upaas/internal/docker/client.go

// Package docker provides Docker client functionality.
package docker

import (
	"bufio"
	"context"
	"errors"
	"fmt"
	"io"
	"log/slog"
	"os"
	"path/filepath"
	"regexp"
	"strconv"
	"strings"

	dockertypes "github.com/docker/docker/api/types"
	"github.com/docker/docker/api/types/container"
	"github.com/docker/docker/api/types/filters"
	"github.com/docker/docker/api/types/image"
	"github.com/docker/docker/api/types/mount"
	"github.com/docker/docker/api/types/network"
	"github.com/docker/docker/client"
	"github.com/docker/docker/pkg/archive"
	"github.com/docker/go-connections/nat"
	"go.uber.org/fx"

	"git.eeqj.de/sneak/upaas/internal/config"
	"git.eeqj.de/sneak/upaas/internal/logger"
)

// sshKeyPermissions is the file permission for SSH private keys.
const sshKeyPermissions = 0o600

// workDirPermissions is the file permission for the work directory.
const workDirPermissions = 0o750

// stopTimeoutSeconds is the timeout for stopping containers.
const stopTimeoutSeconds = 10

// gitImage is the Docker image used for git operations.
// alpine/git v2.47.2 - pulled 2025-12-30
const gitImage = "alpine/git@sha256:d86f367afb53d022acc4377741e7334bc20add161bb10234272b91b459b4b7d8"

// ErrNotConnected is returned when Docker client is not connected.
var ErrNotConnected = errors.New("docker client not connected")

// ErrGitCloneFailed is returned when git clone fails.
var ErrGitCloneFailed = errors.New("git clone failed")

// ErrInvalidBranch is returned when a branch name contains invalid characters.
var ErrInvalidBranch = errors.New("invalid branch name")

// ErrInvalidCommitSHA is returned when a commit SHA is not a valid hex string.
var ErrInvalidCommitSHA = errors.New("invalid commit SHA")

// validBranchRe matches safe git branch names.
var validBranchRe = regexp.MustCompile(`^[a-zA-Z0-9._/\-]+$`)

// validCommitSHARe matches a full-length hex commit SHA.
var validCommitSHARe = regexp.MustCompile(`^[0-9a-f]{40}$`)

// Params contains dependencies for Client.
type Params struct {
	fx.In

	Logger *logger.Logger
	Config *config.Config
}

// Client wraps the Docker client.
type Client struct {
	docker *client.Client
	log    *slog.Logger
	params *Params
}

// New creates a new Docker Client.
func New(lifecycle fx.Lifecycle, params Params) (*Client, error) {
	dockerClient := &Client{
		log:    params.Logger.Get(),
		params: &params,
	}

	// For testing, if lifecycle is nil, skip connection (tests mock Docker)
	if lifecycle == nil {
		return dockerClient, nil
	}

	lifecycle.Append(fx.Hook{
		OnStart: func(ctx context.Context) error {
			return dockerClient.connect(ctx)
		},
		OnStop: func(_ context.Context) error {
			return dockerClient.close()
		},
	})

	return dockerClient, nil
}

// IsConnected returns true if the Docker client is connected.
func (c *Client) IsConnected() bool {
	return c.docker != nil
}

// BuildImageOptions contains options for building an image.
type BuildImageOptions struct {
	ContextDir     string
	DockerfilePath string
	Tags           []string
	LogWriter      io.Writer // Optional writer for build output
}

// BuildImage builds a Docker image from a context directory.
func (c *Client) BuildImage(
	ctx context.Context,
	opts BuildImageOptions,
) (ImageID, error) {
	if c.docker == nil {
		return "", ErrNotConnected
	}

	c.log.Info(
		"building docker image",
		"context", opts.ContextDir,
		"dockerfile", opts.DockerfilePath,
	)

	imageID, err := c.performBuild(ctx, opts)
	if err != nil {
		return "", err
	}

	return imageID, nil
}

// CreateContainerOptions contains options for creating a container.
type CreateContainerOptions struct {
	Name    string
	Image   string
	Env     map[string]string
	Labels  map[string]string
	Volumes []VolumeMount
	Ports   []PortMapping
	Network string
}

// VolumeMount represents a volume mount.
type VolumeMount struct {
	HostPath      string
	ContainerPath string
	ReadOnly      bool
}

// PortMapping represents a port mapping.
type PortMapping struct {
	HostPort      int
	ContainerPort int
	Protocol      string // "tcp" or "udp"
}

// buildPortConfig converts port mappings to Docker port configuration.
func buildPortConfig(ports []PortMapping) (nat.PortSet, nat.PortMap) {
	exposedPorts := make(nat.PortSet)
	portBindings := make(nat.PortMap)

	for _, p := range ports {
		proto := p.Protocol
		if proto == "" {
			proto = "tcp"
		}

		containerPort := nat.Port(fmt.Sprintf("%d/%s", p.ContainerPort, proto))
		exposedPorts[containerPort] = struct{}{}
		portBindings[containerPort] = []nat.PortBinding{
			{
				HostIP:   "0.0.0.0",
				HostPort: strconv.Itoa(p.HostPort),
			},
		}
	}

	return exposedPorts, portBindings
}

// CreateContainer creates a new container.
func (c *Client) CreateContainer(
	ctx context.Context,
	opts CreateContainerOptions,
) (ContainerID, error) {
	if c.docker == nil {
		return "", ErrNotConnected
	}

	c.log.Info("creating container", "name", opts.Name, "image", opts.Image)

	// Convert env map to slice
	envSlice := make([]string, 0, len(opts.Env))

	for key, val := range opts.Env {
		envSlice = append(envSlice, key+"="+val)
	}

	// Convert volumes to mounts
	mounts := make([]mount.Mount, 0, len(opts.Volumes))

	for _, vol := range opts.Volumes {
		mounts = append(mounts, mount.Mount{
			Type:     mount.TypeBind,
			Source:   vol.HostPath,
			Target:   vol.ContainerPath,
			ReadOnly: vol.ReadOnly,
		})
	}

	// Convert ports to exposed ports and port bindings
	exposedPorts, portBindings := buildPortConfig(opts.Ports)

	// Create container
	resp, err := c.docker.ContainerCreate(ctx,
		&container.Config{
			Image:        opts.Image,
			Env:          envSlice,
			Labels:       opts.Labels,
			ExposedPorts: exposedPorts,
		},
		&container.HostConfig{
			Mounts:       mounts,
			PortBindings: portBindings,
			NetworkMode:  container.NetworkMode(opts.Network),
			RestartPolicy: container.RestartPolicy{
				Name: container.RestartPolicyUnlessStopped,
			},
		},
		&network.NetworkingConfig{},
		nil,
		opts.Name,
	)
	if err != nil {
		return "", fmt.Errorf("failed to create container: %w", err)
	}

	return ContainerID(resp.ID), nil
}

// StartContainer starts a container.
func (c *Client) StartContainer(ctx context.Context, containerID ContainerID) error {
	if c.docker == nil {
		return ErrNotConnected
	}

	c.log.Info("starting container", "id", containerID)

	err := c.docker.ContainerStart(ctx, containerID.String(), container.StartOptions{})
	if err != nil {
		return fmt.Errorf("failed to start container: %w", err)
	}

	return nil
}

// StopContainer stops a container.
func (c *Client) StopContainer(ctx context.Context, containerID ContainerID) error {
	if c.docker == nil {
		return ErrNotConnected
	}

	c.log.Info("stopping container", "id", containerID)

	timeout := stopTimeoutSeconds

	err := c.docker.ContainerStop(ctx, containerID.String(), container.StopOptions{Timeout: &timeout})
	if err != nil {
		return fmt.Errorf("failed to stop container: %w", err)
	}

	return nil
}

// RemoveContainer removes a container.
func (c *Client) RemoveContainer(
	ctx context.Context,
	containerID ContainerID,
	force bool,
) error {
	if c.docker == nil {
		return ErrNotConnected
	}

	c.log.Info("removing container", "id", containerID, "force", force)

	err := c.docker.ContainerRemove(ctx, containerID.String(), container.RemoveOptions{Force: force})
	if err != nil {
		return fmt.Errorf("failed to remove container: %w", err)
	}

	return nil
}

// ContainerLogs returns the logs for a container.
func (c *Client) ContainerLogs(
	ctx context.Context,
	containerID ContainerID,
	tail string,
) (string, error) {
	if c.docker == nil {
		return "", ErrNotConnected
	}

	opts := container.LogsOptions{
		ShowStdout: true,
		ShowStderr: true,
		Tail:       tail,
	}

	reader, err := c.docker.ContainerLogs(ctx, containerID.String(), opts)
	if err != nil {
		return "", fmt.Errorf("failed to get container logs: %w", err)
	}

	defer func() {
		closeErr := reader.Close()
		if closeErr != nil {
			c.log.Error("failed to close log reader", "error", closeErr)
		}
	}()

	logs, err := io.ReadAll(reader)
	if err != nil {
		return "", fmt.Errorf("failed to read container logs: %w", err)
	}

	return string(logs), nil
}

// IsContainerRunning checks if a container is running.
func (c *Client) IsContainerRunning(
	ctx context.Context,
	containerID ContainerID,
) (bool, error) {
	if c.docker == nil {
		return false, ErrNotConnected
	}

	inspect, err := c.docker.ContainerInspect(ctx, containerID.String())
	if err != nil {
		return false, fmt.Errorf("failed to inspect container: %w", err)
	}

	return inspect.State.Running, nil
}

// IsContainerHealthy checks if a container is healthy.
func (c *Client) IsContainerHealthy(
	ctx context.Context,
	containerID ContainerID,
) (bool, error) {
	if c.docker == nil {
		return false, ErrNotConnected
	}

	inspect, err := c.docker.ContainerInspect(ctx, containerID.String())
	if err != nil {
		return false, fmt.Errorf("failed to inspect container: %w", err)
	}

	// If no health check defined, consider running as healthy
	if inspect.State.Health == nil {
		return inspect.State.Running, nil
	}

	return inspect.State.Health.Status == "healthy", nil
}

// LabelUpaasID is the Docker label key used to identify containers managed by upaas.
const LabelUpaasID = "upaas.id"

// ContainerInfo contains basic information about a container.
type ContainerInfo struct {
	ID      ContainerID
	Running bool
}

// FindContainerByAppID finds a container by the upaas.id label.
// Returns nil if no container is found.
//
//nolint:nilnil // returning nil,nil is idiomatic for "not found"
func (c *Client) FindContainerByAppID(
	ctx context.Context,
	appID string,
) (*ContainerInfo, error) {
	if c.docker == nil {
		return nil, ErrNotConnected
	}

	filterArgs := filters.NewArgs()
	filterArgs.Add("label", LabelUpaasID+"="+appID)

	containers, err := c.docker.ContainerList(ctx, container.ListOptions{
		All:     true,
		Filters: filterArgs,
	})
	if err != nil {
		return nil, fmt.Errorf("failed to list containers: %w", err)
	}

	if len(containers) == 0 {
		return nil, nil
	}

	// Return the first matching container
	ctr := containers[0]

	return &ContainerInfo{
		ID:      ContainerID(ctr.ID),
		Running: ctr.State == "running",
	}, nil
}

// cloneConfig holds configuration for a git clone operation.
type cloneConfig struct {
	repoURL       string
	branch        string
	commitSHA     string // Optional: specific commit to checkout
	sshPrivateKey string
	containerDir  string // Path inside the upaas container (for file operations)
	hostDir       string // Path on the Docker host (for bind mounts)
	keyFile       string // Container path to SSH key file
	hostKeyFile   string // Host path to SSH key file
}

// CloneResult contains the result of a git clone operation.
type CloneResult struct {
	Output    string // Combined stdout/stderr from git clone
	CommitSHA string // The HEAD commit SHA after clone/checkout
}

// CloneRepo clones a git repository using SSH and optionally checks out a specific commit.
// containerDir is the path inside the upaas container (for writing files).
// hostDir is the corresponding path on the Docker host (for bind mounts).
// If commitSHA is provided, that specific commit will be checked out.
func (c *Client) CloneRepo(
	ctx context.Context,
	repoURL, branch, commitSHA, sshPrivateKey, containerDir, hostDir string,
) (*CloneResult, error) {
	// Validate inputs to prevent shell injection
	if !validBranchRe.MatchString(branch) {
		return nil, fmt.Errorf("%w: %q", ErrInvalidBranch, branch)
	}

	if commitSHA != "" && !validCommitSHARe.MatchString(commitSHA) {
		return nil, fmt.Errorf("%w: %q", ErrInvalidCommitSHA, commitSHA)
	}

	if c.docker == nil {
		return nil, ErrNotConnected
	}

	c.log.Info("cloning repository",
		"url", repoURL,
		"branch", branch,
		"commit", commitSHA,
		"containerDir", containerDir,
		"hostDir", hostDir,
	)

	// Clone to 'work' subdirectory, SSH key stays in build directory root
	cfg := &cloneConfig{
		repoURL:       repoURL,
		branch:        branch,
		commitSHA:     commitSHA,
		sshPrivateKey: sshPrivateKey,
		containerDir:  filepath.Join(containerDir, "work"),
		hostDir:       filepath.Join(hostDir, "work"),
		keyFile:       filepath.Join(containerDir, "deploy_key"),
		hostKeyFile:   filepath.Join(hostDir, "deploy_key"),
	}

	return c.performClone(ctx, cfg)
}

// RemoveImage removes a Docker image by ID or tag.
// It returns nil if the image was successfully removed or does not exist.
func (c *Client) RemoveImage(ctx context.Context, imageID ImageID) error {
	_, err := c.docker.ImageRemove(ctx, imageID.String(), image.RemoveOptions{
		Force:         true,
		PruneChildren: true,
	})
	if err != nil && !client.IsErrNotFound(err) {
		return fmt.Errorf("failed to remove image %s: %w", imageID, err)
	}

	return nil
}

func (c *Client) performBuild(
	ctx context.Context,
	opts BuildImageOptions,
) (ImageID, error) {
	// Create tar archive of build context
	tarArchive, err := archive.TarWithOptions(opts.ContextDir, &archive.TarOptions{})
	if err != nil {
		return "", fmt.Errorf("failed to create build context: %w", err)
	}

	defer func() {
		closeErr := tarArchive.Close()
		if closeErr != nil {
			c.log.Error("failed to close tar archive", "error", closeErr)
		}
	}()

	// Build image
	resp, err := c.docker.ImageBuild(ctx, tarArchive, dockertypes.ImageBuildOptions{
		Dockerfile: opts.DockerfilePath,
		Tags:       opts.Tags,
		Remove:     true,
		NoCache:    false,
	})
	if err != nil {
		return "", fmt.Errorf("failed to build image: %w", err)
	}

	defer func() {
		closeErr := resp.Body.Close()
		if closeErr != nil {
			c.log.Error("failed to close response body", "error", closeErr)
		}
	}()

	// Stream build output line by line for real-time log updates
	err = c.streamBuildOutput(resp.Body, opts.LogWriter)
	if err != nil {
		return "", err
	}

	// Get image ID
	if len(opts.Tags) > 0 {
		inspect, _, inspectErr := c.docker.ImageInspectWithRaw(ctx, opts.Tags[0])
		if inspectErr != nil {
			return "", fmt.Errorf("failed to inspect image: %w", inspectErr)
		}

		return ImageID(inspect.ID), nil
	}

	return "", nil
}

// scannerInitialBufferSize is the initial buffer size for the build log scanner.
const scannerInitialBufferSize = 64 * 1024 // 64KB

// scannerMaxBufferSize is the max buffer size for build log lines (base64 layers can be large).
const scannerMaxBufferSize = 1024 * 1024 // 1MB

// streamBuildOutput reads Docker build output line by line and writes to stdout and optional log writer.
// Docker sends newline-delimited JSON, so reading line by line ensures each log entry is written immediately.
func (c *Client) streamBuildOutput(body io.Reader, logWriter io.Writer) error {
	scanner := bufio.NewScanner(body)
	buf := make([]byte, 0, scannerInitialBufferSize)
	scanner.Buffer(buf, scannerMaxBufferSize)

	newline := []byte{'\n'}

	for scanner.Scan() {
		line := scanner.Bytes()
		// Write to stdout
		_, _ = os.Stdout.Write(line)
		_, _ = os.Stdout.Write(newline)
		// Write to log writer if provided
		if logWriter != nil {
			_, _ = logWriter.Write(line)
			_, _ = logWriter.Write(newline)
		}
	}

	scanErr := scanner.Err()
	if scanErr != nil {
		return fmt.Errorf("failed to read build output: %w", scanErr)
	}

	return nil
}

func (c *Client) performClone(ctx context.Context, cfg *cloneConfig) (*CloneResult, error) {
	// Create work directory for clone destination
	err := os.MkdirAll(cfg.containerDir, workDirPermissions)
	if err != nil {
		return nil, fmt.Errorf("failed to create work dir: %w", err)
	}

	// Write SSH key to temp file
	err = os.WriteFile(cfg.keyFile, []byte(cfg.sshPrivateKey), sshKeyPermissions)
	if err != nil {
		return nil, fmt.Errorf("failed to write SSH key: %w", err)
	}

	defer func() {
		removeErr := os.Remove(cfg.keyFile)
		if removeErr != nil {
			c.log.Error("failed to remove SSH key file", "error", removeErr)
		}
	}()

	gitContainerID, err := c.createGitContainer(ctx, cfg)
	if err != nil {
		return nil, err
	}

	defer func() {
		_ = c.docker.ContainerRemove(ctx, gitContainerID.String(), container.RemoveOptions{Force: true})
	}()

	return c.runGitClone(ctx, gitContainerID)
}

func (c *Client) createGitContainer(
	ctx context.Context,
	cfg *cloneConfig,
) (ContainerID, error) {
	gitSSHCmd := "ssh -i /keys/deploy_key -o StrictHostKeyChecking=no"

	// Build the git command using environment variables to avoid shell injection.
	// Arguments are passed via env vars and quoted in the shell script.
	var script string
	if cfg.commitSHA != "" {
		// Clone without depth limit so we can checkout any commit, then checkout specific SHA
		script = `git clone --branch "$CLONE_BRANCH" "$CLONE_URL" /repo` +
			` && cd /repo && git checkout "$CLONE_SHA"` +
			` && echo COMMIT:$(git rev-parse HEAD)`
	} else {
		// Shallow clone of branch HEAD, then output commit SHA
		script = `git clone --depth 1 --branch "$CLONE_BRANCH" "$CLONE_URL" /repo` +
			` && cd /repo && echo COMMIT:$(git rev-parse HEAD)`
	}

	env := []string{
		"GIT_SSH_COMMAND=" + gitSSHCmd,
		"CLONE_URL=" + cfg.repoURL,
		"CLONE_BRANCH=" + cfg.branch,
	}
	if cfg.commitSHA != "" {
		env = append(env, "CLONE_SHA="+cfg.commitSHA)
	}

	entrypoint := []string{}
	cmd := []string{"sh", "-c", script}

	// Use host paths for Docker bind mounts (Docker runs on the host, not in our container)
	resp, err := c.docker.ContainerCreate(ctx,
		&container.Config{
			Image:      gitImage,
			Entrypoint: entrypoint,
			Cmd:        cmd,
			Env:        env,
			WorkingDir: "/",
		},
		&container.HostConfig{
			Mounts: []mount.Mount{
				{Type: mount.TypeBind, Source: cfg.hostDir, Target: "/repo"},
				{
					Type:     mount.TypeBind,
					Source:   cfg.hostKeyFile,
					Target:   "/keys/deploy_key",
					ReadOnly: true,
				},
			},
		},
		nil,
		nil,
		"",
	)
	if err != nil {
		return "", fmt.Errorf("failed to create git container: %w", err)
	}

	return ContainerID(resp.ID), nil
}

func (c *Client) runGitClone(ctx context.Context, containerID ContainerID) (*CloneResult, error) {
	err := c.docker.ContainerStart(ctx, containerID.String(), container.StartOptions{})
	if err != nil {
		return nil, fmt.Errorf("failed to start git container: %w", err)
	}

	statusCh, errCh := c.docker.ContainerWait(ctx, containerID.String(), container.WaitConditionNotRunning)

	select {
	case err := <-errCh:
		return nil, fmt.Errorf("error waiting for git container: %w", err)
	case status := <-statusCh:
		// Always capture logs for the result
		logs, _ := c.ContainerLogs(ctx, containerID, "100")

		if status.StatusCode != 0 {
			return nil, fmt.Errorf(
				"%w with status %d: %s",
				ErrGitCloneFailed,
				status.StatusCode,
				logs,
			)
		}

		// Parse commit SHA from output (looks for "COMMIT:<sha>" line)
		commitSHA := parseCommitSHA(logs)

		return &CloneResult{Output: logs, CommitSHA: commitSHA}, nil
	}
}

// commitMarker is the prefix used to identify commit SHA in clone output.
const commitMarker = "COMMIT:"

// parseCommitSHA extracts the commit SHA from git clone output.
// It looks for a line starting with "COMMIT:" and returns the SHA after it.
func parseCommitSHA(output string) string {
	for line := range strings.SplitSeq(output, "\n") {
		line = strings.TrimSpace(line)

		sha, found := strings.CutPrefix(line, commitMarker)
		if found {
			return strings.TrimSpace(sha)
		}
	}

	return ""
}

func (c *Client) connect(ctx context.Context) error {
	opts := []client.Opt{
		client.FromEnv,
		client.WithAPIVersionNegotiation(),
	}

	if c.params.Config.DockerHost != "" {
		opts = append(opts, client.WithHost(c.params.Config.DockerHost))
	}

	docker, err := client.NewClientWithOpts(opts...)
	if err != nil {
		return fmt.Errorf("failed to create Docker client: %w", err)
	}

	// Test connection
	_, err = docker.Ping(ctx)
	if err != nil {
		return fmt.Errorf("failed to ping Docker: %w", err)
	}

	c.docker = docker
	c.log.Info("docker client connected")

	return nil
}

func (c *Client) close() error {
	if c.docker != nil {
		err := c.docker.Close()
		if err != nil {
			return fmt.Errorf("failed to close docker client: %w", err)
		}
	}

	return nil
}