[CI] issue: HPCINFRA-3936 Valgrind worker threads mode

Today we are running the Valgrind test in R2C mode only, meaning we are
not testing the new worker threads mode in our CI process.

Add Worker threads mode test to the vg.sh based on WORKER_THREADS env
var

Signed-off-by: NirWolfer <[email protected]>

Author: NirWolfer

.ci/matrix_job.yaml

@@ -123,6 +123,18 @@ runs_on_dockers:
      runAsUser: '0',
      runAsGroup: '0'
     }
+  - {
+     arch: 'x86_64',
+     name: 'vg-worker-threads',
+     url: 'harbor.mellanox.com/swx-infra/media/xlio/x86_64/ubuntu22.04/vg:20250219',
+     category: 'tool',
+     annotations: [{ key: 'k8s.v1.cni.cncf.io/networks', value: 'sriov-cx6dx-p2' }],
+     limits: '{memory: 10Gi, cpu: 10000m, hugepages-2Mi: 10Gi, nvidia.com/sriov-cx6dx-p2: 1}',
+     requests: '{memory: 10Gi, cpu: 10000m, hugepages-2Mi: 10Gi, nvidia.com/sriov-cx6dx-p2: 1}',
+     caps_add: '[ IPC_LOCK, SYS_RESOURCE ]',
+     runAsUser: '0',
+     runAsGroup: '0'
+    }
   - {
       file: '.ci/dockerfiles/Dockerfile.ubuntu22.04',
       arch: 'x86_64',
@@ -139,12 +151,9 @@ runs_on_dockers:
       runAsGroup: '0'
     }
   - {
-      file: '.ci/dockerfiles/Dockerfile.ubuntu22.04',
       arch: 'x86_64',
       name: 'sockperf-worker-threads',
-      uri: 'xlio/$arch/ubuntu22.04/$name',
-      tag: '20251015',
-      build_args: '--no-cache --target test',
+      url: 'harbor.mellanox.com/swx-infra/media/xlio/x86_64/ubuntu22.04/sockperf:20251015',
       category: 'tests',
       annotations: [{ key: 'k8s.v1.cni.cncf.io/networks', value: 'sriov-cx6dx-p1' }],
       limits: '{memory: 10Gi, cpu: 10000m, hugepages-2Mi: 10Gi, nvidia.com/sriov-cx6dx-p1: 1}',
@@ -208,6 +217,7 @@ steps:
       - "{category: 'base'}"
       - "{category: 'tests'}"
       - "{name: 'vg'}"
+      - "{name: 'vg-worker-threads'}"
     agentSelector:
       - "{nodeLabel: 'skip-agent'}"
     run: |
@@ -441,7 +451,23 @@ steps:
       - "{nodeLabel: 'skip-agent'}"
     run: |
       [ "x${do_valgrind}" == "xtrue" ] && action=yes || action=no
-      env WORKSPACE=$PWD TARGET=${flags} jenkins_test_vg=${action} ./contrib/test_jenkins.sh
+      env WORKSPACE=$PWD TARGET=${flags} jenkins_test_vg=${action} WORKER_THREADS=false ./contrib/test_jenkins.sh
+    parallel: false
+    onfail: |
+      ./.ci/artifacts.sh
+    archiveArtifacts-onfail: |
+      jenkins/**/arch-*.tar.gz,
+      jenkins/**/vg/*valgrind*.log
+
+  - name: Valgrind (Worker thread mode)
+    enable: ${do_valgrind}
+    containerSelector:
+      - "{name: 'vg-worker-threads'}"
+    agentSelector:
+      - "{nodeLabel: 'skip-agent'}"
+    run: |
+      [ "x${do_valgrind}" == "xtrue" ] && action=yes || action=no
+      env WORKSPACE=$PWD TARGET=${flags} jenkins_test_vg=${action} WORKER_THREADS=true ./contrib/test_jenkins.sh
     parallel: false
     onfail: |
       ./.ci/artifacts.sh
@@ -469,6 +495,7 @@ steps:
     containerSelector:
       - "{category: 'base'}"
       - "{name: 'vg'}"
+      - "{name: 'vg-worker-threads'}"
       - "{category: 'tests'}"
     agentSelector:
       - "{nodeLabel: 'beni09'}"

contrib/jenkins_tests/vg.sh

@@ -1,14 +1,14 @@
 #!/bin/bash -eExl
 
+ulimit -n 50000
+ulimit -l unlimited
 source $(dirname $0)/globals.sh
 
 # Fix hugepages for docker environments
 do_hugepages
 
 echo "Checking for valgrind ..."
 
-#do_module "tools/valgrind-3.12.0"
-
 set +eE
 
 cd $WORKSPACE
@@ -45,12 +45,25 @@ if [ "$test_ip_list" == "eth_ip4: eth_ip6:" ] || [ -z "${test_ip_list}" ]; then
 	exit 1
 fi
 
-test_list="tcp:--tcp udp:"
 test_lib=${vg_dir}/install/lib/${prj_lib}
-test_lib_env="XLIO_MEM_ALLOC_TYPE=ANON XLIO_MEMORY_LIMIT=256MB XLIO_TX_WRE=2000 XLIO_RX_WRE=2000 XLIO_STRQ=off"
+test_lib_env="XLIO_MEM_ALLOC_TYPE=ANON XLIO_TX_WRE=2000 XLIO_RX_WRE=2000 XLIO_STRQ=off LD_PRELOAD=$test_lib"
 test_app=sockperf
 test_app_path=${test_dir}/sockperf/install/bin/sockperf
 vg_tool=/bin/valgrind
+WORKER_THREADS=${WORKER_THREADS:-"false"}
+
+# Enable params based on the mode choice - given by WORKER_THREADS variable
+if [[ "$WORKER_THREADS" == "true" ]]; then
+	echo "Testing mode: Worker Threads"
+	test_list="tcp:--tcp"
+	test_lib_env="${test_lib_env} XLIO_WORKER_THREADS=1 XLIO_MEMORY_LIMIT=512MB"
+	test_params="--nonblocked"
+else
+	echo "Testing mode: R2C"
+	test_list="tcp:--tcp udp:"
+	test_lib_env="${test_lib_env} XLIO_MEMORY_LIMIT=256MB"
+	test_params=""
+fi
 
 if [ $(command -v $test_app_path >/dev/null 2>&1 || echo $?) ]; then
 	test_app_path=sockperf
@@ -92,10 +105,11 @@ for test_link in $test_ip_list; do
 			--undef-value-errors=yes --track-fds=yes --num-callers=32 \
 			--fullpath-after=${WORKSPACE} --gen-suppressions=all \
 			--suppressions=${WORKSPACE}/contrib/valgrind/valgrind_xlio.supp \
+			--fair-sched=yes \
 			"
-		eval "${sudo_cmd} $timeout_exe env $test_lib_env LD_PRELOAD=$test_lib \
+		eval "${sudo_cmd} $timeout_exe env $test_lib_env \
 			${vg_tool} --log-file=${vg_dir}/${test_name}-valgrind-sr.log $vg_args \
-			$test_app_path sr ${test_opt} -i ${test_ip} 2>&1 | tee ${vg_dir}/${test_name}-output-sr.log &"
+			$test_app_path sr ${test_opt} -i ${test_ip} ${test_params} 2>&1 | tee ${vg_dir}/${test_name}-output-sr.log &"
 
 		wait=0
 		while [ $wait -lt $sockperf_max_wait ]; do
@@ -108,9 +122,9 @@ for test_link in $test_ip_list; do
 			sleep 2
 		done
 
-		eval "${sudo_cmd} $timeout_exe_short env $test_lib_env LD_PRELOAD=$test_lib \
+		eval "${sudo_cmd} $timeout_exe_short env $test_lib_env \
 			${vg_tool} --log-file=${vg_dir}/${test_name}-valgrind-cl.log $vg_args \
-			$test_app_path pp ${test_opt} -i ${test_ip} -t 10 2>&1 | tee ${vg_dir}/${test_name}-output-cl.log"
+			$test_app_path pp ${test_opt} -i ${test_ip} -t 10 ${test_params} 2>&1 | tee ${vg_dir}/${test_name}-output-cl.log"
 
 		if [ `ps -ef | grep $test_app | wc -l` -gt 1 ];
 		then