From 12ff1fd506b1267b21d9924b98868fb96ec40cea Mon Sep 17 00:00:00 2001
From: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
Date: Fri, 16 Apr 2021 09:22:19 +0200
Subject: [PATCH] garlicd: send logs to the builder

---
 garlic/garlicd/garlicd | 41 +++++++++++++++++++++++++++++++++--------
 garlic/pp/store.nix    |  9 +++++++++
 2 files changed, 42 insertions(+), 8 deletions(-)

diff --git a/garlic/garlicd/garlicd b/garlic/garlicd/garlicd
index cc67089..17e3f9d 100755
--- a/garlic/garlicd/garlicd
+++ b/garlic/garlicd/garlicd
@@ -2,9 +2,14 @@
 
 set -e
 
-msg() {
+emsg() {
   >&2 echo "garlicd: $@"
-} 
+}
+
+msg() {
+  emsg "$@"
+  if [ ! -z "$st" ]; then echo "garlicd: $@" >>"$st"; fi
+}
 
 if [ ! -z "$1" ]; then
   >&2 echo "usage: garlicd"
@@ -18,8 +23,8 @@ garlic_sandbox=$(nix show-config |\
   grep -o '/garlic=[^ ]*' || true)
 
 if [ -z "$garlic_sandbox" ]; then
-  msg "Missing extra-sandbox-paths /garlic mountpoint"
-  msg "Check the ~/.config/nix/nix.conf file"
+  emsg "Missing extra-sandbox-paths /garlic mountpoint"
+  emsg "Check the ~/.config/nix/nix.conf file"
   exit 1
 fi
 
@@ -28,6 +33,7 @@ mountdir=$(readlink -f "$mountdir_rel")
 run="$mountdir/run"
 completed="$mountdir/completed"
 wipe="$mountdir/wipe"
+st="$mountdir/st"
 
 handle_bad_signal() {
   msg "cleaning FIFO pipes"
@@ -46,21 +52,40 @@ for fifo in "$run" "$completed" "$wipe"; do
 done
 
 while true; do
-  msg "Waiting for experiments ..."
+  emsg "--- Waiting for experiments ---"
   tre=$(head -1 "$run")
 
+  # Truncate state file
+  printf "" > "$st"
+
   msg "Attempting to run: $tre"
   msg "Copying files to MN4..."
   # It fails if the user doesn't have nix-store, but is already copied
   # with the post build hook
   nix copy --to ssh://mn1 $tre || true
 
+  set +e
+
   msg "Launching the experiment..."
-  garlic -R "$tre"
+  garlic -R "$tre" 2>>"$st"
+  if [ "$?" != 0 ]; then
+    echo ERROR > "$completed"
+    msg "Failed to run the experiment :-("
+    continue
+  fi
 
   msg "Fetching results..."
-  garlic -FKv "$tre"
+  garlic -FKv "$tre" 2>>"$st"
 
+  if [ "$?" != 0 ]; then
+    echo ERROR > "$completed"
+    msg "The experiment failed :-("
+    continue
+  fi
+
+  set -e
+
+  msg "Signalling nix build..."
   echo -n "$tre" >> "$completed"
 
   msg "Waiting for nix to finish the build..."
@@ -72,7 +97,7 @@ while true; do
   fi
 
   msg "Removing temporal files..."
-  garlic -D "$tre"
+  garlic -D "$tre" 2>>"$st"
 
   echo -n "$tre" >> "$completed"
 
diff --git a/garlic/pp/store.nix b/garlic/pp/store.nix
index 40d5591..6005efb 100644
--- a/garlic/pp/store.nix
+++ b/garlic/pp/store.nix
@@ -26,8 +26,17 @@ in
 
         echo ${trebuchet} >> /garlic/run
         echo "resultTree: waiting for experiment results..."
+        stdbuf -o 0 tail -f /garlic/st &
+        stpid=$!
         res=$(cat /garlic/completed)
 
+        kill -TERM $stpid
+
+        if [ "$res" == "ERROR" ]; then
+          echo "resultTree: the experiment failed"
+          exit 1
+        fi
+
         if [ "$res" != "${trebuchet}" ]; then
           echo "resultTree: unknown trebuchet received"
           exit 1