contrib/intel/jenkins: Make UCX build on compute node

UCX has version/environment issues when building on head node and running on a compute node. Changing its build args and forcing to build on a compute node will solve the test failures. Signed-off-by: Zach Dworkin <[email protected]>
ofiwg · Oct 2, 2023 · 102a18d · 102a18d
1 parent 862428b
commit 102a18d
Show file tree

Hide file tree

Showing 3 changed files with 29 additions and 15 deletions.
diff --git a/contrib/intel/jenkins/Jenkinsfile b/contrib/intel/jenkins/Jenkinsfile
@@ -314,6 +314,27 @@ pipeline {
             }
           }
         }
+        stage ('build_ucx') {
+          steps {
+            script {
+              dir ("${CUSTOM_WORKSPACE}/ucx") {
+                checkout scm
+                def prefix = "python$PYTHON_VERSION ${RUN_LOCATION}/build.py"
+                def opts = ""
+                def build_cmd = ""
+                for (mode in BUILD_MODES) {
+                  for (item in ["libfabric", "fabtests"]) {
+                    opts = "--build_item=${item} --ofi_build_mode=${mode} --ucx"
+                    build_cmd = "${build_cmd} ${prefix} ${opts}; "
+                  }
+                }
+                slurm_batch("squirtle,totodile", "1",
+                            "${env.LOG_DIR}/libfabric_ucx_build_log",
+                            "${build_cmd}")
+              }
+            }
+          }
+        }
         stage ('build-daos') {
           agent {
             node {
@@ -471,14 +492,6 @@ pipeline {
         stage('ucx') {
           steps {
             script {
-              dir (CUSTOM_WORKSPACE) {
-                for (mode in BUILD_MODES) {
-                  echo "Building Libfabric $mode"
-                  build("libfabric", "${mode}", null, false, "--ucx")
-                  echo "Building Fabtests $mode"
-                  build("fabtests", "${mode}", null, false, "--ucx")
-                }
-              }
               dir (RUN_LOCATION) {
                 run_fabtests("ucx", "totodile", "2", "ucx")
               }

diff --git a/contrib/intel/jenkins/build.py b/contrib/intel/jenkins/build.py
@@ -24,6 +24,7 @@ def build_libfabric(libfab_install_path, mode, cluster=None, ucx=None):
 		config_cmd.append('--enable-debug')
 	elif (mode == 'dl'):
 		enable_prov_val='dl'
+
 	if (cluster == 'daos'):
 		prov_list = common.daos_prov_list
 	elif (cluster == 'gpu'):
@@ -32,16 +33,16 @@ def build_libfabric(libfab_install_path, mode, cluster=None, ucx=None):
 		prov_list = common.default_prov_list
 
 	for prov in prov_list:
-		if (prov == 'ucx'):
-			if (ucx):
-				config_cmd.append('--enable-ucx=yes')
+		if (ucx):
+			config_cmd.append('--enable-ucx=yes')
+			break
 		else:
 			config_cmd.append(f'--enable-{prov}={enable_prov_val}')
 
 	for op in common.common_disable_list:
 		config_cmd.append(f'--enable-{op}=no')
 
-	if (cluster == 'default' and build_item != 'libfabric_mpich'):
+	if (cluster == 'default' and build_item != 'libfabric_mpich' and not ucx):
 		for op in common.default_enable_list:
 			config_cmd.append(f'--enable-{op}')
 
@@ -187,7 +188,8 @@ def log_dir(install_path, release=False):
 	libfab_install_path = f'{cloudbees_config.install_dir}/{jobname}/{buildno}/{ofi_build_mode}'
 
 	if (ucx):
-		libfab_install_path += "/ucx"
+		libfab_install_path += '/ucx'
+		workspace += '/ucx'
 
 	p = re.compile('mpi*')
 

diff --git a/contrib/intel/jenkins/common.py b/contrib/intel/jenkins/common.py
@@ -115,8 +115,7 @@ def run(self):
     'sockets',
     'udp',
     'shm',
-    'psm3',
-    'ucx'
+    'psm3'
 ]
 daos_prov_list = [
     'verbs',
-Original file line number
+Diff line change
@@ Expand Up / @@ -115,8 +115,7 @@ def run(self): @@
         'sockets',
         'udp',
         'shm',
-        'psm3',
-        'ucx'
+        'psm3'
     ]
     daos_prov_list = [
         'verbs',
@@ Expand Down @@