| 1 | // Package jmxbridge provides process management for JSON-based Java helper bridges. |
| 2 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 3 | |
| 4 | package jmxbridge |
| 5 | |
| 6 | import ( |
| 7 | "bufio" |
| 8 | "context" |
| 9 | "encoding/json" |
| 10 | "errors" |
| 11 | "fmt" |
| 12 | "io" |
| 13 | "os" |
| 14 | "os/exec" |
| 15 | "strings" |
| 16 | "sync" |
| 17 | ) |
| 18 | |
| 19 | // Logger is the minimal logging contract required by the bridge. |
| 20 | type Logger interface { |
| 21 | Debugf(format string, args ...any) |
| 22 | Infof(format string, args ...any) |
| 23 | Warningf(format string, args ...any) |
| 24 | Errorf(format string, args ...any) |
| 25 | } |
| 26 | |
| 27 | // Config contains options for launching the helper process. |
| 28 | type Config struct { |
| 29 | JavaExecPath string |
| 30 | JarPath string |
| 31 | JarData []byte |
| 32 | JarFileName string |
| 33 | WorkingDir string |
| 34 | } |
| 35 | |
| 36 | // Command represents a JSON command sent to the helper. |
| 37 | type Command map[string]any |
| 38 | |
| 39 | // Response represents a generic helper response. |
| 40 | type Response struct { |
| 41 | Status string `json:"status"` |
| 42 | Message string `json:"message,omitempty"` |
| 43 | Details string `json:"details,omitempty"` |
| 44 | Recoverable bool `json:"recoverable,omitempty"` |
| 45 | Data map[string]any `json:"data,omitempty"` |
| 46 | } |
| 47 | |
| 48 | // Option configures a Client. |
| 49 | type Option func(*Client) |
| 50 | |
| 51 | // ProcessFactory builds a process abstraction. It may be overridden for tests. |
| 52 | type ProcessFactory func(ctx context.Context, javaPath string, jarPath string) (process, error) |
| 53 | |
| 54 | type process interface { |
| 55 | Start() error |
| 56 | Stdin() io.WriteCloser |
| 57 | Stdout() io.ReadCloser |
| 58 | Stderr() io.ReadCloser |
| 59 | Wait() error |
| 60 | } |
| 61 | |
| 62 | // Client manages the lifecycle of the helper process. |
| 63 | type Client struct { |
| 64 | cfg Config |
| 65 | logger Logger |
| 66 | |
| 67 | processFactory ProcessFactory |
| 68 | |
| 69 | mu sync.Mutex |
| 70 | proc process |
| 71 | cancel context.CancelFunc |
| 72 | stdin io.WriteCloser |
| 73 | scanner *bufio.Scanner |
| 74 | stderr io.ReadCloser |
| 75 | stderrWG sync.WaitGroup |
| 76 | running bool |
| 77 | |
| 78 | jarPath string |
| 79 | removeJar bool |
| 80 | } |
| 81 | |
| 82 | // NewClient creates a Client with optional customisations. |
| 83 | func NewClient(cfg Config, logger Logger, opts ...Option) (*Client, error) { |
| 84 | if logger == nil { |
| 85 | return nil, errors.New("jmxbridge: logger is required") |
| 86 | } |
| 87 | |
| 88 | if cfg.JarPath == "" && len(cfg.JarData) == 0 { |
| 89 | return nil, errors.New("jmxbridge: either JarPath or JarData must be provided") |
| 90 | } |
| 91 | |
| 92 | if cfg.JarFileName == "" { |
| 93 | cfg.JarFileName = "netdata_jmx_helper.jar" |
| 94 | } |
| 95 | |
| 96 | client := &Client{ |
| 97 | cfg: cfg, |
| 98 | logger: logger, |
| 99 | processFactory: defaultProcessFactory, |
| 100 | } |
| 101 | |
| 102 | for _, opt := range opts { |
| 103 | opt(client) |
| 104 | } |
| 105 | |
| 106 | return client, nil |
| 107 | } |
| 108 | |
| 109 | // WithProcessFactory overrides the process factory (useful for tests). |
| 110 | func WithProcessFactory(factory ProcessFactory) Option { |
| 111 | return func(c *Client) { |
| 112 | c.processFactory = factory |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | // Start launches the helper and sends the initial command. |
| 117 | func (c *Client) Start(ctx context.Context, initCmd Command) error { |
| 118 | c.mu.Lock() |
| 119 | defer c.mu.Unlock() |
| 120 | |
| 121 | if c.running { |
| 122 | return nil |
| 123 | } |
| 124 | |
| 125 | jarPath, removeJar, err := c.prepareJar() |
| 126 | if err != nil { |
| 127 | return err |
| 128 | } |
| 129 | |
| 130 | procCtx, cancel := context.WithCancel(context.Background()) |
| 131 | proc, err := c.processFactory(procCtx, c.cfg.JavaExecPath, jarPath) |
| 132 | if err != nil { |
| 133 | cancel() |
| 134 | if removeJar { |
| 135 | _ = os.Remove(jarPath) |
| 136 | } |
| 137 | return err |
| 138 | } |
| 139 | |
| 140 | stdin := proc.Stdin() |
| 141 | stdout := proc.Stdout() |
| 142 | stderr := proc.Stderr() |
| 143 | |
| 144 | if err := proc.Start(); err != nil { |
| 145 | cancel() |
| 146 | if removeJar { |
| 147 | _ = os.Remove(jarPath) |
| 148 | } |
| 149 | return fmt.Errorf("jmxbridge: failed to start helper: %w", err) |
| 150 | } |
| 151 | |
| 152 | c.stderrWG.Add(1) |
| 153 | go c.consumeStderr(stderr) |
| 154 | |
| 155 | c.proc = proc |
| 156 | c.cancel = cancel |
| 157 | c.stdin = stdin |
| 158 | c.scanner = bufio.NewScanner(stdout) |
| 159 | c.stderr = stderr |
| 160 | c.running = true |
| 161 | c.jarPath = jarPath |
| 162 | c.removeJar = removeJar |
| 163 | |
| 164 | if initCmd != nil { |
| 165 | c.mu.Unlock() |
| 166 | _, err := c.Send(ctx, initCmd) |
| 167 | c.mu.Lock() |
| 168 | if err != nil { |
| 169 | c.internalShutdownLocked() |
| 170 | return fmt.Errorf("jmxbridge: helper init failed: %w", err) |
| 171 | } |
| 172 | } |
| 173 | |
| 174 | c.logger.Infof("jmxbridge: helper started") |
| 175 | return nil |
| 176 | } |
| 177 | |
| 178 | // Send transmits a command and waits for the corresponding response. |
| 179 | func (c *Client) Send(ctx context.Context, cmd Command) (*Response, error) { |
| 180 | c.mu.Lock() |
| 181 | defer c.mu.Unlock() |
| 182 | |
| 183 | if !c.running { |
| 184 | return nil, errors.New("jmxbridge: helper not running") |
| 185 | } |
| 186 | |
| 187 | payload, err := json.Marshal(cmd) |
| 188 | if err != nil { |
| 189 | return nil, fmt.Errorf("jmxbridge: failed to marshal command: %w", err) |
| 190 | } |
| 191 | |
| 192 | if _, err := c.stdin.Write(append(payload, '\n')); err != nil { |
| 193 | return nil, fmt.Errorf("jmxbridge: failed to send command: %w", err) |
| 194 | } |
| 195 | |
| 196 | scanner := c.scanner |
| 197 | if scanner == nil { |
| 198 | return nil, errors.New("jmxbridge: response scanner not initialised") |
| 199 | } |
| 200 | |
| 201 | type result struct { |
| 202 | line string |
| 203 | err error |
| 204 | } |
| 205 | |
| 206 | resCh := make(chan result, 1) |
| 207 | go func(s *bufio.Scanner) { |
| 208 | if s.Scan() { |
| 209 | resCh <- result{line: s.Text()} |
| 210 | } else { |
| 211 | resCh <- result{err: s.Err()} |
| 212 | } |
| 213 | }(scanner) |
| 214 | |
| 215 | select { |
| 216 | case <-ctx.Done(): |
| 217 | return nil, fmt.Errorf("jmxbridge: waiting for response cancelled: %w", ctx.Err()) |
| 218 | case res := <-resCh: |
| 219 | if res.err != nil { |
| 220 | return nil, fmt.Errorf("jmxbridge: failed to read response: %w", res.err) |
| 221 | } |
| 222 | var resp Response |
| 223 | if err := json.Unmarshal([]byte(res.line), &resp); err != nil { |
| 224 | return nil, fmt.Errorf("jmxbridge: invalid response: %w", err) |
| 225 | } |
| 226 | status := strings.ToUpper(resp.Status) |
| 227 | if status != "OK" && status != "SUCCESS" { |
| 228 | return &resp, fmt.Errorf("jmxbridge: helper returned status %s: %s", resp.Status, resp.Message) |
| 229 | } |
| 230 | return &resp, nil |
| 231 | } |
| 232 | } |
| 233 | |
| 234 | // Shutdown terminates the helper process and cleans resources. |
| 235 | func (c *Client) Shutdown() { |
| 236 | c.mu.Lock() |
| 237 | defer c.mu.Unlock() |
| 238 | c.internalShutdownLocked() |
| 239 | } |
| 240 | |
| 241 | func (c *Client) internalShutdownLocked() { |
| 242 | if !c.running { |
| 243 | return |
| 244 | } |
| 245 | |
| 246 | if c.cancel != nil { |
| 247 | c.cancel() |
| 248 | } |
| 249 | |
| 250 | if c.stdin != nil { |
| 251 | _ = c.stdin.Close() |
| 252 | } |
| 253 | |
| 254 | if c.proc != nil { |
| 255 | _ = c.proc.Wait() |
| 256 | } |
| 257 | |
| 258 | if c.stderr != nil { |
| 259 | c.stderr.Close() |
| 260 | } |
| 261 | c.stderrWG.Wait() |
| 262 | |
| 263 | if c.removeJar { |
| 264 | _ = os.Remove(c.jarPath) |
| 265 | } |
| 266 | |
| 267 | c.proc = nil |
| 268 | c.stdin = nil |
| 269 | c.scanner = nil |
| 270 | c.stderr = nil |
| 271 | c.cancel = nil |
| 272 | c.running = false |
| 273 | c.logger.Infof("jmxbridge: helper stopped") |
| 274 | } |
| 275 | |
| 276 | func (c *Client) prepareJar() (string, bool, error) { |
| 277 | if c.cfg.JarPath != "" { |
| 278 | return c.cfg.JarPath, false, nil |
| 279 | } |
| 280 | |
| 281 | dir := c.cfg.WorkingDir |
| 282 | if dir == "" { |
| 283 | dir = os.TempDir() |
| 284 | } |
| 285 | |
| 286 | f, err := os.CreateTemp(dir, c.cfg.JarFileName) |
| 287 | if err != nil { |
| 288 | return "", false, fmt.Errorf("jmxbridge: failed to create temp jar: %w", err) |
| 289 | } |
| 290 | if _, err := f.Write(c.cfg.JarData); err != nil { |
| 291 | _ = f.Close() |
| 292 | _ = os.Remove(f.Name()) |
| 293 | return "", false, fmt.Errorf("jmxbridge: failed to write jar: %w", err) |
| 294 | } |
| 295 | if err := f.Close(); err != nil { |
| 296 | _ = os.Remove(f.Name()) |
| 297 | return "", false, fmt.Errorf("jmxbridge: failed to close jar: %w", err) |
| 298 | } |
| 299 | |
| 300 | return f.Name(), true, nil |
| 301 | } |
| 302 | |
| 303 | func (c *Client) consumeStderr(r io.Reader) { |
| 304 | defer c.stderrWG.Done() |
| 305 | scanner := bufio.NewScanner(r) |
| 306 | for scanner.Scan() { |
| 307 | c.logger.Debugf("jmxbridge stderr: %s", scanner.Text()) |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | func defaultProcessFactory(ctx context.Context, javaPath, jarPath string) (process, error) { |
| 312 | if javaPath == "" { |
| 313 | javaPath = "java" |
| 314 | } |
| 315 | |
| 316 | cmd := exec.CommandContext(ctx, javaPath, "-jar", jarPath) |
| 317 | stdin, err := cmd.StdinPipe() |
| 318 | if err != nil { |
| 319 | return nil, err |
| 320 | } |
| 321 | stdout, err := cmd.StdoutPipe() |
| 322 | if err != nil { |
| 323 | return nil, err |
| 324 | } |
| 325 | stderr, err := cmd.StderrPipe() |
| 326 | if err != nil { |
| 327 | return nil, err |
| 328 | } |
| 329 | |
| 330 | return &execProcess{ |
| 331 | cmd: cmd, |
| 332 | stdin: stdin, |
| 333 | stdout: stdout, |
| 334 | stderr: stderr, |
| 335 | }, nil |
| 336 | } |
| 337 | |
| 338 | type execProcess struct { |
| 339 | cmd *exec.Cmd |
| 340 | stdin io.WriteCloser |
| 341 | stdout io.ReadCloser |
| 342 | stderr io.ReadCloser |
| 343 | } |
| 344 | |
| 345 | func (p *execProcess) Start() error { return p.cmd.Start() } |
| 346 | func (p *execProcess) Stdin() io.WriteCloser { return p.stdin } |
| 347 | func (p *execProcess) Stdout() io.ReadCloser { return p.stdout } |
| 348 | func (p *execProcess) Stderr() io.ReadCloser { return p.stderr } |
| 349 | func (p *execProcess) Wait() error { return p.cmd.Wait() } |