Merge cockroachdb#119906

119906: roachprod: RAID0 only network disks if both local and network are present r=srosenberg a=nameisbhaskar

Today, if a machine has both local and network disks, both the disks are selected for RAID'ing.

But, RAID'ing different types of disks causes performance differences.

To address this, local disks are ignored for RAID'ing only if network disks are present.

Fixes: cockroachdb#98783
Epic: none

Co-authored-by: Bhaskarjyoti Bora <[email protected]>

pkg/cmd/roachtest/tests/restore.go

@@ -33,7 +33,6 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/roachpb"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/install"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/logger"
-	"github.com/cockroachdb/cockroach/pkg/roachprod/vm"
 	"github.com/cockroachdb/cockroach/pkg/testutils"
 	"github.com/cockroachdb/cockroach/pkg/testutils/sqlutils"
 	"github.com/cockroachdb/cockroach/pkg/ts/tspb"
@@ -537,18 +536,6 @@ func (hw hardwareSpecs) makeClusterSpecs(r registry.Registry, backupCloud string
 	}
 	s := r.MakeClusterSpec(hw.nodes+addWorkloadNode, clusterOpts...)
 
-	if backupCloud == spec.AWS && s.VolumeSize != 0 {
-		// Work around an issue that RAID0s local NVMe and GP3 storage together:
-		// https://github.com/cockroachdb/cockroach/issues/98783.
-		//
-		// TODO(srosenberg): Remove this workaround when 98783 is addressed.
-		// TODO(miral): This now returns an error instead of panicking, so even though
-		// we haven't panicked here before, we should handle the error. Moot if this is
-		// removed as per TODO above.
-		s.AWS.MachineType, _, _ = spec.SelectAWSMachineType(s.CPUs, s.Mem, false /* shouldSupportLocalSSD */, vm.ArchAMD64)
-		s.AWS.MachineType = strings.Replace(s.AWS.MachineType, "d.", ".", 1)
-		s.Arch = vm.ArchAMD64
-	}
 	return s
 }
 

pkg/roachprod/vm/aws/support.go

@@ -50,19 +50,44 @@ mount_opts="defaults"
 
 use_multiple_disks='{{if .UseMultipleDisks}}true{{end}}'
 
-disks=()
 mount_prefix="/mnt/data"
 
+# if the use_multiple_disks is not set and there are more than 1 disk (excluding the boot disk),
+# then the disks will be selected for RAID'ing. If there are both EC2 NVMe Instance Storage and
+# EBS, RAID'ing in this case can cause performance differences. So, to avoid this,
+# EC2 NVMe Instance Storage are ignored.
+# Scenarios: 
+#   (local SSD = 0, Network Disk - 1)  - no RAID'ing and mount network disk
+#   (local SSD = 1, Network Disk - 0)  - no RAID'ing and mount local SSD
+#   (local SSD >= 1, Network Disk = 1) - no RAID'ing and mount network disk
+#   (local SSD > 1, Network Disk = 0)  - local SSDs selected for RAID'ing
+#   (local SSD >= 0, Network Disk > 1) - network disks selected for RAID'ing
+# Keep track of the Local SSDs and EBS volumes for RAID'ing
+local_disks=()
+ebs_volumes=()
+
 # On different machine types, the drives are either called nvme... or xvdd.
 for d in $(ls /dev/nvme?n1 /dev/xvdd); do
   if ! mount | grep ${d}; then
-    disks+=("${d}")
+		if udevadm info --query=property --name=${d} | grep "ID_MODEL=Amazon Elastic Block Store"; then
+			echo "EBS Volume ${d} identified!"
+			ebs_volumes+=("${d}")
+		else
+			local_disks+=("${d}")
+		fi
     echo "Disk ${d} not mounted, need to mount..."
   else
     echo "Disk ${d} already mounted, skipping..."
   fi
 done
 
+# use only EBS volumes if available and ignore EC2 NVMe Instance Storage
+disks=()
+if [ "${#ebs_volumes[@]}" -gt "0" ]; then
+	disks=("${ebs_volumes[@]}")
+else
+	disks=("${local_disks[@]}")
+fi
 
 if [ "${#disks[@]}" -eq "0" ]; then
   mountpoint="${mount_prefix}1"

pkg/roachprod/vm/gce/utils.go

@@ -72,7 +72,23 @@ for d in $(ls /dev/nvme?n? /dev/disk/by-id/google-persistent-disk-[1-9]); do
   zpool list -v -P | grep ${d} > /dev/null
   if [ $? -ne 0 ]; then
 {{ else }}
-for d in $(ls /dev/disk/by-id/google-local-* /dev/disk/by-id/google-persistent-disk-[1-9]); do
+# if the use_multiple_disks is not set and there are more than 1 disk (excluding the boot disk),
+# then the disks will be selected for RAID'ing. If there are both Local SSDs and Persistent disks,
+# RAID'ing in this case can cause performance differences. So, to avoid this, local SSDs are ignored.
+# Scenarios: 
+#   (local SSD = 0, Persistent Disk - 1) - no RAID'ing and Persistent Disk mounted
+#   (local SSD = 1, Persistent Disk - 0) - no RAID'ing and local SSD mounted
+#   (local SSD >= 1, Persistent Disk = 1) - no RAID'ing and Persistent Disk mounted
+#   (local SSD > 1, Persistent Disk = 0) - local SSDs selected for RAID'ing
+#   (local SSD >= 0, Persistent Disk > 1) - network disks selected for RAID'ing
+disk_list=()
+if [ "$(ls /dev/disk/by-id/google-persistent-disk-[1-9]|wc -l)" -eq "0" ]; then
+	disk_list=$(ls /dev/disk/by-id/google-local-*)
+else
+  echo "Only persistent disks are selected."
+	disk_list=$(ls /dev/disk/by-id/google-persistent-disk-[1-9])
+fi
+for d in ${disk_list}; do
   if ! mount | grep ${d}; then
 {{ end }}
     disks+=("${d}")