Slave.java 19.0 KB
Newer Older
K
kohsuke 已提交
1 2 3 4
package hudson.model;

import hudson.FilePath;
import hudson.Launcher;
5
import hudson.Launcher.RemoteLauncher;
K
kohsuke 已提交
6
import hudson.Util;
7 8
import hudson.maven.agent.Main;
import hudson.maven.agent.PluginManagerInterceptor;
K
kohsuke 已提交
9
import hudson.model.Descriptor.FormException;
K
kohsuke 已提交
10 11
import hudson.remoting.Callable;
import hudson.remoting.Channel;
12
import hudson.remoting.Channel.Listener;
K
kohsuke 已提交
13
import hudson.remoting.VirtualChannel;
14
import hudson.remoting.Which;
15 16 17
import hudson.tasks.DynamicLabeler;
import hudson.tasks.LabelFinder;
import hudson.util.*;
18 19
import org.kohsuke.stapler.StaplerRequest;
import org.kohsuke.stapler.StaplerResponse;
K
kohsuke 已提交
20

21 22
import javax.servlet.ServletException;
import javax.servlet.http.HttpServletResponse;
23
import java.io.*;
24 25
import java.net.URL;
import java.net.URLConnection;
26
import java.util.*;
K
kohsuke 已提交
27
import java.util.logging.Level;
K
kohsuke 已提交
28
import java.util.logging.LogRecord;
29
import java.util.logging.Logger;
K
kohsuke 已提交
30

K
kohsuke 已提交
31 32 33 34 35
/**
 * Information about a Hudson slave node.
 *
 * @author Kohsuke Kawaguchi
 */
K
kohsuke 已提交
36
public final class Slave implements Node, Serializable {
K
kohsuke 已提交
37
    /**
K
kohsuke 已提交
38
     * Name of this slave node.
K
kohsuke 已提交
39
     */
K
kohsuke 已提交
40
    protected final String name;
K
kohsuke 已提交
41 42 43 44 45 46 47 48

    /**
     * Description of this node.
     */
    private final String description;

    /**
     * Path to the root of the workspace
K
kohsuke 已提交
49
     * from the view point of this node, such as "/hudson"
K
kohsuke 已提交
50
     */
K
kohsuke 已提交
51
    protected final String remoteFS;
K
kohsuke 已提交
52 53 54 55 56 57 58 59 60 61 62

    /**
     * Number of executors of this node.
     */
    private int numExecutors = 2;

    /**
     * Job allocation strategy.
     */
    private Mode mode;

K
kohsuke 已提交
63 64 65 66 67 68
    /**
     * Command line to launch the agent, like
     * "ssh myslave java -jar /path/to/hudson-remoting.jar"
     */
    private String agentCommand;

69 70 71 72 73 74 75 76 77 78
    /**
     * Whitespace-separated labels.
     */
    private String label="";

    /**
     * Lazily computed set of labels from {@link #label}.
     */
    private transient volatile Set<Label> labels;

79 80 81
    private transient volatile Set<Label> dynamicLabels;
    private transient volatile int dynamicLabelsInstanceHash;

K
kohsuke 已提交
82 83 84
    /**
     * @stapler-constructor
     */
85
    public Slave(String name, String description, String command, String remoteFS, int numExecutors, Mode mode,
86
                 String label) throws FormException {
K
kohsuke 已提交
87 88 89 90
        this.name = name;
        this.description = description;
        this.numExecutors = numExecutors;
        this.mode = mode;
K
kohsuke 已提交
91 92
        this.agentCommand = command;
        this.remoteFS = remoteFS;
93 94
        this.label = Util.fixNull(label).trim();
        getAssignedLabels();    // compute labels now
K
kohsuke 已提交
95

K
d'oh!  
kohsuke 已提交
96
        if (name.equals(""))
K
kohsuke 已提交
97
            throw new FormException("Invalid slave configuration. Name is empty", null);
K
kohsuke 已提交
98

99 100 101 102 103
        // this prevents the config from being saved when slaves are offline.
        // on a large deployment with a lot of slaves, some slaves are bound to be offline,
        // so this check is harmful.
        //if (!localFS.exists())
        //    throw new FormException("Invalid slave configuration for " + name + ". No such directory exists: " + localFS, null);
K
kohsuke 已提交
104 105
        if (remoteFS.equals(""))
            throw new FormException("Invalid slave configuration for " + name + ". No remote directory given", null);
K
kohsuke 已提交
106 107 108
    }

    public String getCommand() {
K
kohsuke 已提交
109
        return agentCommand;
K
kohsuke 已提交
110 111 112 113 114 115
    }

    public String getRemoteFS() {
        return remoteFS;
    }

K
kohsuke 已提交
116 117
    public String getNodeName() {
        return name;
K
kohsuke 已提交
118 119 120 121 122 123 124 125 126 127 128 129 130 131
    }

    public String getNodeDescription() {
        return description;
    }

    public int getNumExecutors() {
        return numExecutors;
    }

    public Mode getMode() {
        return mode;
    }

132 133 134
    public String getLabelString() {
        return Util.fixNull(label).trim();
    }
135

136
    public Set<Label> getAssignedLabels() {
137 138
        // todo refactor to make dynamic labels a bit less hacky
        if(labels==null || isChangedDynamicLabels()) {
139 140 141 142 143 144 145 146
            Set<Label> r = new HashSet<Label>();
            String ls = getLabelString();
            if(ls.length()>0) {
                for( String l : ls.split(" +")) {
                    r.add(Hudson.getInstance().getLabel(l));
                }
            }
            r.add(getSelfLabel());
147
            r.addAll(getDynamicLabels());
148 149 150 151 152
            this.labels = Collections.unmodifiableSet(r);
        }
        return labels;
    }

153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198
    /**
     * Check if we should rebuild the list of dynamic labels.
     * @todo make less hacky
     * @return
     */
    private boolean isChangedDynamicLabels() {
        Computer comp = getComputer();
        if (comp == null) {
            return dynamicLabelsInstanceHash != 0;
        } else {
            if (dynamicLabelsInstanceHash == comp.hashCode()) {
                return false;
            }
            dynamicLabels = null; // force a re-calc
            return true;
        }
    }

    /**
     * Returns the possibly empty set of labels that it has been determined as supported by this node.
     *
     * @todo make less hacky
     * @see hudson.tasks.LabelFinder
     */
    public Set<Label> getDynamicLabels() {
        if (dynamicLabels == null) {
            synchronized (this) {
                if (dynamicLabels == null) {
                    dynamicLabels = new HashSet<Label>();
                    Computer computer = getComputer();
                    VirtualChannel channel;
                    if (computer != null && (channel = computer.getChannel()) != null) {
                        dynamicLabelsInstanceHash = computer.hashCode();
                        for (DynamicLabeler labeler : LabelFinder.LABELERS) {
                            for (String label : labeler.findLabels(channel)) {
                                dynamicLabels.add(Hudson.getInstance().getLabel(label));
                            }
                        }
                    } else {
                        dynamicLabelsInstanceHash = 0;
                    }
                }
            }
        }
        return dynamicLabels;
    }
199 200 201 202 203

    public Label getSelfLabel() {
        return Hudson.getInstance().getLabel(name);
    }

204
    public ClockDifference getClockDifference() throws IOException, InterruptedException {
K
kohsuke 已提交
205
        VirtualChannel channel = getComputer().getChannel();
K
kohsuke 已提交
206 207
        if(channel==null)
            throw new IOException(getNodeName()+" is offline");
K
kohsuke 已提交
208

K
kohsuke 已提交
209 210 211 212 213 214 215
        long startTime = System.currentTimeMillis();
        long slaveTime = channel.call(new Callable<Long,RuntimeException>() {
            public Long call() {
                return System.currentTimeMillis();
            }
        });
        long endTime = System.currentTimeMillis();
K
kohsuke 已提交
216

217
        return new ClockDifference((startTime+endTime)/2 - slaveTime);
K
kohsuke 已提交
218 219
    }

K
kohsuke 已提交
220 221 222 223
    public Computer createComputer() {
        return new ComputerImpl(this);
    }

224
    public FilePath getWorkspaceFor(TopLevelItem item) {
K
kohsuke 已提交
225 226 227
        FilePath r = getWorkspaceRoot();
        if(r==null)     return null;    // offline
        return r.child(item.getName());
228 229
    }

K
kohsuke 已提交
230 231 232 233 234 235
    public FilePath getRootPath() {
        VirtualChannel ch = getComputer().getChannel();
        if(ch==null)    return null;    // offline
        return new FilePath(ch,remoteFS);
    }

K
kohsuke 已提交
236 237
    /**
     * Root directory on this slave where all the job workspaces are laid out.
K
kohsuke 已提交
238 239
     * @return
     *      null if not connected.
K
kohsuke 已提交
240 241
     */
    public FilePath getWorkspaceRoot() {
K
kohsuke 已提交
242 243 244
        FilePath r = getRootPath();
        if(r==null) return null;
        return r.child("workspace");
K
kohsuke 已提交
245
    }
K
kohsuke 已提交
246

K
kohsuke 已提交
247 248
    public static final class ComputerImpl extends Computer {
        private volatile Channel channel;
249
        private Boolean isUnix;
K
kohsuke 已提交
250

K
kohsuke 已提交
251 252 253 254 255 256
        /**
         * This is where the log from the remote agent goes.
         */
        private File getLogFile() {
            return new File(Hudson.getInstance().getRootDir(),"slave-"+nodeName+".log");
        }
K
kohsuke 已提交
257

K
kohsuke 已提交
258 259 260 261
        private ComputerImpl(Slave slave) {
            super(slave);
        }

262 263 264 265 266 267 268 269 270
        public Slave getNode() {
            return (Slave)super.getNode();
        }

        @Override
        public boolean isJnlpAgent() {
            return getNode().getCommand().length()==0;
        }

271 272 273 274 275 276 277
        /**
         * Gets the formatted current time stamp.
         */
        private static String getTimestamp() {
            return String.format("[%1$tD %1$tT]",new Date());
        }

K
kohsuke 已提交
278 279 280 281 282 283
        /**
         * Launches a remote agent.
         */
        private void launch(final Slave slave) {
            closeChannel();

K
kohsuke 已提交
284
            final OutputStream launchLog = openLogFile();
K
kohsuke 已提交
285

286 287 288 289 290 291 292 293
            if(slave.agentCommand.length()>0) {
                // launch the slave agent asynchronously
                threadPoolForRemoting.execute(new Runnable() {
                    // TODO: do this only for nodes that are so configured.
                    // TODO: support passive connection via JNLP
                    public void run() {
                        final StreamTaskListener listener = new StreamTaskListener(launchLog);
                        try {
294
                            listener.getLogger().printf("%s Launching slave agent\n",getTimestamp());
295 296 297 298 299 300 301 302 303 304 305
                            listener.getLogger().println("$ "+slave.agentCommand);
                            final Process proc = Runtime.getRuntime().exec(slave.agentCommand);

                            // capture error information from stderr. this will terminate itself
                            // when the process is killed.
                            new StreamCopyThread("stderr copier for remote agent on "+slave.getNodeName(),
                                proc.getErrorStream(), launchLog).start();

                            setChannel(proc.getInputStream(),proc.getOutputStream(),launchLog,new Listener() {
                                public void onClosed(Channel channel, IOException cause) {
                                    if(cause!=null)
K
kohsuke 已提交
306
                                        cause.printStackTrace(listener.error("%s slave agent was terminated\n",getTimestamp()));
307 308 309 310 311 312
                                    proc.destroy();
                                }
                            });

                            logger.info("slave agent launched for "+slave.getNodeName());

313 314
                        } catch (InterruptedException e) {
                            e.printStackTrace(listener.error("aborted"));
315 316 317 318 319 320 321 322 323 324
                        } catch (IOException e) {
                            Util.displayIOException(e,listener);

                            String msg = Util.getWin32ErrorMessage(e);
                            if(msg==null)   msg="";
                            else            msg=" : "+msg;
                            msg = "Unable to launch the slave agent for " + slave.getNodeName() + msg;
                            logger.log(Level.SEVERE,msg,e);
                            e.printStackTrace(listener.error(msg));
                        }
K
kohsuke 已提交
325
                    }
326 327
                });
            }
K
kohsuke 已提交
328
        }
K
kohsuke 已提交
329

K
kohsuke 已提交
330 331 332 333 334 335 336 337 338 339 340
        public OutputStream openLogFile() {
            OutputStream os;
            try {
                os = new FileOutputStream(getLogFile());
            } catch (FileNotFoundException e) {
                logger.log(Level.SEVERE, "Failed to create log file "+getLogFile(),e);
                os = new NullStream();
            }
            return os;
        }

341 342
        private final Object channelLock = new Object();

K
kohsuke 已提交
343 344 345
        /**
         * Creates a {@link Channel} from the given stream and sets that to this slave.
         */
346
        public void setChannel(InputStream in, OutputStream out, OutputStream launchLog, Listener listener) throws IOException, InterruptedException {
347
            synchronized(channelLock) {
K
kohsuke 已提交
348 349 350
                if(this.channel!=null)
                    throw new IllegalStateException("Already connected");

351
                Channel channel = new Channel(nodeName,threadPoolForRemoting, Channel.Mode.NEGOTIATE, 
K
kohsuke 已提交
352 353 354 355 356 357 358
                    in,out, launchLog);
                channel.addListener(new Listener() {
                    public void onClosed(Channel c,IOException cause) {
                        ComputerImpl.this.channel = null;
                    }
                });
                channel.addListener(listener);
359

360 361
                PrintWriter log = new PrintWriter(launchLog,true);

362 363 364 365
                {// send jars that we need for our operations
                    // TODO: maybe I should generalize this kind of "post initialization" processing
                    FilePath dst = new FilePath(channel,getNode().getRemoteFS());
                    new FilePath(Which.jarFile(Main.class)).copyTo(dst.child("maven-agent.jar"));
366
                    log.println("Copied maven-agent.jar");
367
                    new FilePath(Which.jarFile(PluginManagerInterceptor.class)).copyTo(dst.child("maven-interceptor.jar"));
368
                    log.println("Copied maven-interceptor.jar");
369 370
                }

K
kohsuke 已提交
371
                isUnix = channel.call(new DetectOS());
372 373
                log.println(isUnix?"This is a Unix slave":"This is a Windows slave");

K
kohsuke 已提交
374
                // install log handler
K
kohsuke 已提交
375
                channel.call(new LogInstaller());
K
kohsuke 已提交
376 377


378 379
                // prevent others from seeing a channel that's not properly initialized yet
                this.channel = channel;
K
kohsuke 已提交
380 381 382 383
            }
            Hudson.getInstance().getQueue().scheduleMaintenance();
        }

K
kohsuke 已提交
384 385 386 387 388
        @Override
        public VirtualChannel getChannel() {
            return channel;
        }

K
kohsuke 已提交
389 390 391 392 393 394 395 396 397 398 399
        public List<LogRecord> getLogRecords() throws IOException, InterruptedException {
            if(channel==null)
                return Collections.emptyList();
            else
                return channel.call(new Callable<List<LogRecord>,RuntimeException>() {
                    public List<LogRecord> call() {
                        return new ArrayList<LogRecord>(SLAVE_LOG_HANDLER.getView());
                    }
                });
        }

K
kohsuke 已提交
400
        public void doDoDisconnect(StaplerRequest req, StaplerResponse rsp) throws IOException, ServletException {
401
            Hudson.getInstance().checkPermission(Hudson.ADMINISTER);
K
kohsuke 已提交
402 403 404 405
            closeChannel();
            rsp.sendRedirect(".");
        }

K
kohsuke 已提交
406 407 408 409
        public void doLaunchSlaveAgent(StaplerRequest req, StaplerResponse rsp) throws IOException, ServletException {
            if(channel!=null) {
                rsp.sendError(HttpServletResponse.SC_NOT_FOUND);
                return;
K
kohsuke 已提交
410 411
            }

412
            launch();
K
kohsuke 已提交
413 414 415 416 417 418

            // TODO: would be nice to redirect the user to "launching..." wait page,
            // then spend a few seconds there and poll for the completion periodically.
            rsp.sendRedirect("log");
        }

419 420 421 422 423
        public void launch() {
            if(channel==null)
                launch(getNode());
        }

K
kohsuke 已提交
424 425 426 427 428 429 430 431 432 433 434 435 436 437
        /**
         * Gets the string representation of the slave log.
         */
        public String getLog() throws IOException {
            return Util.loadFile(getLogFile());
        }

        /**
         * Handles incremental log.
         */
        public void doProgressiveLog( StaplerRequest req, StaplerResponse rsp) throws IOException {
            new LargeText(getLogFile(),false).doProgressText(req,rsp);
        }

438 439 440
        /**
         * Serves jar files for JNLP slave agents.
         */
K
kohsuke 已提交
441 442
        public JnlpJar getJnlpJars(String fileName) {
            return new JnlpJar(fileName);
443 444
        }

K
kohsuke 已提交
445 446 447 448 449 450 451 452 453
        @Override
        protected void kill() {
            super.kill();
            closeChannel();
        }

        private void closeChannel() {
            Channel c = channel;
            channel = null;
454
            isUnix=null;
K
kohsuke 已提交
455 456 457 458 459
            if(c!=null)
                try {
                    c.close();
                } catch (IOException e) {
                    logger.log(Level.SEVERE, "Failed to terminate channel to "+getDisplayName(),e);
K
kohsuke 已提交
460
                }
K
kohsuke 已提交
461 462 463 464 465 466 467 468 469 470 471
        }

        @Override
        protected void setNode(Node node) {
            super.setNode(node);
            if(channel==null)
                // maybe the configuration was changed to relaunch the slave, so try it now.
                launch((Slave)node);
        }

        private static final Logger logger = Logger.getLogger(ComputerImpl.class.getName());
K
kohsuke 已提交
472 473 474 475 476 477

        private static final class DetectOS implements Callable<Boolean,IOException> {
            public Boolean call() throws IOException {
                return File.pathSeparatorChar==':';
            }
        }
K
kohsuke 已提交
478 479
    }

480 481 482 483
    /**
     * Web-bound object used to serve jar files for JNLP.
     */
    public static final class JnlpJar {
K
kohsuke 已提交
484
        private final String fileName;
485

K
kohsuke 已提交
486 487
        public JnlpJar(String fileName) {
            this.fileName = fileName;
488 489 490
        }

        public void doIndex( StaplerRequest req, StaplerResponse rsp) throws IOException, ServletException {
K
kohsuke 已提交
491 492 493 494
            URL res = req.getServletContext().getResource("/WEB-INF/" + fileName);
            if(res==null) {
                // during the development this path doesn't have the files.
                res = new URL(new File(".").getAbsoluteFile().toURL(),"target/generated-resources/WEB-INF/"+fileName);
495 496
            }

K
kohsuke 已提交
497
            URLConnection con = res.openConnection();
498 499 500 501 502 503 504
            InputStream in = con.getInputStream();
            rsp.serveFile(req, in, con.getLastModified(), con.getContentLength(), "*.jar" );
            in.close();
        }

    }

K
kohsuke 已提交
505
    public Launcher createLauncher(TaskListener listener) {
506 507
        ComputerImpl c = getComputer();
        return new RemoteLauncher(listener, c.getChannel(), c.isUnix);
K
kohsuke 已提交
508 509
    }

K
kohsuke 已提交
510 511 512
    /**
     * Gets th ecorresponding computer object.
     */
513 514
    public ComputerImpl getComputer() {
        return (ComputerImpl)Hudson.getInstance().getComputer(getNodeName());
K
kohsuke 已提交
515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
    }

    public boolean equals(Object o) {
        if (this == o) return true;
        if (o == null || getClass() != o.getClass()) return false;

        final Slave that = (Slave) o;

        return name.equals(that.name);
    }

    public int hashCode() {
        return name.hashCode();
    }

K
kohsuke 已提交
530 531 532 533 534 535 536 537 538 539 540 541
    /**
     * Invoked by XStream when this object is read into memory.
     */
    private Object readResolve() {
        // convert the old format to the new one
        if(command!=null && agentCommand==null) {
            if(command.length()>0)  command += ' ';
            agentCommand = command+"java -jar ~/bin/slave.jar";
        }
        return this;
    }

K
kohsuke 已提交
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557
    /**
     * This field is used on each slave node to record log records on the slave.
     */
    private static final RingBufferLogHandler SLAVE_LOG_HANDLER = new RingBufferLogHandler();

    private static class LogInstaller implements Callable<Void,RuntimeException> {
        public Void call() {
            // avoid double installation of the handler
            Logger logger = Logger.getLogger("hudson");
            logger.removeHandler(SLAVE_LOG_HANDLER);
            logger.addHandler(SLAVE_LOG_HANDLER);
            return null;
        }
        private static final long serialVersionUID = 1L;
    }

K
kohsuke 已提交
558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576
//
// backwrad compatibility
//
    /**
     * In Hudson < 1.69 this was used to store the local file path
     * to the remote workspace. No longer in use.
     *
     * @deprecated
     *      ... but still in use during the transition.
     */
    private File localFS;

    /**
     * In Hudson < 1.69 this was used to store the command
     * to connect to the remote machine, like "ssh myslave".
     *
     * @deprecated
     */
    private transient String command;
K
kohsuke 已提交
577
}