Sbatch options

.gitignore

@@ -1,7 +1,7 @@
 # Maven projects
 /computation-mpi/target/
 /computation-slurm/target/
-/computation-slurm-tests/target/
+/computation-slurm-test/target/
 /distribution-hpc/target/
 /target
 

computation-slurm-test/src/test/resources/myapps/myecho.sh

@@ -1,2 +1,6 @@
 #!/bin/sh
-echo $1 > $2
+if [ $1 = "oddIn3" ]; then
+  echoooo
+else
+  echo $1 > $2
+fi

computation-slurm/src/main/java/com/powsybl/computation/slurm/CommandUtils.java

@@ -35,7 +35,16 @@ static String commandToString(String program, Collection<String> args) {
         requireNonNull(program);
         requireNonNull(args);
 
-        String argStr = args.stream().collect(getWrapperAndJoiner());
-        return program + " " + argStr;
+        return program + " " + commandArgsToString(args);
+    }
+
+    /**
+     * Generates a command's argu string, with each argument wrapped with quotes.
+     * @param args
+     * @return the argu's string
+     */
+    static String commandArgsToString(Collection<String> args) {
+        requireNonNull(args);
+        return args.stream().collect(getWrapperAndJoiner());
     }
 }

computation-slurm/src/main/java/com/powsybl/computation/slurm/FlagFilesMonitor.java

@@ -53,7 +53,7 @@ public void run() {
                         commandRunner.execute("rm " + flagDir + "/" + line);
                         // cancel following jobs(which depends on this job) if there are errors
                         if (line.startsWith("myerror_")) {
-                            taskStore.getCompletableFuture(workingDirName).cancel(true);
+                            taskStore.getCompletableFuture(workingDirName).cancelBySlurm(new SlurmException("An error job found"));
                         } else if (line.startsWith("mydone_")) {
                             String id = line.substring(lastIdx + 1);
                             taskStore.untracing(Long.parseLong(id));

computation-slurm/src/main/java/com/powsybl/computation/slurm/SbatchCmdBuilder.java → computation-slurm/src/main/java/com/powsybl/computation/slurm/SbatchArguments.java

@@ -6,129 +6,142 @@
  */
 package com.powsybl.computation.slurm;
 
-import com.google.common.base.Preconditions;
-
 import javax.annotation.Nullable;
 import java.nio.file.Path;
 import java.util.*;
 import java.util.stream.Collectors;
 
 /**
- * Builds an {@link SbatchCmd}, used to submit script execution to Slurm.
+ * This class contains two map and one set for three types of arguments in sbatch.
  * @see <a href="https://slurm.schedmd.com/sbatch.html">Sbatch</a>
  *
  * @author Yichen Tang <yichen.tang at rte-france.com>
  */
-class SbatchCmdBuilder {
+class SbatchArguments {
 
-    private static final String DATETIME_FORMATTER = "`date -d \"%d seconds\" \"+%%Y-%%m-%%dT%%H:%%M:%%S\"`";
+    private static final String PRECEDENT = "#SBATCH ";
+    private static final String DENPENDENCY_OPT = "dependency";
 
     private final Map<String, String> sbatchArgsByName = new HashMap<>();
     private final Map<Character, String> sbatchArgsByCharacter = new HashMap<>();
     private final TreeSet<String> sbatchOptions = new TreeSet<>();
-    private String script;
 
-    SbatchCmdBuilder jobName(String jobName) {
+    SbatchArguments() {
+        killOnInvalidDep();
+    }
+
+    SbatchArguments jobName(String jobName) {
         sbatchArgsByName.put("job-name", jobName);
         return this;
     }
 
-    SbatchCmdBuilder array(int i) {
-        if (i < 0) {
+    /**
+     *
+     * @param i if equals 1, --array would not be set. If negative, exception would be thrown.
+     * @return this builder
+     */
+    SbatchArguments array(int i) {
+        if (i <= 0) {
             throw new IllegalArgumentException(i + " is not validate for array.");
         }
-        if (i == 1) {
-            sbatchArgsByName.put("array", Integer.toString(0));
-        }
-        if (i != 1) {
+        if (i > 1) {
             sbatchArgsByName.put("array", "0-" + (i - 1));
         }
         return this;
     }
 
-    SbatchCmdBuilder aftercorr(List<Long> jobIds) {
+    SbatchArguments aftercorr(List<Long> jobIds) {
         Objects.requireNonNull(jobIds);
         if (!jobIds.isEmpty()) {
             String coll = jobIds.stream().map(Object::toString).collect(Collectors.joining(":", "aftercorr:", ""));
-            sbatchArgsByName.put("dependency", coll);
+            sbatchArgsByName.put(DENPENDENCY_OPT, coll);
         }
         return this;
     }
 
-    SbatchCmdBuilder nodes(int i) {
+    SbatchArguments aftercorr(@Nullable Long preMasterJob) {
+        if (preMasterJob != null) {
+            sbatchArgsByName.put(DENPENDENCY_OPT, "aftercorr:" + preMasterJob);
+        }
+        return this;
+    }
+
+    SbatchArguments afternotok(Long lastMasterJob) {
+        if (lastMasterJob != null) {
+            sbatchArgsByName.put(DENPENDENCY_OPT, "afternotok:" + lastMasterJob);
+        }
+        return this;
+    }
+
+    SbatchArguments nodes(int i) {
         if (i < 1) {
             throw new IllegalArgumentException(i + " is not validate for nodes.");
         }
         sbatchArgsByName.put("nodes", Integer.toString(i));
         return this;
     }
 
-    SbatchCmdBuilder ntasks(int i) {
+    SbatchArguments ntasks(int i) {
         if (i < 1) {
             throw new IllegalArgumentException(i + " is not validate for ntasks.");
         }
         sbatchArgsByName.put("ntasks", Integer.toString(i));
         return this;
     }
 
-    SbatchCmdBuilder tasksPerNode(int i) {
+    SbatchArguments tasksPerNode(int i) {
         if (i < 1) {
             throw new IllegalArgumentException(i + " is not validate for tasksPerNode.");
         }
         sbatchArgsByName.put("ntasks-per-node", Integer.toString(i));
         return this;
     }
 
-    SbatchCmdBuilder cpusPerTask(int i) {
+    SbatchArguments cpusPerTask(int i) {
         sbatchArgsByName.put("cpus-per-task", Integer.toString(i));
         return this;
     }
 
-    SbatchCmdBuilder error(String pattern) {
+    // TODO test
+    SbatchArguments error(String pattern) {
         Objects.requireNonNull(pattern);
         sbatchArgsByName.put("error", pattern);
         return this;
     }
 
-    SbatchCmdBuilder output(String pattern) {
+    SbatchArguments output(String pattern) {
         Objects.requireNonNull(pattern);
         sbatchArgsByName.put("output", pattern);
         return this;
     }
 
-    SbatchCmdBuilder partition(String partition) {
+    SbatchArguments partition(String partition) {
         Objects.requireNonNull(partition);
         sbatchArgsByName.put("partition", partition);
         return this;
     }
 
-    SbatchCmdBuilder oversubscribe() {
+    SbatchArguments oversubscribe() {
         sbatchOptions.add("oversubscribe");
         return this;
     }
 
-    SbatchCmdBuilder script(String name) {
-        this.script = Objects.requireNonNull(name);
+    private SbatchArguments killOnInvalidDep() {
+        sbatchArgsByName.put("kill-on-invalid-dep", "yes");
         return this;
     }
 
-    SbatchCmdBuilder workDir(Path dir) {
+    SbatchArguments workDir(Path dir) {
         Objects.requireNonNull(dir);
         sbatchArgsByCharacter.put('D', dir.toAbsolutePath().toString());
         return this;
     }
 
-    SbatchCmdBuilder timeout(@Nullable String duration) {
+    SbatchArguments timeout(@Nullable String duration) {
         sbatchArgsByName.put("time", checkTimeout(duration));
         return this;
     }
 
-    SbatchCmdBuilder deadline(long seconds) {
-        Preconditions.checkArgument(seconds > 0, "Invalid seconds({}) for deadline: must be 1 or greater", seconds);
-        sbatchArgsByName.put("deadline", String.format(DATETIME_FORMATTER, seconds));
-        return this;
-    }
-
     private static String checkTimeout(@Nullable String duration) {
         if (duration == null) {
             return "UNLIMITED";
@@ -137,20 +150,18 @@ private static String checkTimeout(@Nullable String duration) {
         return duration;
     }
 
-    SbatchCmdBuilder qos(String qos) {
+    SbatchArguments qos(String qos) {
         Objects.requireNonNull(qos);
         sbatchArgsByName.put("qos", qos);
         return this;
     }
 
-    SbatchCmd build() {
-        validate();
-        return new SbatchCmd(sbatchArgsByName, sbatchArgsByCharacter, sbatchOptions, script);
+    List<String> toScript() {
+        List<String> list = new ArrayList<>();
+        sbatchArgsByCharacter.forEach((k, v) -> list.add(PRECEDENT + "-" + k + " " + v));
+        sbatchArgsByName.forEach((k, v) -> list.add(PRECEDENT + "--" + k + "=" + v));
+        sbatchOptions.forEach(o -> list.add(PRECEDENT + "--" + o));
+        return list;
     }
 
-    private void validate() {
-        if (null == script) {
-            throw new SlurmException("Script is null in cmd");
-        }
-    }
 }

computation-slurm/src/main/java/com/powsybl/computation/slurm/SbatchCmd.java

@@ -6,35 +6,34 @@
  */
 package com.powsybl.computation.slurm;
 
-import java.util.Map;
+import com.google.common.base.Preconditions;
+
 import java.util.Objects;
-import java.util.TreeSet;
 
 /**
- * A submission command to Slurm: sbatch options scriptName.
+ * A submission command to Slurm: sbatch [--deadline] scriptName.
  *
  * Sbatch scripts will be submitted using that kind of command.
  *
  * @author Yichen Tang <yichen.tang at rte-france.com>
  */
 class SbatchCmd extends AbstractSlurmCmd<SbatchCmdResult> {
 
-    private static final String SBATCH = "sbatch";
-
-    private final Map<String, String> argsByName;
-    private final Map<Character, String> argsByChar;
-
-    private final TreeSet<String> options;
+    private static final String SBATCH = "sbatch ";
+    private static final String DATETIME_FORMATTER = "--deadline=`date -d \"%d seconds\" \"+%%Y-%%m-%%dT%%H:%%M:%%S\"` ";
 
     private final String scriptName;
 
-    private String cmd;
+    private long deadLine = 0;
 
-    SbatchCmd(Map<String, String> argsByName, Map<Character, String> argsByChar, TreeSet<String> options, String scriptName) {
-        this.argsByName = Objects.requireNonNull(argsByName);
-        this.argsByChar = Objects.requireNonNull(argsByChar);
+    SbatchCmd(String scriptName) {
         this.scriptName = Objects.requireNonNull(scriptName);
-        this.options = Objects.requireNonNull(options);
+    }
+
+    SbatchCmd deadLine(long seconds) {
+        Preconditions.checkArgument(seconds > 0, "Invalid seconds({}) for deadline: must be 1 or greater", seconds);
+        deadLine = seconds;
+        return this;
     }
 
     SbatchCmdResult send(CommandExecutor commandExecutor) throws SlurmCmdNonZeroException {
@@ -44,24 +43,10 @@ SbatchCmdResult send(CommandExecutor commandExecutor) throws SlurmCmdNonZeroExce
 
     @Override
     public String toString() {
-        if (cmd == null) {
-            StringBuilder sb = new StringBuilder();
-            sb.append(SBATCH);
-            argsByChar.forEach((c, v) -> {
-                sb.append(" -").append(c);
-                sb.append(" ").append(v);
-            });
-            argsByName.forEach((k, v) -> {
-                sb.append(" --").append(k);
-                sb.append("=").append(v);
-            });
-            options.forEach(opt -> {
-                sb.append(" --");
-                sb.append(opt);
-            });
-            sb.append(" ").append(scriptName);
-            cmd = sb.toString();
+        StringBuilder sb = new StringBuilder().append(SBATCH);
+        if (deadLine > 0) {
+            sb.append(String.format(DATETIME_FORMATTER, deadLine));
         }
-        return cmd;
+        return sb.append(scriptName).toString();
     }
 }