Update switch.sh to provide more details. (#113)

* Add HPECRAY-23 log settings for PAXs Signed-off-by: Anthony Floeder <[email protected]> * switch.sh: Add device and PAX ID info to all commands and support a VERBOSE option switch.sh: Add PCI info to the status command. switch.sh: Don't display slots 1 and 13 Signed-off-by: Anthony Floeder <[email protected]> --------- Signed-off-by: Anthony Floeder <[email protected]>
NearNodeFlash · Dec 5, 2024 · 51a5364 · 51a5364
1 parent afc4d0c
commit 51a5364
Show file tree

Hide file tree

Showing 6 changed files with 323 additions and 267 deletions.
diff --git a/tools/_util.sh b/tools/_util.sh
diff --git a/tools/create-write-delete-namespaces.sh b/tools/create-write-delete-namespaces.sh
@@ -16,62 +16,80 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+set -euo pipefail
+
+DURATION=${1:-"30s"}
+OP=${2:-"randread"}
 
 usage() {
     cat <<EOF
-Create a namespace on every drive, create a logical volume from those namespaces, write the logical volume, then delete everything
-Usage: $0 [NAMESPACE-SIZE-IN-BYTES]
+Create a namespace on every drive, create a logical volume from those namespaces, operate over that logical volume, then delete everything
 
-NAMESPACE-SIZE-IN-BYTES: 0 -> (Allocate entire drive)
+Usage: $0 [duration: default 30s] [operation: randread|randwrite: default randread]
+Examples:
+   $0 60s  randread                 # random read test for 60 seconds
+   $0 60m  randread                 # random read test for 60 minutes
+   $0 300s randwrite                # random write test for 300 seconds
 EOF
 }
 
-SIZE=${1:-0}
+while getopts "h" OPTION
+do
+    case "${OPTION}" in
+        'h')
+            usage
+            exit 0
+            ;;
+        *)
+            ;;
+    esac
+done
+shift $((OPTIND - 1))
+
+SIZE=0
 DELAY_FOR_DEVICES=2
-DELAY_TO_COOL_CPU=0
-
-case "$SIZE" in
-    ''|*[!0-9]*)
-        usage
-        exit 1
-    ;;
-    *)
-        # Create and attach a namespace on each drive to the RABBIT's processor
-        ./nvme.sh create "$SIZE"
-        ./nvme.sh attach
-
-        printf "Sleeping %d seconds to allow all devices to be available for logical volume\n" "$DELAY_FOR_DEVICES"
-	    sleep "$DELAY_FOR_DEVICES"
-
-        # Create a logical volume spanning all the namespaces
-        ./lvm.sh create
-
-        # Write a little something to the logical volume to give the drives some work to do on delete-ns operation
-        fio --direct=1 --rw=randwrite --bs=32M --ioengine=libaio --iodepth=128 --numjobs=4 --runtime=5m --time_based --group_reporting --name=rabbit --eta-newline=1 --filename=/dev/rabbit/rabbit
-
-        if [ $DELAY_TO_COOL_CPU != 0 ]
-        then
-            printf "Sleeping %d seconds to cool the CPU\n" "$DELAY_TO_COOL_CPU"
-            sleep "$DELAY_TO_COOL_CPU"
-        fi
-        # Delete the logical volume to tidy up
-        ./lvm.sh delete
-
-    	# Format the namespace to speed up deletion
-	    ./nvme.sh cmd format --force --namspace-id=1
-
-        # Wait for the format to finish
-        areWeDone=$(nvme list | grep KIO | awk '{if ($6 > 0) { print "wait" } else { print "done" }}')
-        while [ "$areWeDone" != "done" ]; do
-            sleep 1
-            printf "Formatting, space left %f\n" "$(nvme list | grep KIO | awk '{print $6}')"
-            areWeDone=$(nvme list | grep KIO | awk '{if ($6 > 0) { print "wait" } else { print "done" }}')
-        done
-
-        # Show the nvme namespaces for the record
-        nvme list | grep KIO
-
-        # Finally, delete the namespaces
-        ./nvme.sh -t delete
-    ;;
-esac
+
+printf "duration %s\n" "$DURATION"
+printf "operation %s\n" "$OP"
+
+# Ensure lvm.conf doesn't get in the way
+sed -i 's/use_lvmlockd = 1/use_lvmlockd = 0/g' /etc/lvm/lvm.conf
+
+# Ensure the /dev/rabbit directory doesn't already exist
+rm -rf /dev/rabbit
+
+# Run nnf-ec just to be sure
+./nnf-ec
+
+# Create and attach a namespace on each drive to the RABBIT's processor
+./nvme.sh create "$SIZE"
+./nvme.sh attach
+
+printf "Sleeping %d seconds to allow all devices to be available for logical volume\n" "$DELAY_FOR_DEVICES"
+sleep "$DELAY_FOR_DEVICES"
+
+# Create a logical volume spanning all the namespaces
+./lvm.sh create
+
+# Write a little something to the logical volume to give the drives some work to do on delete-ns operation
+fio --direct=1 --rw="$OP" --bs=32M --ioengine=libaio --iodepth=128 --numjobs=4 --runtime="$DURATION" --time_based --group_reporting --name=rabbit --eta-newline=1 --filename=/dev/rabbit/rabbit
+
+# Delete the logical volume to tidy up
+./lvm.sh delete
+
+# Format the namespace to speed up deletion
+./nvme.sh cmd format --force --namespace-id=1
+
+# Wait for the format to finish
+gbToFormat=$(nvme list | grep KIO | awk '{print $6}' | tr ' ' '\n' | paste -sd+ - | bc | awk '{print ($1 == int($1)) ? int($1) : int($1) + 1}')
+while (("$gbToFormat" > 0)); do
+    sleep 1
+    printf "Formatting, space left %d\n" "$(nvme list | grep KIO | awk '{print $6}' | tr ' ' '\n' | paste -sd+ - | bc | awk '{print ($1 == int($1)) ? $1 : int($1) + 1}')"
+    gbToFormat=$(nvme list | grep KIO | awk '{print $6}' | tr ' ' '\n' | paste -sd+ - | bc | awk '{print ($1 == int($1)) ? int($1) : int($1) + 1}')
+done
+
+# Show the nvme namespaces for the record
+nvme list | grep KIO
+
+# Finally, delete the namespaces
+./nvme.sh -t delete
diff --git a/tools/devices.sh b/tools/devices.sh
diff --git a/tools/nvme.sh b/tools/nvme.sh
@@ -16,14 +16,9 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+set -eo pipefail
 shopt -s expand_aliases
 
-# Pull in common utility functions
-SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
-# shellcheck source="$SCRIPT_DIR"/_util.sh
-source "$SCRIPT_DIR"/_util.sh
-
 usage() {
     cat <<EOF
 Run various NVMe Namespace commands over the NVMe switch fabric using
@@ -122,6 +117,38 @@ executeParallel() {
     rm _result*
 }
 
+# Retrieve the Physical Device Fabric IDs used to iterate through a list of nvme drives
+function getPDFIDs() {
+    local SWITCH=$1 FUNCTION="${2:-0}"
+
+    switchtec fabric gfms-dump "$SWITCH" | grep "Function $FUNCTION " -A2 | grep PDFID | awk '{print $2}'
+}
+
+function getDriveList() {
+    # DRIVES=$1
+    # for DRIVE in $(ls /dev/nvme* | grep -E "nvme[[:digit:]]+$");
+    for DRIVE in /dev/nvme[0-9]*;
+    do
+        # shellcheck disable=SC2086
+        if [ "$(nvme id-ctrl ${DRIVE} | grep -e KIOXIA -e 'SAMSUNG MZ3LO1T9HCJR')" != "" ];
+        then
+            # SerialNumber=$(nvme id-ctrl ${DRIVE} | grep -E "^sn " | awk '{print $3}')
+            # Mfg=$(nvme id-ctrl ${DRIVE} | grep -E "^mn " | awk '{print $3}')
+            # FW=$(nvme id-ctrl ${DRIVE} | grep -E "^fr " | awk '{print $3}')
+            # printf "%s\t%s\t%s\t%s\n" "$DRIVE" "$Mfg" "$SerialNumber" "$FW"
+
+            DRIVES+=("${DRIVE}")
+        fi
+    done
+
+    DriveCount="${#DRIVES[@]}"
+    if ((DriveCount == 0));
+    then
+        printf "No drives found: Did you run nnf-ec?\n"
+    fi
+}
+
+
 alias TIME=""
 while getopts "th:" OPTION
 do