CSIT-360: Parallel test sets run

- use parallel test set runs on as many VIRL sessions as needed
  for defined test sets

Change-Id: I7640f15894a1451aba963989fab4dc838f2ab6d8
Signed-off-by: Jan Gelety <jgelety@cisco.com>

1 files changed, 247 insertions, 83 deletions

diff --git a/bootstrap.sh b/bootstrap.sh
index f822badde5..2796e5031c 100755
--- a/bootstrap.sh
+++ b/bootstrap.sh
@@ -22,7 +22,6 @@ sudo apt-get -y update
 sudo apt-get -y install libpython2.7-dev python-virtualenv
 
 VIRL_SERVERS=("10.30.51.28" "10.30.51.29" "10.30.51.30")
-VIRL_SERVER=""
 
 VIRL_USERNAME=jenkins-in
 VIRL_PKEY=priv_key
@@ -31,6 +30,18 @@ VIRL_SERVER_EXPECTED_STATUS="PRODUCTION"
 
 SSH_OPTIONS="-i ${VIRL_PKEY} -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -o BatchMode=yes -o LogLevel=error"
 
+TEST_GROUPS=("bridge_domain,dhcp,gre,honeycomb,l2_xconnect,lisp,softwire" "cop,ipfix,ipsec,ipv6,rpf,tap,vrf" "fds,iacl,ipv4,policer,vlan,vxlan")
+SUITE_PATH="tests.func"
+
+SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+LOG_PATH="${SCRIPT_DIR}"
+
+# Create temp dir for tarballs
+mkdir ${SCRIPT_DIR}/../temp
+export TMPDIR="${SCRIPT_DIR}/../temp"
+
+SHARED_MEMORY_PATH="/run/shm"
+
 function ssh_do() {
     echo
     echo "### "  ssh $@
@@ -70,42 +81,44 @@ EOF
 chmod 600 ${VIRL_PKEY}
 
 #
-# Pick a random host from the array of VIRL servers, and attempt
-# to reach it and verify it's status.
+# Pick a random host from the array of VIRL servers for every test set run
+# and attempt to reach it and verify it's status.
 #
-# The server must be reachable, and have a "status" file with
-# the content "PRODUCTION", to be selected.
+# The server must be reachable and have a "status" file with
+# the content "PRODUCTION" to be selected.
 #
-# If the server is not reachable, or does not have the correct
-# status, remove it from the array and start again.
+# If the server is not reachable or does not have the correct
+# status remove it from the array and start again.
 #
 # Abort if there are no more servers left in the array.
 #
-while [[ ! "$VIRL_SERVER" ]]
-do
-    num_hosts=${#VIRL_SERVERS[@]}
-    if [ $num_hosts == 0 ]
-    then
-        echo "No more VIRL candidate hosts available, failing."
-        exit 127
-    fi
-    element=$[ $RANDOM % $num_hosts ]
-    virl_server_candidate=${VIRL_SERVERS[$element]}
-    virl_server_status=$(ssh ${SSH_OPTIONS} ${VIRL_USERNAME}@${virl_server_candidate} cat $VIRL_SERVER_STATUS_FILE 2>&1)
-    echo VIRL HOST $virl_server_candidate status is \"$virl_server_status\"
-    if [ "$virl_server_status" == "$VIRL_SERVER_EXPECTED_STATUS" ]
-    then
-        # Candidate is in good status. Select this server.
-        VIRL_SERVER="$virl_server_candidate"
-    else
-        # Candidate is in bad status. Remove from array.
-        VIRL_SERVERS=("${VIRL_SERVERS[@]:0:$element}" "${VIRL_SERVERS[@]:$[$element+1]}")
-    fi
+for index in "${!TEST_GROUPS[@]}"; do
+    VIRL_SERVER[${index}]=""
+    while [[ ! "${VIRL_SERVER[${index}]}" ]]; do
+        num_hosts=${#VIRL_SERVERS[@]}
+        if [ $num_hosts == 0 ]
+        then
+            echo "No more VIRL candidate hosts available, failing."
+            exit 127
+        fi
+        element=$[ $RANDOM % $num_hosts ]
+        virl_server_candidate=${VIRL_SERVERS[$element]}
+        virl_server_status=$(ssh ${SSH_OPTIONS} ${VIRL_USERNAME}@${virl_server_candidate} cat $VIRL_SERVER_STATUS_FILE 2>&1)
+        echo VIRL HOST $virl_server_candidate status is \"$virl_server_status\"
+        if [ "$virl_server_status" == "$VIRL_SERVER_EXPECTED_STATUS" ]
+        then
+            # Candidate is in good status. Select this server.
+            VIRL_SERVER[${index}]="$virl_server_candidate"
+        else
+            # Candidate is in bad status. Remove from array.
+            VIRL_SERVERS=("${VIRL_SERVERS[@]:0:$element}" "${VIRL_SERVERS[@]:$[$element+1]}")
+        fi
+    done
 done
 
-# Temporarily download VPP packages from nexus.fd.io
-SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+echo "Selected VIRL servers: ${VIRL_SERVER[@]}"
 
+# Temporarily download VPP packages from nexus.fd.io
 if [ "${#}" -ne "0" ]; then
     arr=(${@})
     echo ${arr[0]}
@@ -136,56 +149,75 @@ done
 echo "Updated file names: " ${VPP_DEBS_FULL[@]}
 
 cat ${VIRL_PKEY}
-# Copy the files to VIRL host
-scp ${SSH_OPTIONS} *.deb \
-    ${VIRL_USERNAME}@${VIRL_SERVER}:${VIRL_DIR_LOC}/
-
-result=$?
-if [ "${result}" -ne "0" ]; then
-    echo "Failed to copy vpp deb files to virl host"
-    echo ${result}
-    exit ${result}
-fi
+
+# Copy the files to VIRL hosts
+DONE=""
+for index in "${!VIRL_SERVER[@]}"; do
+    # Do not copy files in case they have already been copied to the VIRL host
+    [[ "${DONE[@]}" =~ "${VIRL_SERVER[${index}]}" ]] && copy=0 || copy=1
+
+    if [ "${copy}" -eq "0" ]; then
+        echo "VPP deb files have already been copied to the VIRL host ${VIRL_SERVER[${index}]}"
+    else
+        scp ${SSH_OPTIONS} *.deb \
+        ${VIRL_USERNAME}@${VIRL_SERVER[${index}]}:${VIRL_DIR_LOC}/
+
+         result=$?
+        if [ "${result}" -ne "0" ]; then
+            echo "Failed to copy VPP deb files to VIRL host ${VIRL_SERVER[${index}]}"
+            echo ${result}
+            exit ${result}
+        else
+            echo "VPP deb files successfully copied to the VIRL host ${VIRL_SERVER[${index}]}"
+        fi
+        DONE+=(${VIRL_SERVER[${index}]})
+    fi
+done
 
 # Start a simulation on VIRL server
-echo "Starting simulation on VIRL server"
 
 function stop_virl_simulation {
-    ssh ${SSH_OPTIONS} ${VIRL_USERNAME}@${VIRL_SERVER}\
-        "stop-testcase ${VIRL_SID}"
+    for index in "${!VIRL_SERVER[@]}"; do
+        ssh ${SSH_OPTIONS} ${VIRL_USERNAME}@${VIRL_SERVER[${index}]}\
+            "stop-testcase ${VIRL_SID[${index}]}"
+    done
 }
 
-VIRL_SID=$(ssh ${SSH_OPTIONS} \
-    ${VIRL_USERNAME}@${VIRL_SERVER} \
-    "start-testcase -c double-ring-nested ${VPP_DEBS_FULL[@]}")
-retval=$?
-if [ "$?" -ne "0" ]; then
-    echo "VIRL simulation start failed"
-    exit ${retval}
-fi
-
-if [[ ! "${VIRL_SID}" =~ session-[a-zA-Z0-9_]{6} ]]; then
-    echo "No VIRL session ID reported."
-    exit 127
-fi
-
 # Upon script exit, cleanup the simulation execution
 trap stop_virl_simulation EXIT
-echo ${VIRL_SID}
 
-ssh_do ${VIRL_USERNAME}@${VIRL_SERVER} cat /scratch/${VIRL_SID}/topology.yaml
+for index in "${!VIRL_SERVER[@]}"; do
+    echo "Starting simulation nr. ${index} on VIRL server ${VIRL_SERVER[${index}]}"
+    VIRL_SID[${index}]=$(ssh ${SSH_OPTIONS} \
+        ${VIRL_USERNAME}@${VIRL_SERVER[${index}]} \
+        "start-testcase -c double-ring-nested ${VPP_DEBS_FULL[@]}")
+    retval=$?
+    if [ "$?" -ne "0" ]; then
+        echo "VIRL simulation start failed on ${VIRL_SERVER[${index}]}"
+        exit ${retval}
+    fi
+    if [[ ! "${VIRL_SID[${index}]}" =~ session-[a-zA-Z0-9_]{6} ]]; then
+        echo "No VIRL session ID reported."
+        exit 127
+    fi
+    echo "VIRL simulation started on ${VIRL_SERVER[${index}]}"
+
+    ssh_do ${VIRL_USERNAME}@${VIRL_SERVER[${index}]}\
+     cat /scratch/${VIRL_SID[${index}]}/topology.yaml
 
-# Download the topology file from virl session
-scp ${SSH_OPTIONS} \
-    ${VIRL_USERNAME}@${VIRL_SERVER}:/scratch/${VIRL_SID}/topology.yaml \
-    topologies/enabled/topology.yaml
+    # Download the topology file from VIRL session and rename it
+    scp ${SSH_OPTIONS} \
+        ${VIRL_USERNAME}@${VIRL_SERVER[${index}]}:/scratch/${VIRL_SID[${index}]}/topology.yaml \
+        topologies/enabled/topology${index}.yaml
 
-retval=$?
-if [ "$?" -ne "0" ]; then
-    echo "Failed to copy topology file from VIRL simulation"
-    exit ${retval}
-fi
+    retval=$?
+    if [ "$?" -ne "0" ]; then
+        echo "Failed to copy topology file from VIRL simulation nr. ${index} on VIRL server ${VIRL_SERVER[${index}]}"
+        exit ${retval}
+    fi
+done
 
+echo ${VIRL_SID[@]}
 
 virtualenv --system-site-packages env
 . env/bin/activate
@@ -193,22 +225,154 @@ virtualenv --system-site-packages env
 echo pip install
 pip install -r ${SCRIPT_DIR}/requirements.txt
 
-pykwalify -s ${SCRIPT_DIR}/resources/topology_schemas/3_node_topology.sch.yaml \
-          -s ${SCRIPT_DIR}/resources/topology_schemas/topology.sch.yaml \
-          -d ${SCRIPT_DIR}/topologies/enabled/topology.yaml \
-          -vvv
+for index in "${!VIRL_SERVER[@]}"; do
+    pykwalify -s ${SCRIPT_DIR}/resources/topology_schemas/3_node_topology.sch.yaml \
+              -s ${SCRIPT_DIR}/resources/topology_schemas/topology.sch.yaml \
+              -d ${SCRIPT_DIR}/topologies/enabled/topology${index}.yaml \
+              -vvv
+    if [ "$?" -ne "0" ]; then
+        echo "Topology${index} schema validation failed."
+        echo "However, the tests will start."
+    fi
+done
+
+function run_test_set() {
+    set +x
+    OLDIFS=$IFS
+    IFS=","
+    nr=$(echo $1)
+    rm -f ${LOG_PATH}/test_run${nr}.log
+    exec &> >(while read line; do echo "$(date +'%H:%M:%S') $line" \
+     >> ${LOG_PATH}/test_run${nr}.log; done;)
+    suite_str=""
+    for suite in ${TEST_GROUPS[${nr}]}; do
+        suite_str="${suite_str} --suite ${SUITE_PATH}.${suite}"
+    done
+    IFS=$OLDIFS
 
-if [ "$?" -ne "0" ]; then
-    echo "Topology schema validation failed."
-    echo "However, the tests will start."
+    echo "PYTHONPATH=`pwd` pybot -L TRACE -W 136\
+        -v TOPOLOGY_PATH:${SCRIPT_DIR}/topologies/enabled/topology${nr}.yaml \
+        ${suite_str} \
+        --include vm_envAND3_node_single_link_topo \
+        --include vm_envAND3_node_double_link_topo \
+        --exclude PERFTEST \
+        --exclude SKIP_PATCH \
+        --noncritical EXPECTED_FAILING \
+        --output ${LOG_PATH}/log_test_set_run${nr} \
+        tests/"
+
+    PYTHONPATH=`pwd` pybot -L TRACE -W 136\
+        -v TOPOLOGY_PATH:${SCRIPT_DIR}/topologies/enabled/topology${nr}.yaml \
+        ${suite_str} \
+        --include vm_envAND3_node_single_link_topo \
+        --include vm_envAND3_node_double_link_topo \
+        --exclude PERFTEST \
+        --exclude SKIP_PATCH \
+        --noncritical EXPECTED_FAILING \
+        --output ${LOG_PATH}/log_test_set_run${nr} \
+        tests/
+
+    local_run_rc=$?
+    echo ${local_run_rc} > ${SHARED_MEMORY_PATH}/rc_test_run${nr}
+    set -x
+}
+
+set +x
+# Send to background an instance of the run_test_set() function for each number,
+# record the pid.
+for index in "${!VIRL_SERVER[@]}"; do
+    run_test_set ${index} &
+    pid=$!
+    echo "Sent to background: Test_set${index} (pid=$pid)"
+    pids[$pid]=$index
+done
+
+echo
+echo -n "Waiting..."
+
+# Watch the stable of background processes.
+# If a pid goes away, remove it from the array.
+while [ -n "${pids[*]}" ]; do
+    for i in $(seq 0 9); do
+        sleep 1
+        echo -n "."
+    done
+    for pid in "${!pids[@]}"; do
+        if ! ps "$pid" >/dev/null; then
+            echo -e "\n"
+            echo "Test_set${pids[$pid]} with PID $pid finished."
+            unset pids[$pid]
+        fi
+    done
+    if [ -z "${!pids[*]}" ]; then
+        break
+    fi
+    echo -n -e "\nStill waiting for test set(s): ${pids[*]} ..."
+done
+
+echo
+echo "All test set runs finished."
+echo
+
+set -x
+
+RC=0
+for index in "${!VIRL_SERVER[@]}"; do
+    echo "Test_set${index} log:"
+    cat ${LOG_PATH}/test_run${index}.log
+    RC_PARTIAL_RUN=$(cat ${SHARED_MEMORY_PATH}/rc_test_run${index})
+    RC=$((RC+RC_PARTIAL_RUN))
+    rm -f ${SHARED_MEMORY_PATH}/rc_test_run${index}
+    rm -f ${LOG_PATH}/test_run${index}.log
+    echo
+done
+
+# Log the final result
+if [ "${RC}" -eq "0" ]; then
+    set +x
+    echo
+    echo "========================================================================================================================================"
+    echo "Final result of all test loops:                                                                                                 | PASS |"
+    echo "All critical tests have passed."
+    echo "========================================================================================================================================"
+    echo
+    set -x
+else
+    if [ "${RC}" -eq "1" ]; then
+        HLP_STR="test has"
+    else
+        HLP_STR="tests have"
+    fi
+    set +x
+    echo
+    echo "========================================================================================================================================"
+    echo "Final result of all test loops:                                                                                                 | FAIL |"
+    echo "${RC} critical ${HLP_STR} failed."
+    echo "========================================================================================================================================"
+    echo
+    set -x
+fi
+
+echo Post-processing test data...
+
+partial_logs=""
+for index in "${!VIRL_SERVER[@]}"; do
+    partial_logs="${partial_logs} ${LOG_PATH}/log_test_set_run${index}.xml"
+done
+
+# Rebot output post-processing
+rebot --noncritical EXPECTED_FAILING \
+      --output output.xml ${partial_logs}
+
+# Remove unnecessary log files
+rm -f ${partial_logs}
+
+echo Post-processing finished.
+
+if [ ${RC} -eq 0 ]; then
+    RETURN_STATUS=0
+else
+    RETURN_STATUS=1
 fi
 
-PYTHONPATH=`pwd` pybot -L TRACE -W 136\
-    -v TOPOLOGY_PATH:${SCRIPT_DIR}/topologies/enabled/topology.yaml \
-    --suite "tests.func" \
-    --include vm_envAND3_node_single_link_topo \
-    --include vm_envAND3_node_double_link_topo \
-    --exclude PERFTEST \
-    --exclude SKIP_PATCH \
-    --noncritical EXPECTED_FAILING \
-    tests/
+exit ${RETURN_STATUS}