Enable GPUs and secondary disks for TdS net, pull external account file (#5031)

* Enable V100 GPUs over 3 regions for TdS cluster

* Turn on secondary config-local drive for tds net

* Enable long args bypass for GPU machine details

* bypass quoted long arg

* Pull external account file from wget

* typo

* Symlink config-local instead of changing the path variables

* Fix link path

ci/testnet-deploy.sh

@@ -119,6 +119,9 @@ while [[ -n $1 ]]; do
     elif [[ $1 = --fullnode-additional-disk-size-gb ]]; then
       maybeFullnodeAdditionalDiskSize="$1 $2"
       shift 2
+    elif [[ $1 == --machine-type* ]]; then # Bypass quoted long args for GPUs
+      shortArgs+=("$1")
+      shift
     else
       usage "Unknown long option: $1"
     fi

ci/testnet-manager.sh

@@ -469,12 +469,23 @@ deploy() {
   tds)
     (
       set -x
-      # TODO: Should we spread the few nodes around multiple zones?
+
+      EXTERNAL_ACCOUNTS_FILE_URL=https://raw.githubusercontent.com/solana-labs/tour-de-sol/master/stage1/validator.yml
+      EXTERNAL_ACCOUNTS_FILE=/tmp/validator.yml
+
+      wget ${EXTERNAL_ACCOUNTS_FILE_URL} -O ${EXTERNAL_ACCOUNTS_FILE}
+
+      # Multiple V100 GPUs are available in us-west1, us-central1 and europe-west4
       # shellcheck disable=SC2068
       # shellcheck disable=SC2086
       NO_LEDGER_VERIFY=1 \
       NO_VALIDATOR_SANITY=1 \
-        ci/testnet-deploy.sh -p tds-solana-com -C gce ${GCE_ZONE_ARGS[0]} \
+        ci/testnet-deploy.sh -p tds-solana-com -C gce \
+          -G "--machine-type n1-standard-16 --accelerator count=2,type=nvidia-tesla-v100" \
+          -d pd-ssd \
+          -z us-west1-a \
+          -z us-central1-a \
+          -z europe-west4-a \
           -t "$CHANNEL_OR_TAG" -n "$GCE_NODE_COUNT" -c 1 -P -u \
           -a tds-solana-com --letsencrypt tds.solana.com \
           --hashes-per-tick auto \
@@ -483,9 +494,10 @@ deploy() {
           ${maybeStop:+-S} \
           ${maybeDelete:+-D} \
           --stake-internal-nodes 1000000000000 \
-          --external-accounts-file /tmp/stakes.yml \
+          --external-accounts-file "$EXTERNAL_ACCOUNTS_FILE" \
           --lamports 8589934592000000000 \
-          --skip-deploy-update
+          --skip-deploy-update \
+          --fullnode-additional-disk-size-gb 32000
 
     )
     ;;

multinode-demo/common.sh

@@ -70,11 +70,14 @@ source "$SOLANA_ROOT"/scripts/configure-metrics.sh
 SOLANA_RSYNC_CONFIG_DIR=$SOLANA_ROOT/config
 
 # Configuration that remains local
+SOLANA_CONFIG_DIR=$SOLANA_ROOT/config-local
+
+# If there is a secondary disk, symlink the config-local dir there
 SECONDARY_DISK_MOUNT_POINT=/mnt/extra-disk
 if [[ -d $SECONDARY_DISK_MOUNT_POINT ]]; then
-  SOLANA_CONFIG_DIR=$SECONDARY_DISK_MOUNT_POINT/config-local
-else
-  SOLANA_CONFIG_DIR=$SOLANA_ROOT/config-local
+  mkdir -p $SECONDARY_DISK_MOUNT_POINT/config-local
+  mkdir -p "$SOLANA_ROOT"
+  ln -s $SECONDARY_DISK_MOUNT_POINT/config-local "$SOLANA_ROOT"
 fi
 
 default_arg() {

net/gce.sh

@@ -159,6 +159,9 @@ while [[ -n $1 ]]; do
     elif [[ $1 = --fullnode-additional-disk-size-gb ]]; then
       fullNodeAdditionalDiskSizeInGb="$2"
       shift 2
+    elif [[ $1 == --machine-type* ]]; then # Bypass quoted long args for GPUs
+      shortArgs+=("$1")
+      shift
     else
       usage "Unknown long option: $1"
     fi