gather_openvswitch

#!/bin/bash
BASE_COLLECTION_PATH="must-gather"
OVS_DUMP_PATH=${OUT:-"${BASE_COLLECTION_PATH}/openvswitch"}

mkdir -p "${NETWORK_LOG_PATH}"/

GLOBAL_OVS_FILE_COPY=(
    "/run/openvswitch/ovsdb-server.pid"
    "/run/openvswitch/ovs-vswitchd.pid"
    "/run/openvswitch/ovs-monitor-ipsec.pid"
    "/var/log/openvswitch/*"
    "/usr/local/var/log/openvswitch/*"
)

function debug_node_gather() {
    PIDS+=($!)
    cat <<EOF | oc debug --to-namespace="default" node/"$1" > "${NODE_DUMP_PATH}"/debug_node.log
echo "chroot to host"
chroot /host

echo "=== dbclient list ==="
ovsdb-client -f list dump

echo "=== ovs-vsctl -t 5 show ==="
ovs-vsctl -t 5 show

echo "List the contents of important runtime directories"
ls -laZ /run/openvswitch
ls -laZ /dev/hugepages/
ls -laZ /dev/vfio
ls -laZ /var/lib/vhost_sockets

echo "=== Capture a list of all bond devices ==="
ovs-appctl bond/list
echo "=== Capture more details from bond devices ==="
ovs-appctl bond/show
echo "=== Capture LACP details ==="
ovs-appctl lacp/show
ovs-appctl lacp/show-stats
echo "=== Capture coverage stats" ==="
ovs-appctl coverage/show
echo "=== Capture cached routes ==="
ovs-appctl ovs/route/show
echo "=== Capture tnl arp table" ==="
ovs-appctl tnl/arp/show
echo "=== Capture a list of listening ports" ==="
ovs-appctl tnl/ports/show -v
echo "=== Capture upcall information ==="
ovs-appctl upcall/show
echo "=== Capture DPDK and other parameters ==="
ovs-vsctl -t 5 get Open_vSwitch . other_config

echo "=== Capture OVS list ==="
ovs-vsctl -t 5 list Open_vSwitch
echo "=== Capture OVS interface list ==="
ovs-vsctl -t 5 list interface
echo "=== Capture OVS detailed information from all the bridges ==="
ovs-vsctl -t 5 list bridge
echo "=== Capture DPDK queue to pmd mapping ==="
ovs-appctl dpif-netdev/pmd-rxq-show
echo "=== Capture DPDK pmd stats ==="
ovs-appctl dpif-netdev/pmd-stats-show
echo "=== Capture DPDK pmd performance counters ==="
ovs-appctl dpif-netdev/pmd-perf-show
echo "=== Capture ofproto tunnel configs ==="
ovs-appctl ofproto/list-tunnels

echo "=== Capture ipsec tunnel information ==="
ovs-appctl -t ovs-monitor-ipsec tunnels/show
ovs-appctl -t ovs-monitor-ipsec xfrm/state
ovs-appctl -t ovs-monitor-ipsec xfrm/policies

echo "=== capture dps ==="
ovs-appctl dpctl/dump-dps

echo "=== capture dp-info ==="
echo "-- dpctl/show ovs-system --"
ovs-appctl dpctl/show -s ovs-system

echo "-- ovs-appctl dpctl/dump-flows --"
ovs-appctl dpctl/dump-flows -m ovs-system

echo "-- ovs-appctl dpctl/dump-conntrack --"
ovs-appctl dpctl/dump-conntrack -m ovs-system

sleep 300
EOF
}

function get_ovs_data_off_node() {
    local debugPod=""

    # Add this process to the PIDs variable
    PIDS+=($!)

    #Get debug pod's name
    debugPod=$(oc debug --to-namespace="default" node/"$1" -o jsonpath='{.metadata.name}')

    NODE_DUMP_PATH="${OVS_DUMP_PATH}"/"$1"
    mkdir "${NODE_DUMP_PATH}"

    #Start Debug pod force it to stay up until removed in "default" namespace
    # Here we will create a file which contains all the data dumps

    debug_node_gather "${NODE_DUMP_PATH}"/debug_node.log

    #Mimic a normal oc call, i.e pause between two successive calls to allow pod to register
    sleep 2
    oc wait -n "default" --for=condition=Ready pod/"$debugPod" --timeout=30s

    if [ -z "$debugPod" ]; then
        echo "Debug pod for node ""$1"" never activated" > "${NODE_DUMP_PATH}"/debug_node.log
    else
        # copy relevant files
        for file in ${GLOBAL_OVS_FILE_COPY}; do
            mkdir -p "${NODE_DUMP_PATH}"$(dirname "${file}")
            oc cp  --loglevel 1 -n "default" "$debugPod":/host"$file" "${NODE_DUMP_PATH}""$file" > /dev/null 2>&1
        done

        
        #clean up debug pod after we are done using them  
        oc delete pod "$debugPod" -n "default"  
    fi
}

function gather_ovs_data() {
    for NODE in ${NODES}; do
        get_ovs_data_off_node "${NODE}" &
    done
}

if [ $# -eq 0 ]; then
    echo "WARNING: ovs: Collecting ovs data from ALL nodes in your cluster."
    echo "WARNING: ovs: This can take a bit of time."
fi

PIDS=()
NODES="${*:-$(oc get nodes -o jsonpath='{.items[?(@.status.nodeInfo.operatingSystem=="linux")].metadata.name}')}"

gather_ovs_data

echo "INFO: Waiting for node core dump collection to complete ..."
wait "${PIDS[@]}"
echo "INFO: Node core dump collection to complete."

sync