Make Requests Processes and create process hierarchy. Associate OpenR…

…epository with context. (#17125)

This PR registers requests with the process manager and manages hierarchy within the processes.

Git repos are then associated with a context, (usually the request's context) - with sub commands using this context as their base context.

Signed-off-by: Andrew Thornton <art27@cantab.net>

cmd/hook.go

@@ -309,7 +309,7 @@ func runHookPostReceive(c *cli.Context) error {
 	defer cancel()
 
 	// First of all run update-server-info no matter what
-	if _, err := git.NewCommand("update-server-info").SetParentContext(ctx).Run(); err != nil {
+	if _, err := git.NewCommandContext(ctx, "update-server-info").Run(); err != nil {
 		return fmt.Errorf("Failed to call 'git update-server-info': %v", err)
 	}
 

modules/context/repo.go

@@ -534,7 +534,7 @@ func RepoAssignment(ctx *Context) (cancel context.CancelFunc) {
 		return
 	}
 
-	gitRepo, err := git.OpenRepository(models.RepoPath(userName, repoName))
+	gitRepo, err := git.OpenRepositoryCtx(ctx, models.RepoPath(userName, repoName))
 	if err != nil {
 		if strings.Contains(err.Error(), "repository does not exist") || strings.Contains(err.Error(), "no such file or directory") {
 			log.Error("Repository %-v has a broken repository on the file system: %s Error: %v", ctx.Repo.Repository, ctx.Repo.Repository.RepoPath(), err)
@@ -792,7 +792,7 @@ func RepoRefByType(refType RepoRefType, ignoreNotExistErr ...bool) func(*Context
 
 		if ctx.Repo.GitRepo == nil {
 			repoPath := models.RepoPath(ctx.Repo.Owner.Name, ctx.Repo.Repository.Name)
-			ctx.Repo.GitRepo, err = git.OpenRepository(repoPath)
+			ctx.Repo.GitRepo, err = git.OpenRepositoryCtx(ctx, repoPath)
 			if err != nil {
 				ctx.ServerError("RepoRef Invalid repo "+repoPath, err)
 				return

modules/git/batch_reader.go

@@ -28,17 +28,15 @@ type WriteCloserError interface {
 }
 
 // CatFileBatchCheck opens git cat-file --batch-check in the provided repo and returns a stdin pipe, a stdout reader and cancel function
-func CatFileBatchCheck(repoPath string) (WriteCloserError, *bufio.Reader, func()) {
+func CatFileBatchCheck(ctx context.Context, repoPath string) (WriteCloserError, *bufio.Reader, func()) {
 	batchStdinReader, batchStdinWriter := io.Pipe()
 	batchStdoutReader, batchStdoutWriter := io.Pipe()
-	ctx, ctxCancel := context.WithCancel(DefaultContext)
+	ctx, ctxCancel := context.WithCancel(ctx)
 	closed := make(chan struct{})
 	cancel := func() {
-		_ = batchStdinReader.Close()
-		_ = batchStdinWriter.Close()
-		_ = batchStdoutReader.Close()
-		_ = batchStdoutWriter.Close()
 		ctxCancel()
+		_ = batchStdoutReader.Close()
+		_ = batchStdinWriter.Close()
 		<-closed
 	}
 
@@ -67,19 +65,17 @@ func CatFileBatchCheck(repoPath string) (WriteCloserError, *bufio.Reader, func()
 }
 
 // CatFileBatch opens git cat-file --batch in the provided repo and returns a stdin pipe, a stdout reader and cancel function
-func CatFileBatch(repoPath string) (WriteCloserError, *bufio.Reader, func()) {
+func CatFileBatch(ctx context.Context, repoPath string) (WriteCloserError, *bufio.Reader, func()) {
 	// We often want to feed the commits in order into cat-file --batch, followed by their trees and sub trees as necessary.
 	// so let's create a batch stdin and stdout
 	batchStdinReader, batchStdinWriter := io.Pipe()
 	batchStdoutReader, batchStdoutWriter := nio.Pipe(buffer.New(32 * 1024))
-	ctx, ctxCancel := context.WithCancel(DefaultContext)
+	ctx, ctxCancel := context.WithCancel(ctx)
 	closed := make(chan struct{})
 	cancel := func() {
-		_ = batchStdinReader.Close()
+		ctxCancel()
 		_ = batchStdinWriter.Close()
 		_ = batchStdoutReader.Close()
-		_ = batchStdoutWriter.Close()
-		ctxCancel()
 		<-closed
 	}
 

modules/git/blame.go

@@ -24,12 +24,12 @@ type BlamePart struct {
 
 // BlameReader returns part of file blame one by one
 type BlameReader struct {
-	cmd     *exec.Cmd
-	pid     int64
-	output  io.ReadCloser
-	reader  *bufio.Reader
-	lastSha *string
-	cancel  context.CancelFunc
+	cmd      *exec.Cmd
+	output   io.ReadCloser
+	reader   *bufio.Reader
+	lastSha  *string
+	cancel   context.CancelFunc   // Cancels the context that this reader runs in
+	finished process.FinishedFunc // Tells the process manager we're finished and it can remove the associated process from the process table
 }
 
 var shaLineRegex = regexp.MustCompile("^([a-z0-9]{40})")
@@ -100,8 +100,8 @@ func (r *BlameReader) NextPart() (*BlamePart, error) {
 
 // Close BlameReader - don't run NextPart after invoking that
 func (r *BlameReader) Close() error {
-	defer process.GetManager().Remove(r.pid)
-	r.cancel()
+	defer r.finished() // Only remove the process from the process table when the underlying command is closed
+	r.cancel()         // However, first cancel our own context early
 
 	_ = r.output.Close()
 
@@ -114,7 +114,7 @@ func (r *BlameReader) Close() error {
 
 // CreateBlameReader creates reader for given repository, commit and file
 func CreateBlameReader(ctx context.Context, repoPath, commitID, file string) (*BlameReader, error) {
-	gitRepo, err := OpenRepository(repoPath)
+	gitRepo, err := OpenRepositoryCtx(ctx, repoPath)
 	if err != nil {
 		return nil, err
 	}
@@ -125,32 +125,31 @@ func CreateBlameReader(ctx context.Context, repoPath, commitID, file string) (*B
 
 func createBlameReader(ctx context.Context, dir string, command ...string) (*BlameReader, error) {
 	// Here we use the provided context - this should be tied to the request performing the blame so that it does not hang around.
-	ctx, cancel := context.WithCancel(ctx)
+	ctx, cancel, finished := process.GetManager().AddContext(ctx, fmt.Sprintf("GetBlame [repo_path: %s]", dir))
+
 	cmd := exec.CommandContext(ctx, command[0], command[1:]...)
 	cmd.Dir = dir
 	cmd.Stderr = os.Stderr
 
 	stdout, err := cmd.StdoutPipe()
 	if err != nil {
-		defer cancel()
+		defer finished()
 		return nil, fmt.Errorf("StdoutPipe: %v", err)
 	}
 
 	if err = cmd.Start(); err != nil {
-		defer cancel()
+		defer finished()
+		_ = stdout.Close()
 		return nil, fmt.Errorf("Start: %v", err)
 	}
 
-	pid := process.GetManager().Add(fmt.Sprintf("GetBlame [repo_path: %s]", dir), cancel)
-
 	reader := bufio.NewReader(stdout)
 
 	return &BlameReader{
-		cmd,
-		pid,
-		stdout,
-		reader,
-		nil,
-		cancel,
+		cmd:      cmd,
+		output:   stdout,
+		reader:   reader,
+		cancel:   cancel,
+		finished: finished,
 	}, nil
 }

modules/git/blob_nogogit.go

@@ -29,7 +29,7 @@ type Blob struct {
 // DataAsync gets a ReadCloser for the contents of a blob without reading it all.
 // Calling the Close function on the result will discard all unread output.
 func (b *Blob) DataAsync() (io.ReadCloser, error) {
-	wr, rd, cancel := b.repo.CatFileBatch()
+	wr, rd, cancel := b.repo.CatFileBatch(b.repo.Ctx)
 
 	_, err := wr.Write([]byte(b.ID.String() + "\n"))
 	if err != nil {
@@ -67,7 +67,7 @@ func (b *Blob) Size() int64 {
 		return b.size
 	}
 
-	wr, rd, cancel := b.repo.CatFileBatchCheck()
+	wr, rd, cancel := b.repo.CatFileBatchCheck(b.repo.Ctx)
 	defer cancel()
 	_, err := wr.Write([]byte(b.ID.String() + "\n"))
 	if err != nil {

modules/git/command.go

@@ -143,8 +143,13 @@ func (c *Command) RunWithContext(rc *RunContext) error {
 		log.Debug("%s: %v", rc.Dir, c)
 	}
 
-	ctx, cancel := context.WithTimeout(c.parentContext, rc.Timeout)
-	defer cancel()
+	desc := c.desc
+	if desc == "" {
+		desc = fmt.Sprintf("%s %s [repo_path: %s]", c.name, strings.Join(c.args, " "), rc.Dir)
+	}
+
+	ctx, cancel, finished := process.GetManager().AddContextTimeout(c.parentContext, rc.Timeout, desc)
+	defer finished()
 
 	cmd := exec.CommandContext(ctx, c.name, c.args...)
 	if rc.Env == nil {
@@ -172,13 +177,6 @@ func (c *Command) RunWithContext(rc *RunContext) error {
 		return err
 	}
 
-	desc := c.desc
-	if desc == "" {
-		desc = fmt.Sprintf("%s %s %s [repo_path: %s]", GitExecutable, c.name, strings.Join(c.args, " "), rc.Dir)
-	}
-	pid := process.GetManager().Add(desc, cancel)
-	defer process.GetManager().Remove(pid)
-
 	if rc.PipelineFunc != nil {
 		err := rc.PipelineFunc(ctx, cancel)
 		if err != nil {

modules/git/commit_info_nogogit.go

@@ -100,7 +100,7 @@ func (tes Entries) GetCommitsInfo(ctx context.Context, commit *Commit, treePath
 }
 
 func getLastCommitForPathsByCache(ctx context.Context, commitID, treePath string, paths []string, cache *LastCommitCache) (map[string]*Commit, []string, error) {
-	wr, rd, cancel := cache.repo.CatFileBatch()
+	wr, rd, cancel := cache.repo.CatFileBatch(ctx)
 	defer cancel()
 
 	var unHitEntryPaths []string
@@ -129,7 +129,7 @@ func GetLastCommitForPaths(ctx context.Context, cache *LastCommitCache, commit *
 		return nil, err
 	}
 
-	batchStdinWriter, batchReader, cancel := commit.repo.CatFileBatch()
+	batchStdinWriter, batchReader, cancel := commit.repo.CatFileBatch(ctx)
 	defer cancel()
 
 	commitsMap := map[string]*Commit{}

modules/git/diff.go

@@ -56,8 +56,8 @@ func GetRepoRawDiffForFile(repo *Repository, startCommit, endCommit string, diff
 		fileArgs = append(fileArgs, "--", file)
 	}
 	// FIXME: graceful: These commands should have a timeout
-	ctx, cancel := context.WithCancel(DefaultContext)
-	defer cancel()
+	ctx, _, finished := process.GetManager().AddContext(repo.Ctx, fmt.Sprintf("GetRawDiffForFile: [repo_path: %s]", repo.Path))
+	defer finished()
 
 	var cmd *exec.Cmd
 	switch diffType {
@@ -90,8 +90,6 @@ func GetRepoRawDiffForFile(repo *Repository, startCommit, endCommit string, diff
 	cmd.Dir = repo.Path
 	cmd.Stdout = writer
 	cmd.Stderr = stderr
-	pid := process.GetManager().Add(fmt.Sprintf("GetRawDiffForFile: [repo_path: %s]", repo.Path), cancel)
-	defer process.GetManager().Remove(pid)
 
 	if err = cmd.Run(); err != nil {
 		return fmt.Errorf("Run: %v - %s", err, stderr)

modules/git/pipeline/lfs_nogogit.go

@@ -63,7 +63,7 @@ func FindLFSFile(repo *git.Repository, hash git.SHA1) ([]*LFSResult, error) {
 
 	// Next feed the commits in order into cat-file --batch, followed by their trees and sub trees as necessary.
 	// so let's create a batch stdin and stdout
-	batchStdinWriter, batchReader, cancel := repo.CatFileBatch()
+	batchStdinWriter, batchReader, cancel := repo.CatFileBatch(repo.Ctx)
 	defer cancel()
 
 	// We'll use a scanner for the revList because it's simpler than a bufio.Reader

modules/git/remote.go

@@ -4,19 +4,22 @@
 
 package git
 
-import "net/url"
+import (
+	"context"
+	"net/url"
+)
 
 // GetRemoteAddress returns the url of a specific remote of the repository.
-func GetRemoteAddress(repoPath, remoteName string) (*url.URL, error) {
+func GetRemoteAddress(ctx context.Context, repoPath, remoteName string) (*url.URL, error) {
 	err := LoadGitVersion()
 	if err != nil {
 		return nil, err
 	}
 	var cmd *Command
 	if CheckGitVersionAtLeast("2.7") == nil {
-		cmd = NewCommand("remote", "get-url", remoteName)
+		cmd = NewCommandContext(ctx, "remote", "get-url", remoteName)
 	} else {
-		cmd = NewCommand("config", "--get", "remote."+remoteName+".url")
+		cmd = NewCommandContext(ctx, "config", "--get", "remote."+remoteName+".url")
 	}
 
 	result, err := cmd.RunInDir(repoPath)

modules/git/repo.go

@@ -211,8 +211,8 @@ type PushOptions struct {
 }
 
 // Push pushs local commits to given remote branch.
-func Push(repoPath string, opts PushOptions) error {
-	cmd := NewCommand("push")
+func Push(ctx context.Context, repoPath string, opts PushOptions) error {
+	cmd := NewCommandContext(ctx, "push")
 	if opts.Force {
 		cmd.AddArguments("-f")
 	}

modules/git/repo_attribute.go

@@ -74,7 +74,7 @@ func (repo *Repository) CheckAttribute(opts CheckAttributeOpts) (map[string]map[
 		}
 	}
 
-	cmd := NewCommand(cmdArgs...)
+	cmd := NewCommandContext(repo.Ctx, cmdArgs...)
 
 	if err := cmd.RunInDirTimeoutEnvPipeline(env, -1, repo.Path, stdOut, stdErr); err != nil {
 		return nil, fmt.Errorf("failed to run check-attr: %v\n%s\n%s", err, stdOut.String(), stdErr.String())

modules/git/repo_base_gogit.go

@@ -9,6 +9,7 @@
 package git
 
 import (
+	"context"
 	"errors"
 	"path/filepath"
 
@@ -30,10 +31,17 @@ type Repository struct {
 	gogitRepo    *gogit.Repository
 	gogitStorage *filesystem.Storage
 	gpgSettings  *GPGSettings
+
+	Ctx context.Context
 }
 
 // OpenRepository opens the repository at the given path.
 func OpenRepository(repoPath string) (*Repository, error) {
+	return OpenRepositoryCtx(DefaultContext, repoPath)
+}
+
+// OpenRepositoryCtx opens the repository at the given path within the context.Context
+func OpenRepositoryCtx(ctx context.Context, repoPath string) (*Repository, error) {
 	repoPath, err := filepath.Abs(repoPath)
 	if err != nil {
 		return nil, err
@@ -60,6 +68,7 @@ func OpenRepository(repoPath string) (*Repository, error) {
 		gogitRepo:    gogitRepo,
 		gogitStorage: storage,
 		tagCache:     newObjectCache(),
+		Ctx:          ctx,
 	}, nil
 }
 

modules/git/repo_base_nogogit.go

@@ -32,10 +32,17 @@ type Repository struct {
 	checkCancel context.CancelFunc
 	checkReader *bufio.Reader
 	checkWriter WriteCloserError
+
+	Ctx context.Context
 }
 
 // OpenRepository opens the repository at the given path.
 func OpenRepository(repoPath string) (*Repository, error) {
+	return OpenRepositoryCtx(DefaultContext, repoPath)
+}
+
+// OpenRepositoryCtx opens the repository at the given path with the provided context.
+func OpenRepositoryCtx(ctx context.Context, repoPath string) (*Repository, error) {
 	repoPath, err := filepath.Abs(repoPath)
 	if err != nil {
 		return nil, err
@@ -46,28 +53,29 @@ func OpenRepository(repoPath string) (*Repository, error) {
 	repo := &Repository{
 		Path:     repoPath,
 		tagCache: newObjectCache(),
+		Ctx:      ctx,
 	}
 
-	repo.batchWriter, repo.batchReader, repo.batchCancel = CatFileBatch(repoPath)
-	repo.checkWriter, repo.checkReader, repo.checkCancel = CatFileBatchCheck(repo.Path)
+	repo.batchWriter, repo.batchReader, repo.batchCancel = CatFileBatch(ctx, repoPath)
+	repo.checkWriter, repo.checkReader, repo.checkCancel = CatFileBatchCheck(ctx, repo.Path)
 
 	return repo, nil
 }
 
 // CatFileBatch obtains a CatFileBatch for this repository
-func (repo *Repository) CatFileBatch() (WriteCloserError, *bufio.Reader, func()) {
+func (repo *Repository) CatFileBatch(ctx context.Context) (WriteCloserError, *bufio.Reader, func()) {
 	if repo.batchCancel == nil || repo.batchReader.Buffered() > 0 {
 		log.Debug("Opening temporary cat file batch for: %s", repo.Path)
-		return CatFileBatch(repo.Path)
+		return CatFileBatch(ctx, repo.Path)
 	}
 	return repo.batchWriter, repo.batchReader, func() {}
 }
 
 // CatFileBatchCheck obtains a CatFileBatchCheck for this repository
-func (repo *Repository) CatFileBatchCheck() (WriteCloserError, *bufio.Reader, func()) {
+func (repo *Repository) CatFileBatchCheck(ctx context.Context) (WriteCloserError, *bufio.Reader, func()) {
 	if repo.checkCancel == nil || repo.checkReader.Buffered() > 0 {
 		log.Debug("Opening temporary cat file batch-check: %s", repo.Path)
-		return CatFileBatchCheck(repo.Path)
+		return CatFileBatchCheck(ctx, repo.Path)
 	}
 	return repo.checkWriter, repo.checkReader, func() {}
 }

modules/git/repo_blame.go

@@ -8,12 +8,12 @@ import "fmt"
 
 // FileBlame return the Blame object of file
 func (repo *Repository) FileBlame(revision, path, file string) ([]byte, error) {
-	return NewCommand("blame", "--root", "--", file).RunInDirBytes(path)
+	return NewCommandContext(repo.Ctx, "blame", "--root", "--", file).RunInDirBytes(path)
 }
 
 // LineBlame returns the latest commit at the given line
 func (repo *Repository) LineBlame(revision, path, file string, line uint) (*Commit, error) {
-	res, err := NewCommand("blame", fmt.Sprintf("-L %d,%d", line, line), "-p", revision, "--", file).RunInDir(path)
+	res, err := NewCommandContext(repo.Ctx, "blame", fmt.Sprintf("-L %d,%d", line, line), "-p", revision, "--", file).RunInDir(path)
 	if err != nil {
 		return nil, err
 	}