Add AWS-based nets

ci/testnet-deploy.sh

@@ -21,12 +21,13 @@ usage() {
     echo "Error: $*"
   fi
   cat <<EOF
-usage: $0 [name] [zone] [options...]
+usage: $0 [name] [cloud] [zone] [options...]
 
 Deploys a CD testnet
 
   name  - name of the network
-  zone  - zone to deploy the network into
+  cloud - cloud provider to use (gce, ec2)
+  zone  - cloud provider zone to deploy the network into
 
   options:
    -s edge|beta|stable  - Deploy the specified Snap release channel
@@ -48,10 +49,12 @@ EOF
 }
 
 netName=$1
-zone=$2
+cloudProvider=$2
+zone=$3
 [[ -n $netName ]] || usage
+[[ -n $cloudProvider ]] || usage "Cloud provider not specified"
 [[ -n $zone ]] || usage "Zone not specified"
-shift 2
+shift 3
 
 while getopts "h?p:Pn:c:s:t:gG:a:d" opt; do
   case $opt in
@@ -108,7 +111,7 @@ while getopts "h?p:Pn:c:s:t:gG:a:d" opt; do
 done
 
 
-gce_create_args=(
+create_args=(
   -a "$leaderAddress"
   -c "$clientNodeCount"
   -n "$validatorNodeCount"
@@ -118,26 +121,26 @@ gce_create_args=(
 
 if $enableGpu; then
   if [[ -z $leaderMachineType ]]; then
-    gce_create_args+=(-g)
+    create_args+=(-g)
   else
-    gce_create_args+=(-G "$leaderMachineType")
+    create_args+=(-G "$leaderMachineType")
   fi
 fi
 
 if $publicNetwork; then
-  gce_create_args+=(-P)
+  create_args+=(-P)
 fi
 
 set -x
 
-echo --- gce.sh delete
-time net/gce.sh delete -z "$zone" -p "$netName"
+echo "--- $cloudProvider.sh delete"
+time net/"$cloudProvider".sh delete -z "$zone" -p "$netName"
 if $delete; then
   exit 0
 fi
 
-echo --- gce.sh create
-time net/gce.sh create "${gce_create_args[@]}"
+echo "--- $cloudProvider.sh create"
+time net/"$cloudProvider".sh create "${create_args[@]}"
 net/init-metrics.sh -e
 
 echo --- net.sh start

ci/testnet-manager.sh

@@ -8,7 +8,7 @@ if [[ -z $BUILDKITE ]]; then
 fi
 
 if [[ -z $SOLANA_METRICS_PARTIAL_CONFIG ]]; then
-  echo SOLANA_METRICS_CONFIG not defined
+  echo SOLANA_METRICS_PARTIAL_CONFIG not defined
   exit 1
 fi
 
@@ -37,6 +37,14 @@ steps:
             value: "testnet-master"
           - label: "testnet-master-perf"
             value: "testnet-master-perf"
+          - label: "testnet-edge"
+            value: "testnet-edge"
+          - label: "testnet-edge-perf"
+            value: "testnet-edge-perf"
+          - label: "testnet-beta"
+            value: "testnet-beta"
+          - label: "testnet-beta-perf"
+            value: "testnet-beta-perf"
       - select: "Operation"
         key: "testnet-operation"
         default: "sanity-or-restart"
@@ -64,10 +72,14 @@ ci/channel-info.sh
 eval "$(ci/channel-info.sh)"
 
 case $TESTNET in
-testnet-master|testnet-master-perf)
+testnet-edge|testnet-edge-perf|testnet-master|testnet-master-perf)
   CHANNEL_OR_TAG=edge
   CHANNEL_BRANCH=$EDGE_CHANNEL
   ;;
+testnet-beta|testnet-beta-perf)
+  CHANNEL_OR_TAG=beta
+  CHANNEL_BRANCH=$BETA_CHANNEL
+  ;;
 testnet|testnet-perf)
   if [[ -n $BETA_CHANNEL_LATEST_TAG ]]; then
     CHANNEL_OR_TAG=$BETA_CHANNEL_LATEST_TAG
@@ -103,20 +115,67 @@ fi
 sanity() {
   echo "--- sanity $TESTNET"
   case $TESTNET in
-  testnet-master-perf)
+  testnet-edge)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
     (
       set -ex
-      ci/testnet-sanity.sh master-perf-testnet-solana-com us-west1-b
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-sanity.sh edge-testnet-solana-com ec2 us-east-1b
     )
     ;;
-  testnet-master)
+  testnet-edge-perf)
     # shellcheck disable=2030
+    # shellcheck disable=2031
     (
       set -ex
       export REJECT_EXTRA_NODES=1
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-sanity.sh master-testnet-solana-com us-west1-b
+      ci/testnet-sanity.sh edge-perf-testnet-solana-com ec2 us-east-1a
+    )
+    ;;
+  testnet-beta)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-sanity.sh beta-testnet-solana-com ec2 us-west-1a
+    )
+    ;;
+  testnet-beta-perf)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export REJECT_EXTRA_NODES=1
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-sanity.sh beta-perf-testnet-solana-com ec2 us-west-2b
+    )
+    ;;
+  testnet-master)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-sanity.sh master-testnet-solana-com gce us-west1-b
+    )
+    ;;
+  testnet-master-perf)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export REJECT_EXTRA_NODES=1
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-sanity.sh master-perf-testnet-solana-com gce us-west1-b
     )
     ;;
   testnet)
@@ -126,7 +185,7 @@ sanity() {
       set -ex
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-sanity.sh testnet-solana-com us-east1-c
+      ci/testnet-sanity.sh testnet-solana-com gce us-east1-c
     )
     ;;
   testnet-perf)
@@ -137,7 +196,7 @@ sanity() {
       export REJECT_EXTRA_NODES=1
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-sanity.sh perf-testnet-solana-com us-west1-b
+      ci/testnet-sanity.sh perf-testnet-solana-com gce us-west1-b
     )
     ;;
   *)
@@ -157,12 +216,51 @@ start() {
   fi
 
   case $TESTNET in
-  testnet-master-perf)
+  testnet-edge)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
     (
       set -ex
-      ci/testnet-deploy.sh master-perf-testnet-solana-com us-west1-b \
-        -G "n1-standard-16 --accelerator count=2,type=nvidia-tesla-v100" \
-        -t "$CHANNEL_OR_TAG" -c 2 \
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-deploy.sh edge-testnet-solana-com ec2 us-east-1b \
+        -s "$CHANNEL_OR_TAG" -n 3 -c 0 -P \
+        ${maybeDelete:+-d}
+    )
+    ;;
+  testnet-edge-perf)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-deploy.sh edge-perf-testnet-solana-com ec2 us-east-1a \
+        -g -t "$CHANNEL_OR_TAG" -c 2 \
+        ${maybeDelete:+-d}
+    )
+    ;;
+  testnet-beta)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-deploy.sh edge-testnet-solana-com ec2 us-west-1a \
+        -s "$CHANNEL_OR_TAG" -n 3 -c 0 -P \
+        ${maybeDelete:+-d}
+    )
+    ;;
+  testnet-beta-perf)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-deploy.sh beta-perf-testnet-solana-com ec2 us-west-2b \
+        -g -t "$CHANNEL_OR_TAG" -c 2 \
         ${maybeDelete:+-d}
     )
     ;;
@@ -173,11 +271,24 @@ start() {
       set -ex
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-deploy.sh master-testnet-solana-com us-west1-b \
+      ci/testnet-deploy.sh master-testnet-solana-com gce us-west1-b \
         -s "$CHANNEL_OR_TAG" -n 3 -c 0 -P -a master-testnet-solana-com \
         ${maybeDelete:+-d}
     )
     ;;
+  testnet-master-perf)
+    # shellcheck disable=2030
+    # shellcheck disable=2031
+    (
+      set -ex
+      export NO_LEDGER_VERIFY=1
+      export NO_VALIDATOR_SANITY=1
+      ci/testnet-deploy.sh master-perf-testnet-solana-com gce us-west1-b \
+        -G "n1-standard-16 --accelerator count=2,type=nvidia-tesla-v100" \
+        -t "$CHANNEL_OR_TAG" -c 2 \
+        ${maybeDelete:+-d}
+    )
+    ;;
   testnet)
     # shellcheck disable=2030
     # shellcheck disable=2031
@@ -185,7 +296,7 @@ start() {
       set -ex
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-deploy.sh testnet-solana-com us-east1-c \
+      ci/testnet-deploy.sh testnet-solana-com gce us-east1-c \
         -s "$CHANNEL_OR_TAG" -n 3 -g -c 0 -P -a testnet-solana-com  \
         ${maybeDelete:+-d}
     )
@@ -197,7 +308,7 @@ start() {
       set -ex
       export NO_LEDGER_VERIFY=1
       export NO_VALIDATOR_SANITY=1
-      ci/testnet-deploy.sh perf-testnet-solana-com us-west1-b \
+      ci/testnet-deploy.sh perf-testnet-solana-com gce us-west1-b \
         -G "n1-standard-16 --accelerator count=2,type=nvidia-tesla-v100" \
         -t "$CHANNEL_OR_TAG" -c 2 \
         ${maybeDelete:+-d}

ci/testnet-sanity.sh

@@ -9,12 +9,13 @@ usage() {
     echo "Error: $*"
   fi
   cat <<EOF
-usage: $0 [name] [zone]
+usage: $0 [name] [cloud] [zone]
 
 Sanity check a CD testnet
 
   name  - name of the network
-  zone  - zone of the network
+  cloud - cloud provider to use (gce, ec2)
+  zone  - cloud provider zone of the network
 
   Note: the SOLANA_METRICS_CONFIG environment variable is used to configure
         metrics
@@ -23,13 +24,15 @@ EOF
 }
 
 netName=$1
-zone=$2
+cloudProvider=$2
+zone=$3
 [[ -n $netName ]] || usage ""
+[[ -n $cloudProvider ]] || usage "Cloud provider not specified"
 [[ -n $zone ]] || usage "Zone not specified"
 
 set -x
-echo --- gce.sh config
-net/gce.sh config -p "$netName" -z "$zone"
+echo "--- $cloudProvider.sh config"
+net/"$cloudProvider".sh config -p "$netName" -z "$zone"
 net/init-metrics.sh -e
 echo --- net.sh sanity
 net/net.sh sanity \

net/gce.sh

@@ -36,8 +36,8 @@ ec2)
   cpuLeaderMachineType=m4.4xlarge
   gpuLeaderMachineType=p2.xlarge
   leaderMachineType=$cpuLeaderMachineType
-  validatorMachineType=m4.4xlarge
-  clientMachineType=m4.4xlarge
+  validatorMachineType=m4.2xlarge
+  clientMachineType=m4.2xlarge
   ;;
 *)
   echo "Error: Unknown cloud provider: $cloudProvider"
@@ -218,13 +218,16 @@ EOF
 
     echo "Waiting for $name to finish booting..."
     (
-      for i in $(seq 1 30); do
-        if (set -x; timeout 20s ssh "${sshOptions[@]}" "$publicIp" "test -f /.instance-startup-complete"); then
-          break
+      set -x
+      for i in $(seq 1 45); do
+        if timeout 20s ssh "${sshOptions[@]}" "$publicIp" "test -f /.instance-startup-complete"; then
+          exit 0
         fi
         sleep 2
         echo "Retry $i..."
       done
+      echo "$name failed to boot."
+      exit 1
     )
     echo "$name has booted."
   }
@@ -389,6 +392,7 @@ $(
     install-libssl-compatability.sh \
     install-rsync.sh \
     network-config.sh \
+
 )
 
 cat > /etc/motd <<EOM

net/remote/remote-node.sh

@@ -85,6 +85,7 @@ local|tar)
 
   ./fetch-perf-libs.sh
   export LD_LIBRARY_PATH="$PWD/target/perf-libs:$LD_LIBRARY_PATH"
+  echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH"
 
   scripts/oom-monitor.sh  > oom-monitor.log 2>&1 &
   scripts/net-stats.sh  > net-stats.log 2>&1 &