker(time.Second * time.Duration(p.configuration.StartPubREQHello))
for {
// fmt.Printf("--- DEBUG : procFunc call:kind=%v, Subject=%v, toNode=%v\n", proc.processKind, proc.subject, proc.subject.ToNode)
d := fmt.Sprintf("Hello from %v\n", p.node)
m := Message{
2021-09-08 04:15:51 +00:00
FileName: "hello.log",
Directory: "hello-messages",
ToNode: Node(p.configuration.CentralNodeName),
FromNode: Node(p.node),
Data: []string{d},
Method: REQHello,
ACKTimeout: 10,
Retries: 1,
}
2021-08-25 06:56:44 +00:00
sam, err := newSubjectAndMessage(m)
if err != nil {
// In theory the system should drop the message before it reaches here.
sendErrorLogMessage(proc.configuration, s.metrics, proc.toRingbufferCh, Node(proc.node), err)
log.Printf("error: ProcessesStart: %v\n", err)
}
proc.toRingbufferCh <- []subjectAndMessage{sam}
select {
case <-ticker.C:
case <-ctx.Done():
2021-08-03 10:42:50 +00:00
er := fmt.Errorf("info: stopped handleFunc for: publisher %v", proc.subject.name())
2021-07-02 16:32:01 +00:00
// sendErrorLogMessage(proc.toRingbufferCh, proc.node, er)
log.Printf("%v\n", er)
return nil
}
}
})
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQToConsole(p process) {
log.Printf("Starting Text To Console subscriber: %#v\n", p.node)
sub := newSubject(REQToConsole, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQCliCommand(p process) {
log.Printf("Starting CLICommand Request subscriber: %#v\n", p.node)
sub := newSubject(REQCliCommand, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQPong(p process) {
log.Printf("Starting Pong subscriber: %#v\n", p.node)
sub := newSubject(REQPong, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQPing(p process) {
log.Printf("Starting Ping Request subscriber: %#v\n", p.node)
sub := newSubject(REQPing, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQErrorLog(p process) {
log.Printf("Starting REQErrorLog subscriber: %#v\n", p.node)
sub := newSubject(REQErrorLog, "errorCentral")
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQHello(p process) {
log.Printf("Starting Hello subscriber: %#v\n", p.node)
sub := newSubject(REQHello, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
proc.procFuncCh = make(chan Message)
// The reason for running the say hello subscriber as a procFunc is that
// a handler are not able to hold state, and we need to hold the state
// of the nodes we've received hello's from in the sayHelloNodes map,
// which is the information we pass along to generate metrics.
proc.procFunc = func(ctx context.Context) error {
2021-06-29 06:21:42 +00:00
sayHelloNodes := make(map[Node]struct{})
2021-04-12 08:51:26 +00:00
for {
// Receive a copy of the message sent from the method handler.
var m Message
select {
case m = <-proc.procFuncCh:
case <-ctx.Done():
2021-08-03 10:42:50 +00:00
er := fmt.Errorf("info: stopped handleFunc for: subscriber %v", proc.subject.name())
2021-07-02 16:32:01 +00:00
// sendErrorLogMessage(proc.toRingbufferCh, proc.node, er)
log.Printf("%v\n", er)
return nil
}
// Add an entry for the node in the map
sayHelloNodes[m.FromNode] = struct{}{}
// update the prometheus metrics
2021-08-18 13:41:53 +00:00
s.metrics.promHelloNodesTotal.Set(float64(len(sayHelloNodes)))
s.metrics.promHelloNodesContactLast.With(prometheus.Labels{"nodeName": string(m.FromNode)}).SetToCurrentTime()
}
}
go proc.spawnWorker(p.processes, p.natsConn)
}
2021-04-13 15:15:13 +00:00
func (s startup) subREQToFile(p process) {
log.Printf("Starting text to file subscriber: %#v\n", p.node)
2021-04-13 15:15:13 +00:00
sub := newSubject(REQToFile, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQToFileAppend(p process) {
log.Printf("Starting text logging subscriber: %#v\n", p.node)
sub := newSubject(REQToFileAppend, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
2021-04-13 09:28:52 +00:00
func (s startup) subREQTailFile(p process) {
log.Printf("Starting tail log files subscriber: %#v\n", p.node)
2021-04-13 09:28:52 +00:00
sub := newSubject(REQTailFile, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
2021-04-13 09:28:52 +00:00
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQCliCommandCont(p process) {
log.Printf("Starting cli command with continous delivery: %#v\n", p.node)
sub := newSubject(REQCliCommandCont, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
func (s startup) subREQToSocket(p process) {
log.Printf("Starting write to socket subscriber: %#v\n", p.node)
sub := newSubject(REQToSocket, string(p.node))
proc := newProcess(p.ctx, s.metrics, p.natsConn, p.processes, p.toRingbufferCh, p.configuration, sub, p.errorCh, processKindSubscriber, nil)
// fmt.Printf("*** %#v\n", proc)
go proc.spawnWorker(p.processes, p.natsConn)
}
2021-08-11 10:23:37 +00:00
// ---------------------------------------------------------------
// Print the content of the processes map.
func (p *processes) printProcessesMap() {
log.Printf("*** Output of processes map :\n")
2021-10-08 10:07:10 +00:00
activeProcs := p.active.getAll()
for _, vSub := range activeProcs {
for _, vID := range vSub.v {
log.Printf("* proc - : %v, id: %v, name: %v\n", vID.processKind, vID.processID, vID.subject.name())
2021-08-11 10:23:37 +00:00
}
}
2021-10-08 10:07:10 +00:00
p.metrics.promProcessesTotal.Set(float64(len(activeProcs)))
2021-08-11 10:23:37 +00:00
}