WIP: Testing with automatic fetching

garlic/exp/nbody/granularity-mpi.nix

@@ -25,7 +25,7 @@ let
     gitBranch = "garlic/mpi+send";
 
     # Repeat the execution of each unit 30 times
-    loops = 30;
+    loops = 10;
 
     # Resources
     qos = "debug";
@@ -34,6 +34,10 @@ let
     time = "02:00:00";
     cpuBind = "sockets,verbose";
     jobName = "nbody-bs-${toString blocksize}-${gitBranch}";
+
+    # Experiment revision: this allows a user to run again a experiment already
+    # executed
+    rev = 0;
   };
 
   # Compute the array of configurations
@@ -47,10 +51,11 @@ let
   };
 
   program = {nextStage, conf, ...}: with conf;
-  let
-    customPkgs = stdexp.replaceMpi conf.mpi;
-  in
-    customPkgs.apps.nbody.override {
+  # FIXME: This is becoming very slow:
+  #let
+  #  customPkgs = stdexp.replaceMpi conf.mpi;
+  #in
+  bsc.garlic.apps.nbody.override {
     inherit cc blocksize mpi gitBranch;
   };
 

garlic/pp/check.nix

@@ -0,0 +1,33 @@
+{
+  stdenv
+}:
+
+resultTree:
+
+stdenv.mkDerivation {
+  name = "check";
+  preferLocalBuild = true;
+  phases = [ "installPhase" ];
+  installPhase = ''
+    echo "checking result tree: ${resultTree}"
+    cd ${resultTree}
+    for exp in *-experiment; do
+      cd ${resultTree}/$exp
+      echo "$exp: checking units"
+      for unit in *-unit; do
+        cd ${resultTree}/$exp/$unit
+        if [ ! -e status ]; then
+          echo "missing $unit/status file, aborting"
+          exit 1
+        fi
+        st=$(cat status)
+        if [ "$st" != "completed" ]; then
+          echo "unit $unit is not complete yet, aborting"
+          exit 1
+        fi
+      done
+      echo "$exp: execution complete"
+    done
+    ln -s $out ${resultTree}
+  '';
+}

garlic/pp/fetch.nix

@@ -31,14 +31,17 @@ in
     name = "fetch";
     preferLocalBuild = true;
 
-    buildInputs = [ rsync openssh curl ];
+    buildInputs = [ rsync openssh curl nix ];
     phases = [ "installPhase" ];
+    # This doesn't work when multiple users have different directories where the
+    # results are stored.
+    #src = /. + "${prefix}${experimentName}";
 
     installPhase = ''
       cat > $out << EOF
       #!/bin/sh -e
       mkdir -p ${garlicTemp}
-      export PATH=${rsync}/bin:${openssh}/bin:${nix}/bin
+      export PATH=$PATH
       rsync -av \
         --copy-links \
         ${rsyncFilter} \
@@ -50,8 +53,11 @@ in
         garlicTemp = "${garlicTemp}"; \
       })')
 
+      rm -rf ${garlicTemp}/${experimentName}
+
       echo "The results for experiment ${experimentName} are at:"
       echo "  \$res"
+
       EOF
       chmod +x $out
     '';

garlic/pp/result2.nix

@@ -0,0 +1,81 @@
+{
+  stdenv
+, garlicTools
+}:
+
+{
+  trebuchetStage
+, experimentStage
+, garlicTemp
+}:
+
+with garlicTools;
+
+let
+  experimentName = baseNameOf (toString experimentStage);
+  garlicOut = "/mnt/garlic-out";
+in
+  stdenv.mkDerivation {
+    name = "result";
+    preferLocalBuild = true;
+    __noChroot = true;
+
+    phases = [ "installPhase" ];
+
+    installPhase = ''
+      expList=$(find ${garlicOut} -maxdepth 2 -name ${experimentName})
+
+      if [ -z "$expList" ]; then
+        echo "ERROR: missing results for ${experimentName}"
+        echo "Execute it by running:"
+        echo
+        echo -e "  \e[30;48;5;2m${trebuchetStage}\e[0m"
+        echo
+        echo "cannot continue building $out, aborting"
+        exit 1
+      fi
+
+      N=$(echo $expList | wc -l)
+      echo "Found $N results: $expList"
+
+      if [ $N -gt 1 ]; then
+        echo 
+        echo "ERROR: multiple results for ${experimentName}:"
+        echo "$expList"
+        echo
+        echo "cannot continue building $out, aborting"
+        exit 1
+      fi
+
+      exp=$expList
+      repeat=1
+      while [ 1 ]; do
+        repeat=0
+        cd $exp
+        echo "$exp: checking units"
+        for unit in *-unit; do
+          cd $exp/$unit
+          if [ ! -e status ]; then
+            echo "$unit: no status"
+            repeat=1
+          else
+            st=$(cat status)
+            echo "$unit: $st"
+            if [ "$st" != "completed" ]; then
+              repeat=1
+            fi
+          fi
+        done
+
+        if [ $repeat -eq 0 ]; then
+          break
+        fi
+        echo "waiting 10 seconds to try again"
+        sleep 10
+      done
+      echo "$exp: execution complete"
+
+      mkdir -p $out
+      cp -aL $exp $out
+    '';
+  }

garlic/stages/experiment.nix

@@ -38,7 +38,7 @@ stdenv.mkDerivation {
     export GARLIC_EXPERIMENT=$(basename $out)
 
     if [ -e "\$GARLIC_EXPERIMENT" ]; then
-      >&2 echo "skipping, experiment path already exists: \$GARLIC_EXPERIMENT"
+      >&2 echo "skipping, path exists: \$(pwd)/\$GARLIC_EXPERIMENT"
       exit 0
     fi
 

garlic/stages/runexp/runexp

@@ -5,13 +5,13 @@ if [ -e /nix ]; then
     exit 1
 fi
 
->&2 echo Running runexp for MN4
->&2 echo PATH=$PATH
+#>&2 echo Running runexp for MN4
+#>&2 echo PATH=$PATH
 
 user=$(id -un)
 group=$(id -gn)
 
-export GARLIC_OUT="/gpfs/projects/$group/$user/garlic-out"
+export GARLIC_OUT="/gpfs/projects/bsc15/garlic/out/$user"
 mkdir -p "$GARLIC_OUT"
 cd "$GARLIC_OUT"
 

overlay.nix

@@ -314,15 +314,14 @@ let
 
       # Post processing tools
       pp = rec {
-        getExpResult = callPackage ./garlic/pp/result.nix {
-          inherit fetchExperiment;
+        getExpResult = callPackage ./garlic/pp/result2.nix {
         };
-        resultFromTrebuchet = trebuchetStage: getExpResult {
+        resultFromTrebuchet = trebuchetStage: (getExpResult {
           garlicTemp = "/tmp/garlic-temp";
           inherit trebuchetStage;
           experimentStage = with self.bsc.garlicTools;
             getExperimentStage trebuchetStage;
-        };
+        });
         fetchExperiment = callPackage ./garlic/pp/fetch.nix { };
         timetable = callPackage ./garlic/pp/timetable.nix { };
         rPlot = callPackage ./garlic/pp/rplot.nix { };
@@ -337,11 +336,16 @@ let
       # Datasets used in the figures
       ds = with self.bsc.garlic; with pp; {
         nbody = with exp.nbody; {
-          test      = merge [ baseline ];
+          test      = merge [ test ];
+          baseline  = merge [ baseline ];
+          jemalloc  = merge [ jemalloc ];
+          freeCpu   = merge [ freeCpu ];
+          cmp = {
             jemalloc = merge [ baseline jemalloc ];
             freeCpu  = merge [ baseline freeCpu ];
           };
         };
+      };
 
       # Figures generated from the experiments
       fig = with self.bsc.garlic; {