Changeset 5991

Timestamp:

Aug 20, 2020, 9:58:52 AM (3 years ago)

Author:

Nicklas Nordborg

Message:

References #1259: Add support for Slurm

Implemented a simple solution for translating options between Open Grid and Slurm. So far, the implemented translation is between 'pe' and 'cpus-per-task' since that is the only option we use in Reggie.

Location:

extensions/net.sf.basedb.opengrid/trunk/src/net/sf/basedb/opengrid

Files:

• config/JobConfig.java (modified) (6 diffs)
• engine/SlurmEngine.java (modified) (2 diffs)

extensions/net.sf.basedb.opengrid/trunk/src/net/sf/basedb/opengrid/config/JobConfig.java

@@ -5 +5 @@
 import java.util.Map;
 import java.util.Set;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 
 import net.sf.basedb.opengrid.JobDefinition;
@@ -10 +12 @@
 /**
   Configuration settings related to a job definition that is
-  about to be run on an Open Grid cluster. The information becomes 
-  readonly when a {@link JobDefinition} instance has been created.
+  about to be run on an Open Grid or Slurm cluster. The information 
+  becomes readonly when a {@link JobDefinition} instance has been created.
   A single configuration instance can be used with multiple jobs.
+  <p>
+  Note that options for the Open Grid Engine and Slurm can be very
+  different and it is best if the client uses either {@link #setQsubOption(String, String)}
+  or {@link #setSbatchOption(String, String)} depending on the type of the 
+  cluster.
+  <p>
+  Some options can be automatically converted between the two systems. This will 
+  only happen if options for one system has been set, but no options for the other.
+  It is also possible to force conversion by directly calling
+  {@link #convertQsubToSbatchOptions()} or {@link #convertSbatchToQsubOptions()}.
   
   @author nicklas
@@ -93 +105 @@
   private Map<String, String> qsubOptions;
   private Map<String, String> sbatchOptions;
+  private boolean autoConvertOptions;
   
   /**
@@ -98 +111 @@
   */
   public JobConfig()
+  {
+    this(true);
+  }
+  
+  /**
+    Create a new job configuration instance.
+    @param autoConvertOptions If TRUE, options for qsub or sbatch are automatically
+      converted
+    @since 1.4
+  */
+  public JobConfig(boolean autoConvertOptions)
   {
     this.failImmediately = true;
@@ -105 +129 @@
     this.qsubOptions = new LinkedHashMap<>();
     this.sbatchOptions = new LinkedHashMap<>();
+    this.autoConvertOptions = autoConvertOptions;
   }
 
@@ -342 +367 @@
       throw new IllegalArgumentException("Priority must be in the range -1023..+1024: " + priority);
     }
-    if (nice != null && (nice < 2147483645 || nice > 2147483645))
+    if (nice != null && (nice < -2147483645 || nice > 2147483645))
     {
       throw new IllegalArgumentException("Nice must be in the range +/- 2147483645: " + nice);
     }
-  }
-
+    if (forLock && autoConvertOptions) autoConvertOptions();
+  }
+
+  private void autoConvertOptions()
+  {
+    if (sbatchOptions.size() == 0 && qsubOptions.size() > 0)
+    {
+      convertQsubToSbatchOptions();
+    }
+    else if (sbatchOptions.size() > 0 && qsubOptions.size() == 0)
+    {
+      convertSbatchToQsubOptions();
+    }
+  }
+  
+  /**
+    Convert options set for 'qsub' (Open Grid) to options for 'sbatch' (Slurm).
+    Supported options are:
+    
+    * -pe smp X-Y: --nodes=1, --ntasks=1, --cpus-per-task=max(X, Y)
+
+    @since 1.4 
+  */
+  public void convertQsubToSbatchOptions()
+  {
+    checkLocked("convertQsubToSbatchOptions()");
+    for (Map.Entry<String, String> entry : qsubOptions.entrySet())
+    {
+      String option = entry.getKey();
+      if ("pe".equals(option))
+      {
+        sbatchOptions.put("nodes", "1");
+        sbatchOptions.put("ntasks", "1");
+        sbatchOptions.put("cpus-per-task", getMaxNumberInString(entry.getValue()));
+      }
+    }
+  }
+  
+  /**
+    Convert options set for 'sbatch' (Slurm) to options for 'qsub' (Open Grid).
+    Supported options are:
+    
+    * --cpus-per-task=N: -pe smp N
+  
+    @since 1.4 
+  */
+  public void convertSbatchToQsubOptions()
+  {
+    checkLocked("convertSbatchToQsubOptions()");
+    for (Map.Entry<String, String> entry : sbatchOptions.entrySet())
+    {
+      String option = entry.getKey();
+      if ("cpus-per-task".equals(option) || "c".equals(option))
+      {
+        // '--cpu-per-task=N' is converted to '-pe smp N'
+        qsubOptions.put("pe", "smp " + entry.getValue());
+      }
+    }
+  }
+  
+  /**
+    Get the highest number that can be found in the given string.
+    Use to extract the number of requested slots in, for example,
+    -pe smp 8-16.
+  */
+  private String getMaxNumberInString(String s)
+  {
+    Pattern p = Pattern.compile("\\d+");
+    Matcher m = p.matcher(s);
+    int max = 1;
+    while (m.find())
+    {
+      int val = Integer.valueOf(m.group());
+      if (val > max) max = val;
+    }
+    return Integer.toString(max);
+  }
   
   /**

extensions/net.sf.basedb.opengrid/trunk/src/net/sf/basedb/opengrid/engine/SlurmEngine.java

@@ -83 +83 @@
     script.append("#SBATCH --output=stdout\n"); // Stdout is saved to this file
     script.append("#SBATCH --error=stderr\n"); // Stderr is saved to this file
-//    script.append("#SBATCH --nodes=1\n"); // 1 task on 1 node
-//    script.append("#SBATCH --ntasks=1\n"); 
-//    script.append("#SBATCH --cpus-per-task=2\n"); 
     
     if (config.getSlurmNice() != null)
@@ -119 +116 @@
     if (config.getFailImmediately()) script.append("set -e\n");
     if (config.getCreatePrivateFiles()) script.append("umask -S u=rwx,g=,o=\n");
-    
-    /*
-    // TODO -- remove before release
-    ScriptBuilder debug = new ScriptBuilder();
-    debug.progress(10, "Starting up...");
-    debug.cmd("date");
-    debug.cmd("sleep 60s");
-    debug.progress(20, "Analyzing...");
-    debug.cmd("nproc");
-    debug.cmd("sleep 60s");
-    debug.progress(80, "Copying result files...");
-    debug.cmd("echo fakefile > ${WD}/files.out");
-    debug.cmd("sleep 60s");
-    debug.progress(90, "Saving to database...");
-    debug.cmd("sleep 60s");
-    debug.newLine();
-    
-    script.append("# -- for debugging\n");
-    script.append(debug.toString());
-    script.append("exit\n");
-    script.append("# -- end debugging\n");
-    // -- 
-    */
-    
+
     script.append("# --- user script ---\n");
     script.append(job.getCmd()+"\n");