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 · Sep 28, 2023 · 0e0e4f6 · 0e0e4f6
1 parent c9020f9
commit 0e0e4f6
Show file tree

Hide file tree

Showing 3 changed files with 30 additions and 10 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
@@ -28,6 +28,8 @@ def build_libfabric(libfab_install_path, mode, cluster=None, ucx=None):
         prov_list = common.daos_prov_list
     elif (cluster == 'gpu'):
         prov_list = common.gpu_prov_list
+    elif (ucx):
+        prov_list = common.ucx_prov_list
     else:
         prov_list = common.default_prov_list
 
@@ -41,7 +43,7 @@ def build_libfabric(libfab_install_path, mode, cluster=None, ucx=None):
     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}')
 
@@ -186,13 +188,15 @@ 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*')
 
     if (build_item == 'libfabric'):
             build_libfabric(libfab_install_path, ofi_build_mode, cluster, ucx)
     elif (build_item == 'libfabric_mpich'):
+            workspace += '/mpich'
             build_libfabric(f'{libfab_install_path}/libfabric_mpich',
                             ofi_build_mode, cluster)
     elif (build_item == 'mpich'):

diff --git a/contrib/intel/jenkins/common.py b/contrib/intel/jenkins/common.py
@@ -118,6 +118,9 @@ def run(self):
     'psm3',
     'ucx'
 ]
+ucx_prov_list = [
+    'ucx'
+]
 daos_prov_list = [
     'verbs',
     'tcp'
-Original file line number
+Diff line change
@@ Expand Up / @@ -118,6 +118,9 @@ def run(self): @@
         'psm3',
         'ucx'
     ]
+    ucx_prov_list = [
+        'ucx'
+    ]
     daos_prov_list = [
         'verbs',
         'tcp'
@@ Expand Down @@