Graceful: Cancel Process on monitor pages & HammerTime (#9213)

* Graceful: Create callbacks to with contexts

* Graceful: Say when Gitea is completely finished

* Graceful: Git and Process within HammerTime

Force all git commands to terminate at HammerTime
Force all process commands to terminate at HammerTime
Move almost all git processes to run as git Commands

* Graceful: Always Hammer after Shutdown

* ProcessManager: Add cancel functionality

* Fix tests

* Make sure that process.Manager.Kill() cancels

* Make threadsafe access to Processes and remove own unused Kill

* Remove cmd from the process manager as it is no longer used

* the default context is the correct context

* get rid of double till
This commit is contained in:
zeripath 2019-11-30 08:40:22 -06:00 committed by Lauris BH
parent 8f8c250ddb
commit 60c5339042
21 changed files with 536 additions and 198 deletions

View file

@ -6,6 +6,7 @@ package git
import (
"bufio"
"context"
"fmt"
"io"
"os"
@ -28,6 +29,7 @@ type BlameReader struct {
output io.ReadCloser
scanner *bufio.Scanner
lastSha *string
cancel context.CancelFunc
}
var shaLineRegex = regexp.MustCompile("^([a-z0-9]{40})")
@ -76,7 +78,8 @@ func (r *BlameReader) NextPart() (*BlamePart, error) {
// Close BlameReader - don't run NextPart after invoking that
func (r *BlameReader) Close() error {
process.GetManager().Remove(r.pid)
defer process.GetManager().Remove(r.pid)
defer r.cancel()
if err := r.cmd.Wait(); err != nil {
return fmt.Errorf("Wait: %v", err)
@ -97,20 +100,24 @@ func CreateBlameReader(repoPath, commitID, file string) (*BlameReader, error) {
}
func createBlameReader(dir string, command ...string) (*BlameReader, error) {
cmd := exec.Command(command[0], command[1:]...)
// FIXME: graceful: This should have a timeout
ctx, cancel := context.WithCancel(DefaultContext)
cmd := exec.CommandContext(ctx, command[0], command[1:]...)
cmd.Dir = dir
cmd.Stderr = os.Stderr
stdout, err := cmd.StdoutPipe()
if err != nil {
defer cancel()
return nil, fmt.Errorf("StdoutPipe: %v", err)
}
if err = cmd.Start(); err != nil {
defer cancel()
return nil, fmt.Errorf("Start: %v", err)
}
pid := process.GetManager().Add(fmt.Sprintf("GetBlame [repo_path: %s]", dir), cmd)
pid := process.GetManager().Add(fmt.Sprintf("GetBlame [repo_path: %s]", dir), cancel)
scanner := bufio.NewScanner(stdout)
@ -120,5 +127,6 @@ func createBlameReader(dir string, command ...string) (*BlameReader, error) {
stdout,
scanner,
nil,
cancel,
}, nil
}

View file

@ -30,8 +30,10 @@ const DefaultLocale = "C"
// Command represents a command with its subcommands or arguments.
type Command struct {
name string
args []string
name string
args []string
parentContext context.Context
desc string
}
func (c *Command) String() string {
@ -47,19 +49,34 @@ func NewCommand(args ...string) *Command {
cargs := make([]string, len(GlobalCommandArgs))
copy(cargs, GlobalCommandArgs)
return &Command{
name: GitExecutable,
args: append(cargs, args...),
name: GitExecutable,
args: append(cargs, args...),
parentContext: DefaultContext,
}
}
// NewCommandNoGlobals creates and returns a new Git Command based on given command and arguments only with the specify args and don't care global command args
func NewCommandNoGlobals(args ...string) *Command {
return &Command{
name: GitExecutable,
args: args,
name: GitExecutable,
args: args,
parentContext: DefaultContext,
}
}
// SetParentContext sets the parent context for this command
func (c *Command) SetParentContext(ctx context.Context) *Command {
c.parentContext = ctx
return c
}
// SetDescription sets the description for this command which be returned on
// c.String()
func (c *Command) SetDescription(desc string) *Command {
c.desc = desc
return c
}
// AddArguments adds new argument(s) to the command.
func (c *Command) AddArguments(args ...string) *Command {
c.args = append(c.args, args...)
@ -92,7 +109,7 @@ func (c *Command) RunInDirTimeoutEnvFullPipelineFunc(env []string, timeout time.
log("%s: %v", dir, c)
}
ctx, cancel := context.WithTimeout(context.Background(), timeout)
ctx, cancel := context.WithTimeout(c.parentContext, timeout)
defer cancel()
cmd := exec.CommandContext(ctx, c.name, c.args...)
@ -110,7 +127,11 @@ func (c *Command) RunInDirTimeoutEnvFullPipelineFunc(env []string, timeout time.
return err
}
pid := process.GetManager().Add(fmt.Sprintf("%s %s %s [repo_path: %s]", GitExecutable, c.name, strings.Join(c.args, " "), dir), cmd)
desc := c.desc
if desc == "" {
desc = fmt.Sprintf("%s %s %s [repo_path: %s]", GitExecutable, c.name, strings.Join(c.args, " "), dir)
}
pid := process.GetManager().Add(desc, cancel)
defer process.GetManager().Remove(pid)
if fn != nil {

View file

@ -6,6 +6,7 @@
package git
import (
"context"
"fmt"
"os/exec"
"runtime"
@ -35,6 +36,9 @@ var (
// Could be updated to an absolute path while initialization
GitExecutable = "git"
// DefaultContext is the default context to run git commands in
DefaultContext = context.Background()
gitVersion string
)

View file

@ -0,0 +1,90 @@
// Copyright 2019 The Gitea Authors. All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.
package graceful
import (
"context"
"fmt"
"time"
)
// Errors for context.Err()
var (
ErrShutdown = fmt.Errorf("Graceful Manager called Shutdown")
ErrHammer = fmt.Errorf("Graceful Manager called Hammer")
ErrTerminate = fmt.Errorf("Graceful Manager called Terminate")
)
// ChannelContext is a context that wraps a channel and error as a context
type ChannelContext struct {
done <-chan struct{}
err error
}
// NewChannelContext creates a ChannelContext from a channel and error
func NewChannelContext(done <-chan struct{}, err error) *ChannelContext {
return &ChannelContext{
done: done,
err: err,
}
}
// Deadline returns the time when work done on behalf of this context
// should be canceled. There is no Deadline for a ChannelContext
func (ctx *ChannelContext) Deadline() (deadline time.Time, ok bool) {
return
}
// Done returns the channel provided at the creation of this context.
// When closed, work done on behalf of this context should be canceled.
func (ctx *ChannelContext) Done() <-chan struct{} {
return ctx.done
}
// Err returns nil, if Done is not closed. If Done is closed,
// Err returns the error provided at the creation of this context
func (ctx *ChannelContext) Err() error {
select {
case <-ctx.done:
return ctx.err
default:
return nil
}
}
// Value returns nil for all calls as no values are or can be associated with this context
func (ctx *ChannelContext) Value(key interface{}) interface{} {
return nil
}
// ShutdownContext returns a context.Context that is Done at shutdown
// Callers using this context should ensure that they are registered as a running server
// in order that they are waited for.
func (g *gracefulManager) ShutdownContext() context.Context {
return &ChannelContext{
done: g.IsShutdown(),
err: ErrShutdown,
}
}
// HammerContext returns a context.Context that is Done at hammer
// Callers using this context should ensure that they are registered as a running server
// in order that they are waited for.
func (g *gracefulManager) HammerContext() context.Context {
return &ChannelContext{
done: g.IsHammer(),
err: ErrHammer,
}
}
// TerminateContext returns a context.Context that is Done at terminate
// Callers using this context should ensure that they are registered as a terminating server
// in order that they are waited for.
func (g *gracefulManager) TerminateContext() context.Context {
return &ChannelContext{
done: g.IsTerminate(),
err: ErrTerminate,
}
}

View file

@ -5,9 +5,12 @@
package graceful
import (
"context"
"time"
"code.gitea.io/gitea/modules/git"
"code.gitea.io/gitea/modules/log"
"code.gitea.io/gitea/modules/process"
"code.gitea.io/gitea/modules/setting"
)
@ -34,9 +37,110 @@ const numberOfServersToCreate = 3
var Manager *gracefulManager
func init() {
Manager = newGracefulManager()
Manager = newGracefulManager(context.Background())
// Set the git default context to the HammerContext
git.DefaultContext = Manager.HammerContext()
// Set the process default context to the HammerContext
process.DefaultContext = Manager.HammerContext()
}
// CallbackWithContext is combined runnable and context to watch to see if the caller has finished
type CallbackWithContext func(ctx context.Context, callback func())
// RunnableWithShutdownFns is a runnable with functions to run at shutdown and terminate
// After the callback to atShutdown is called and is complete, the main function must return.
// Similarly the callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atShutdown and atTerminate callbacks will create go-routines that will wait till their respective signals
// - users must therefore be careful to only call these as necessary.
// If run is not expected to run indefinitely RunWithShutdownChan is likely to be more appropriate.
type RunnableWithShutdownFns func(atShutdown, atTerminate func(context.Context, func()))
// RunWithShutdownFns takes a function that has both atShutdown and atTerminate callbacks
// After the callback to atShutdown is called and is complete, the main function must return.
// Similarly the callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atShutdown and atTerminate callbacks will create go-routines that will wait till their respective signals
// - users must therefore be careful to only call these as necessary.
// If run is not expected to run indefinitely RunWithShutdownChan is likely to be more appropriate.
func (g *gracefulManager) RunWithShutdownFns(run RunnableWithShutdownFns) {
g.runningServerWaitGroup.Add(1)
defer g.runningServerWaitGroup.Done()
run(func(ctx context.Context, atShutdown func()) {
go func() {
select {
case <-g.IsShutdown():
atShutdown()
case <-ctx.Done():
return
}
}()
}, func(ctx context.Context, atTerminate func()) {
g.RunAtTerminate(ctx, atTerminate)
})
}
// RunnableWithShutdownChan is a runnable with functions to run at shutdown and terminate.
// After the atShutdown channel is closed, the main function must return once shutdown is complete.
// (Optionally IsHammer may be waited for instead however, this should be avoided if possible.)
// The callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atTerminate function will create a go-routine that will wait till terminate - users must therefore be careful to only call this as necessary.
type RunnableWithShutdownChan func(atShutdown <-chan struct{}, atTerminate CallbackWithContext)
// RunWithShutdownChan takes a function that has channel to watch for shutdown and atTerminate callbacks
// After the atShutdown channel is closed, the main function must return once shutdown is complete.
// (Optionally IsHammer may be waited for instead however, this should be avoided if possible.)
// The callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atTerminate function will create a go-routine that will wait till terminate - users must therefore be careful to only call this as necessary.
func (g *gracefulManager) RunWithShutdownChan(run RunnableWithShutdownChan) {
g.runningServerWaitGroup.Add(1)
defer g.runningServerWaitGroup.Done()
run(g.IsShutdown(), func(ctx context.Context, atTerminate func()) {
g.RunAtTerminate(ctx, atTerminate)
})
}
// RunWithShutdownContext takes a function that has a context to watch for shutdown.
// After the provided context is Done(), the main function must return once shutdown is complete.
// (Optionally the HammerContext may be obtained and waited for however, this should be avoided if possible.)
func (g *gracefulManager) RunWithShutdownContext(run func(context.Context)) {
g.runningServerWaitGroup.Add(1)
defer g.runningServerWaitGroup.Done()
run(g.ShutdownContext())
}
// RunAtTerminate adds to the terminate wait group and creates a go-routine to run the provided function at termination
func (g *gracefulManager) RunAtTerminate(ctx context.Context, terminate func()) {
g.terminateWaitGroup.Add(1)
go func() {
select {
case <-g.IsTerminate():
terminate()
case <-ctx.Done():
}
g.terminateWaitGroup.Done()
}()
}
// RunAtShutdown creates a go-routine to run the provided function at shutdown
func (g *gracefulManager) RunAtShutdown(ctx context.Context, shutdown func()) {
go func() {
select {
case <-g.IsShutdown():
shutdown()
case <-ctx.Done():
}
}()
}
// RunAtHammer creates a go-routine to run the provided function at shutdown
func (g *gracefulManager) RunAtHammer(ctx context.Context, hammer func()) {
go func() {
select {
case <-g.IsHammer():
hammer()
case <-ctx.Done():
}
}()
}
func (g *gracefulManager) doShutdown() {
if !g.setStateTransition(stateRunning, stateShuttingDown) {
return
@ -50,6 +154,8 @@ func (g *gracefulManager) doShutdown() {
}
go func() {
g.WaitForServers()
// Mop up any remaining unclosed events.
g.doHammerTime(0)
<-time.After(1 * time.Second)
g.doTerminate()
}()

View file

@ -7,6 +7,7 @@
package graceful
import (
"context"
"errors"
"os"
"os/signal"
@ -31,19 +32,19 @@ type gracefulManager struct {
terminateWaitGroup sync.WaitGroup
}
func newGracefulManager() *gracefulManager {
func newGracefulManager(ctx context.Context) *gracefulManager {
manager := &gracefulManager{
isChild: len(os.Getenv(listenFDs)) > 0 && os.Getppid() > 1,
lock: &sync.RWMutex{},
}
manager.createServerWaitGroup.Add(numberOfServersToCreate)
manager.Run()
manager.Run(ctx)
return manager
}
func (g *gracefulManager) Run() {
func (g *gracefulManager) Run(ctx context.Context) {
g.setState(stateRunning)
go g.handleSignals()
go g.handleSignals(ctx)
c := make(chan struct{})
go func() {
defer close(c)
@ -69,9 +70,7 @@ func (g *gracefulManager) Run() {
}
}
func (g *gracefulManager) handleSignals() {
var sig os.Signal
func (g *gracefulManager) handleSignals(ctx context.Context) {
signalChannel := make(chan os.Signal, 1)
signal.Notify(
@ -86,35 +85,40 @@ func (g *gracefulManager) handleSignals() {
pid := syscall.Getpid()
for {
sig = <-signalChannel
switch sig {
case syscall.SIGHUP:
if setting.GracefulRestartable {
log.Info("PID: %d. Received SIGHUP. Forking...", pid)
err := g.doFork()
if err != nil && err.Error() != "another process already forked. Ignoring this one" {
log.Error("Error whilst forking from PID: %d : %v", pid, err)
}
} else {
log.Info("PID: %d. Received SIGHUP. Not set restartable. Shutting down...", pid)
select {
case sig := <-signalChannel:
switch sig {
case syscall.SIGHUP:
if setting.GracefulRestartable {
log.Info("PID: %d. Received SIGHUP. Forking...", pid)
err := g.doFork()
if err != nil && err.Error() != "another process already forked. Ignoring this one" {
log.Error("Error whilst forking from PID: %d : %v", pid, err)
}
} else {
log.Info("PID: %d. Received SIGHUP. Not set restartable. Shutting down...", pid)
g.doShutdown()
}
case syscall.SIGUSR1:
log.Info("PID %d. Received SIGUSR1.", pid)
case syscall.SIGUSR2:
log.Warn("PID %d. Received SIGUSR2. Hammering...", pid)
g.doHammerTime(0 * time.Second)
case syscall.SIGINT:
log.Warn("PID %d. Received SIGINT. Shutting down...", pid)
g.doShutdown()
case syscall.SIGTERM:
log.Warn("PID %d. Received SIGTERM. Shutting down...", pid)
g.doShutdown()
case syscall.SIGTSTP:
log.Info("PID %d. Received SIGTSTP.", pid)
default:
log.Info("PID %d. Received %v.", pid, sig)
}
case syscall.SIGUSR1:
log.Info("PID %d. Received SIGUSR1.", pid)
case syscall.SIGUSR2:
log.Warn("PID %d. Received SIGUSR2. Hammering...", pid)
g.doHammerTime(0 * time.Second)
case syscall.SIGINT:
log.Warn("PID %d. Received SIGINT. Shutting down...", pid)
case <-ctx.Done():
log.Warn("PID: %d. Background context for manager closed - %v - Shutting down...", pid, ctx.Err())
g.doShutdown()
case syscall.SIGTERM:
log.Warn("PID %d. Received SIGTERM. Shutting down...", pid)
g.doShutdown()
case syscall.SIGTSTP:
log.Info("PID %d. Received SIGTSTP.", pid)
default:
log.Info("PID %d. Received %v.", pid, sig)
}
}
}

View file

@ -8,6 +8,7 @@
package graceful
import (
"context"
"os"
"strconv"
"sync"
@ -29,6 +30,7 @@ const (
)
type gracefulManager struct {
ctx context.Context
isChild bool
lock *sync.RWMutex
state state
@ -40,10 +42,11 @@ type gracefulManager struct {
terminateWaitGroup sync.WaitGroup
}
func newGracefulManager() *gracefulManager {
func newGracefulManager(ctx context.Context) *gracefulManager {
manager := &gracefulManager{
isChild: false,
lock: &sync.RWMutex{},
ctx: ctx,
}
manager.createServerWaitGroup.Add(numberOfServersToCreate)
manager.Run()
@ -89,23 +92,29 @@ func (g *gracefulManager) Execute(args []string, changes <-chan svc.ChangeReques
waitTime := 30 * time.Second
loop:
for change := range changes {
switch change.Cmd {
case svc.Interrogate:
status <- change.CurrentStatus
case svc.Stop, svc.Shutdown:
for {
select {
case <-g.ctx.Done():
g.doShutdown()
waitTime += setting.GracefulHammerTime
break loop
case hammerCode:
g.doShutdown()
g.doHammerTime(0 * time.Second)
break loop
default:
log.Debug("Unexpected control request: %v", change.Cmd)
case change := <-changes:
switch change.Cmd {
case svc.Interrogate:
status <- change.CurrentStatus
case svc.Stop, svc.Shutdown:
g.doShutdown()
waitTime += setting.GracefulHammerTime
break loop
case hammerCode:
g.doShutdown()
g.doHammerTime(0 * time.Second)
break loop
default:
log.Debug("Unexpected control request: %v", change.Cmd)
}
}
}
status <- svc.Status{
State: svc.StopPending,
WaitHint: uint32(waitTime / time.Millisecond),

View file

@ -12,6 +12,7 @@ import (
"fmt"
"io"
"os/exec"
"sort"
"sync"
"time"
)
@ -24,14 +25,17 @@ var (
// ErrExecTimeout represent a timeout error
ErrExecTimeout = errors.New("Process execution timeout")
manager *Manager
// DefaultContext is the default context to run processing commands in
DefaultContext = context.Background()
)
// Process represents a working process inherit from Gogs.
// Process represents a working process inheriting from Gitea.
type Process struct {
PID int64 // Process ID, not system one.
Description string
Start time.Time
Cmd *exec.Cmd
Cancel context.CancelFunc
}
// Manager knows about all processes and counts PIDs.
@ -39,28 +43,28 @@ type Manager struct {
mutex sync.Mutex
counter int64
Processes map[int64]*Process
processes map[int64]*Process
}
// GetManager returns a Manager and initializes one as singleton if there's none yet
func GetManager() *Manager {
if manager == nil {
manager = &Manager{
Processes: make(map[int64]*Process),
processes: make(map[int64]*Process),
}
}
return manager
}
// Add a process to the ProcessManager and returns its PID.
func (pm *Manager) Add(description string, cmd *exec.Cmd) int64 {
func (pm *Manager) Add(description string, cancel context.CancelFunc) int64 {
pm.mutex.Lock()
pid := pm.counter + 1
pm.Processes[pid] = &Process{
pm.processes[pid] = &Process{
PID: pid,
Description: description,
Start: time.Now(),
Cmd: cmd,
Cancel: cancel,
}
pm.counter = pid
pm.mutex.Unlock()
@ -71,10 +75,32 @@ func (pm *Manager) Add(description string, cmd *exec.Cmd) int64 {
// Remove a process from the ProcessManager.
func (pm *Manager) Remove(pid int64) {
pm.mutex.Lock()
delete(pm.Processes, pid)
delete(pm.processes, pid)
pm.mutex.Unlock()
}
// Cancel a process in the ProcessManager.
func (pm *Manager) Cancel(pid int64) {
pm.mutex.Lock()
process, ok := pm.processes[pid]
pm.mutex.Unlock()
if ok {
process.Cancel()
}
}
// Processes gets the processes in a thread safe manner
func (pm *Manager) Processes() []*Process {
pm.mutex.Lock()
processes := make([]*Process, 0, len(pm.processes))
for _, process := range pm.processes {
processes = append(processes, process)
}
pm.mutex.Unlock()
sort.Sort(processList(processes))
return processes
}
// Exec a command and use the default timeout.
func (pm *Manager) Exec(desc, cmdName string, args ...string) (string, string, error) {
return pm.ExecDir(-1, "", desc, cmdName, args...)
@ -110,7 +136,7 @@ func (pm *Manager) ExecDirEnvStdIn(timeout time.Duration, dir, desc string, env
stdOut := new(bytes.Buffer)
stdErr := new(bytes.Buffer)
ctx, cancel := context.WithTimeout(context.Background(), timeout)
ctx, cancel := context.WithTimeout(DefaultContext, timeout)
defer cancel()
cmd := exec.CommandContext(ctx, cmdName, args...)
@ -126,7 +152,7 @@ func (pm *Manager) ExecDirEnvStdIn(timeout time.Duration, dir, desc string, env
return "", "", err
}
pid := pm.Add(desc, cmd)
pid := pm.Add(desc, cancel)
err := cmd.Wait()
pm.Remove(pid)
@ -137,21 +163,16 @@ func (pm *Manager) ExecDirEnvStdIn(timeout time.Duration, dir, desc string, env
return stdOut.String(), stdErr.String(), err
}
// Kill and remove a process from list.
func (pm *Manager) Kill(pid int64) error {
if proc, exists := pm.Processes[pid]; exists {
pm.mutex.Lock()
if proc.Cmd != nil &&
proc.Cmd.Process != nil &&
proc.Cmd.ProcessState != nil &&
!proc.Cmd.ProcessState.Exited() {
if err := proc.Cmd.Process.Kill(); err != nil {
return fmt.Errorf("failed to kill process(%d/%s): %v", pid, proc.Description, err)
}
}
delete(pm.Processes, pid)
pm.mutex.Unlock()
}
type processList []*Process
return nil
func (l processList) Len() int {
return len(l)
}
func (l processList) Less(i, j int) bool {
return l[i].PID < l[j].PID
}
func (l processList) Swap(i, j int) {
l[i], l[j] = l[j], l[i]
}

View file

@ -1,7 +1,7 @@
package process
import (
"os/exec"
"context"
"testing"
"time"
@ -9,27 +9,42 @@ import (
)
func TestManager_Add(t *testing.T) {
pm := Manager{Processes: make(map[int64]*Process)}
pm := Manager{processes: make(map[int64]*Process)}
pid := pm.Add("foo", exec.Command("foo"))
pid := pm.Add("foo", nil)
assert.Equal(t, int64(1), pid, "expected to get pid 1 got %d", pid)
pid = pm.Add("bar", exec.Command("bar"))
pid = pm.Add("bar", nil)
assert.Equal(t, int64(2), pid, "expected to get pid 2 got %d", pid)
}
func TestManager_Remove(t *testing.T) {
pm := Manager{Processes: make(map[int64]*Process)}
func TestManager_Cancel(t *testing.T) {
pm := Manager{processes: make(map[int64]*Process)}
pid1 := pm.Add("foo", exec.Command("foo"))
ctx, cancel := context.WithCancel(context.Background())
pid := pm.Add("foo", cancel)
pm.Cancel(pid)
select {
case <-ctx.Done():
default:
assert.Fail(t, "Cancel should cancel the provided context")
}
}
func TestManager_Remove(t *testing.T) {
pm := Manager{processes: make(map[int64]*Process)}
pid1 := pm.Add("foo", nil)
assert.Equal(t, int64(1), pid1, "expected to get pid 1 got %d", pid1)
pid2 := pm.Add("bar", exec.Command("bar"))
pid2 := pm.Add("bar", nil)
assert.Equal(t, int64(2), pid2, "expected to get pid 2 got %d", pid2)
pm.Remove(pid2)
_, exists := pm.Processes[pid2]
_, exists := pm.processes[pid2]
assert.False(t, exists, "PID %d is in the list but shouldn't", pid2)
}