Add more metrics

net/gce.sh

@@ -110,7 +110,7 @@ done
 
 
 prepareInstancesAndWriteConfigFile() {
-  $metricsWriteDatapoint "testnet-deploy net-config-start=1"
+  $metricsWriteDatapoint "testnet-deploy net-config-begin=1"
 
   cat >> "$configFile" <<EOF
 # autogenerated at $(date)
@@ -183,15 +183,15 @@ EOF
 
 case $command in
 delete)
-  $metricsWriteDatapoint "testnet-deploy net-delete-start=1"
+  $metricsWriteDatapoint "testnet-deploy net-delete-begin=1"
 
   gcloud_FindInstances "name~^$prefix-"
 
   if [[ ${#instances[@]} -eq 0 ]]; then
     echo "No instances found matching '^$prefix-'"
-    exit 0
+  else
+    gcloud_DeleteInstances "$yes"
   fi
-  gcloud_DeleteInstances "$yes"
   rm -f "$configFile"
 
   $metricsWriteDatapoint "testnet-deploy net-delete-complete=1"
@@ -200,7 +200,7 @@ delete)
 create)
   [[ -n $validatorNodeCount ]] || usage "Need number of nodes"
 
-  $metricsWriteDatapoint "testnet-deploy net-create=1"
+  $metricsWriteDatapoint "testnet-deploy net-create-begin=1"
 
   echo "Network composition:"
   echo "Leader = $leaderMachineType (GPU=${leaderAccelerator:-none})"

net/net.sh

@@ -188,13 +188,20 @@ startClient() {
 
 sanity() {
   declare expectedNodeCount=$((${#validatorIpList[@]} + 1))
+  declare ok=true
+
   echo "--- Sanity"
+  $metricsWriteDatapoint "testnet-deploy net-sanity-begin=1"
+
   (
     set -x
     # shellcheck disable=SC2029 # remote-client.sh args are expanded on client side intentionally
     ssh "${sshOptions[@]}" "$leaderIp" \
       "./solana/net/remote/remote-sanity.sh $sanityExtraArgs"
-  ) || exit 1
+  ) || ok=false
+
+  $metricsWriteDatapoint "testnet-deploy net-sanity-complete=1"
+  $ok || exit 1
 }
 
 start() {
@@ -231,6 +238,7 @@ start() {
   esac
 
   echo "Deployment started at $(date)"
+  $metricsWriteDatapoint "testnet-deploy net-start-begin=1"
 
   SECONDS=0
   declare leaderDeployTime=
@@ -263,7 +271,7 @@ start() {
     startClient "$ipAddress" "$netLogDir/client-$ipAddress.log"
   done
   clientDeployTime=$SECONDS
-  $metricsWriteDatapoint "testnet-deploy start=1"
+  $metricsWriteDatapoint "testnet-deploy net-start-complete=1"
 
   if [[ $deployMethod = "snap" ]]; then
     IFS=\  read -r _ networkVersion _ < <(
@@ -303,8 +311,7 @@ stop_node() {
 
 stop() {
   SECONDS=0
-
-  $metricsWriteDatapoint "testnet-deploy stop=1"
+  $metricsWriteDatapoint "testnet-deploy net-stop-begin=1"
 
   stop_node "$leaderIp"
 
@@ -312,6 +319,7 @@ stop() {
     stop_node "$ipAddress"
   done
 
+  $metricsWriteDatapoint "testnet-deploy net-stop-complete=1"
   echo "Stopping nodes took $SECONDS seconds"
 }
 

net/remote/remote-client.sh

@@ -54,16 +54,13 @@ esac
 
 scripts/oom-monitor.sh > oom-monitor.log 2>&1 &
 
+set +e
 while true; do
   echo "=== Client start: $(date)" >> client.log
+  $metricsWriteDatapoint "testnet-deploy client-begin=1"
   clientCommand="$solana_bench_tps --num-nodes $numNodes --seconds 600 --sustained --threads $threadCount"
   echo "$ $clientCommand" >> client.log
-
-  set +e
   $clientCommand >> client.log 2>&1
-  set -e
-
-  $metricsWriteDatapoint "testnet-deploy,name=$netBasename clientexit=1"
-  echo Error: bench-tps should never exit | tee -a client.log
+  $metricsWriteDatapoint "testnet-deploy client-complete=1"
 done