Merge branch 'master' into fix_gpupod_check

red-hat-data-services · Oct 2, 2024 · 563a939 · 563a939
2 parents e4c68ba + e8f3883
commit 563a939
Show file tree

Hide file tree

Showing 13 changed files with 386 additions and 31 deletions.
diff --git a/ods_ci/libs/DataSciencePipelinesKfp.py b/ods_ci/libs/DataSciencePipelinesKfp.py
@@ -282,6 +282,12 @@ def wait_for_run_completion(self, run_id, timeout=160, sleep_duration=5):
         response = self.client.wait_for_run_completion(run_id=run_id, timeout=timeout, sleep_duration=sleep_duration)
         return response.state
 
+    @keyword
+    def get_run_status(self, run_id):
+        """###Gets run status"""
+        response = self.client.get_run(run_id)
+        return response.state
+
     @keyword
     def check_run_status(self, run_id, timeout=160):
         """Waits for a run to complete"""

diff --git a/ods_ci/tests/Resources/CLI/DataSciencePipelines/DataSciencePipelinesBackend.resource b/ods_ci/tests/Resources/CLI/DataSciencePipelines/DataSciencePipelinesBackend.resource
@@ -1,10 +1,11 @@
 *** Settings ***
 Documentation    Collection of keywords to interact with Data Science Pipelines via CLI
-Library          OperatingSystem
 Library          String
 Library          ../../../../libs/DataSciencePipelinesAPI.py
 Library          ../../../../libs/DataSciencePipelinesKfp.py
 Resource         ../../../Resources/OCP.resource
+Resource         ../../../Resources/Common.robot
+
 
 *** Variables ***
 ${DSPA_PATH}=      tests/Resources/Files/pipeline-samples/v2/dspa
@@ -42,10 +43,16 @@ Create Pipeline Server
     ...    -p OBJECT_STORAGE_REGION=${object_storage_region}
     ...    -p OBJECT_STORAGE_BUCKET=${object_storage_bucket_name}
 
-    Run      oc process -f ${DSPA_PATH}/dspa-template.yaml ${template_parameters} | oc apply -n ${namespace} -f -
+    Run And Verify Command    oc process -f ${DSPA_PATH}/dspa-template.yaml ${template_parameters} | oc apply -n ${namespace} -f -    # robocop: off=line-too-long
 
     IF  ${configure_pip_index}   Create Pipelines ConfigMap With Custom Pip Index Url And Trusted Host  ${namespace}
 
+Get DSP Version
+    [Documentation]    Returns dspVersion of the DSPA deployed in ${namespace}
+    [Arguments]     ${namespace}
+    ${dsp_version}=    Run And Verify Command    oc get datasciencepipelinesapplications -n ${namespace} -o json | jq -r '.items[0].spec.dspVersion'    # robocop: off=line-too-long
+    RETURN    ${dsp_version}
+
 # robocop: disable:line-too-long
 Create PipelineServer Using Custom DSPA
     [Documentation]    Install and verifies that DataSciencePipelinesApplication CRD is installed and working
@@ -55,21 +62,25 @@ Create PipelineServer Using Custom DSPA
     [Arguments]     ${namespace}    ${dspa_file}=data-science-pipelines-sample.yaml
     ...    ${assert_install}=${TRUE}    ${configure_pip_index}=${TRUE}
 
-    Run     oc apply -f "${DSPA_PATH}/${dspa_file}" -n ${namespace}
+    Run And Verify Command     oc apply -f "${DSPA_PATH}/${dspa_file}" -n ${namespace}
     IF    ${assert_install}==True
-        ${generation_value}    Run    oc get datasciencepipelinesapplications -n ${namespace} -o json | jq '.items[0].metadata.generation'
+        ${generation_value}=    Run And Verify Command    oc get datasciencepipelinesapplications -n ${namespace} -o json | jq '.items[0].metadata.generation'    # robocop: off=line-too-long
         Should Be True    ${generation_value} == 2    DataSciencePipelinesApplication created
     END
 
     IF  ${configure_pip_index}   Create Pipelines ConfigMap With Custom Pip Index Url And Trusted Host  ${namespace}
 
 Verify Pipeline Server Deployments    # robocop: disable
-    [Documentation]    Verifies the correct deployment of DS Pipelines in the rhods namespace
+    [Documentation]    Verifies the correct deployment of a DSPv2 DataSciencePipelineApplication
     [Arguments]    ${namespace}
 
     @{all_pods}=  Oc Get    kind=Pod    namespace=${namespace}
     ...    label_selector=component=data-science-pipelines
-    Run Keyword And Continue On Failure    Length Should Be    ${all_pods}    7
+
+    ${pods_count}=   Get Length    ${all_pods}
+    IF    ${pods_count} < 7
+         Fail    DSPA requires at least 7 pods running in the namespace
+    END
 
     @{pipeline_api_server}=  Oc Get    kind=Pod    namespace=${namespace}
     ...    label_selector=app=ds-pipeline-dspa
@@ -106,12 +117,52 @@ Verify Pipeline Server Deployments    # robocop: disable
     ${containerNames}=  Create List  mariadb
     Verify Deployment    ${mariadb}  1  1  ${containerNames}
 
+Verify DSPv1 Pipeline Server Deployments
+    [Documentation]    Verifies the correct deployment of a DSPv1 DataSciencePipelineApplication
+    [Arguments]    ${namespace}
+
+    @{all_pods}=  Oc Get    kind=Pod    namespace=${namespace}
+    ...    label_selector=component=data-science-pipelines
+
+    ${pods_count}=   Get Length    ${all_pods}
+    IF    ${pods_count} < 4
+         Fail    DSPA requires at least 4 pods running in the namespace
+    END
+
+    @{pipeline_api_server}=  Oc Get    kind=Pod    namespace=${namespace}
+    ...    label_selector=app=ds-pipeline-dspa
+    ${containerNames}=  Create List  oauth-proxy    ds-pipeline-api-server
+    Verify Deployment    ${pipeline_api_server}  1  2  ${containerNames}
+
+    @{pipeline_persistenceagent}=  Oc Get    kind=Pod    namespace=${namespace}
+    ...    label_selector=app=ds-pipeline-persistenceagent-dspa
+    ${containerNames}=  Create List  ds-pipeline-persistenceagent
+    Verify Deployment    ${pipeline_persistenceagent}  1  1  ${containerNames}
+
+    @{pipeline_scheduledworkflow}=  Oc Get    kind=Pod    namespace=${namespace}
+    ...    label_selector=app=ds-pipeline-scheduledworkflow-dspa
+    ${containerNames}=  Create List  ds-pipeline-scheduledworkflow
+    Verify Deployment    ${pipeline_scheduledworkflow}  1  1  ${containerNames}
+
+    @{mariadb}=  Oc Get    kind=Pod    namespace=${namespace}
+    ...    label_selector=app=mariadb-dspa
+    ${containerNames}=  Create List  mariadb
+    Verify Deployment    ${mariadb}  1  1  ${containerNames}
+
 Wait Until Pipeline Server Is Deployed
     [Documentation]    Waits until all the expected pods of the pipeline server
     ...                are running
     [Arguments]    ${namespace}
-    Wait Until Keyword Succeeds    10 times    10s
-    ...    Verify Pipeline Server Deployments    namespace=${namespace}
+
+    ${dspVersion}=    Get DSP Version    ${namespace}
+    IF    "${dspVersion}" == "v2"
+            Wait Until Keyword Succeeds    10 times    10s
+            ...    Verify Pipeline Server Deployments    namespace=${namespace}
+    ELSE
+             Wait Until Keyword Succeeds    10 times    10s
+            ...    Verify DSPv1 Pipeline Server Deployments    namespace=${namespace}
+    END
+
 
 Wait Until Pipeline Server Is Deleted
     [Documentation]    Waits until all pipeline server pods are deleted
@@ -128,13 +179,13 @@ Create Pipelines ConfigMap With Custom Pip Index Url And Trusted Host
     [Documentation]     Creates a Configmap (ds-pipeline-custom-env-vars) in the project,
     ...    storing the values for pip_index_url and pip_trusted_host
     [Arguments]    ${namespace}
-    Run    oc create configmap ds-pipeline-custom-env-vars -n ${namespace} --from-literal=pip_index_url=${PIP_INDEX_URL} --from-literal=pip_trusted_host=${PIP_TRUSTED_HOST}
+    Run And Verify Command    oc create configmap ds-pipeline-custom-env-vars -n ${namespace} --from-literal=pip_index_url=${PIP_INDEX_URL} --from-literal=pip_trusted_host=${PIP_TRUSTED_HOST}    # robocop: off=line-too-long
 
 Create Secret With Pipelines Object Storage Information
     [Documentation]     Creates a secret needed to create a pipeline server containing the object storage credentials
     [Arguments]    ${namespace}    ${object_storage_access_key}    ${object_storage_secret_key}
-    Run    oc create secret generic dashboard-dspa-secret -n ${namespace} --from-literal=AWS_ACCESS_KEY_ID=${object_storage_access_key} --from-literal=AWS_SECRET_ACCESS_KEY=${object_storage_secret_key}
-    Run    oc label secret dashboard-dspa-secret -n ${namespace} opendatahub.io/dashboard=true
+    Run And Verify Command    oc create secret generic dashboard-dspa-secret -n ${namespace} --from-literal=AWS_ACCESS_KEY_ID=${object_storage_access_key} --from-literal=AWS_SECRET_ACCESS_KEY=${object_storage_secret_key}    # robocop: off=line-too-long
+    Run And Verify Command    oc label secret dashboard-dspa-secret -n ${namespace} opendatahub.io/dashboard=true
 
 
 Import Pipeline And Create Run
@@ -164,6 +215,20 @@ Import Pipeline And Create Run
 
     RETURN    ${pipeline_id}    ${pipeline_version_id}    ${pipeline_run_id}    ${experiment_id}
 
+Verify Run Status
+    [Documentation]  Verifies pipeline run status matches ${pipeline_run_expected_status}
+    [Arguments]    ${namespace}           ${username}                  ${password}
+    ...    ${pipeline_run_id}    ${pipeline_run_expected_status}="SUCCEEDED"
+
+    DataSciencePipelinesKfp.Setup Client    user=${username}    pwd=${password}    project=${namespace}
+
+    ${pipeline_run_status}=    DataSciencePipelinesKfp.Get Run Status    run_id=${pipeline_run_id}
+    IF   "${pipeline_run_status}" != "${pipeline_run_expected_status}"
+        ${error_msg}=    Catenate    Expected pipeline status was ${pipeline_run_expected_status} but pipeline run
+        ...    has status=${pipeline_run_status}
+        Fail    ${error_msg}
+    END
+
 Wait For Run Completion And Verify Status
     [Documentation]
     [Arguments]    ${namespace}           ${username}                  ${password}
@@ -175,11 +240,8 @@ Wait For Run Completion And Verify Status
     ${pipeline_run_status}=    DataSciencePipelinesKfp.Wait For Run Completion    run_id=${pipeline_run_id}
     ...    timeout=${pipeline_run_timeout}     sleep_duration=${5}
 
-    IF   "${pipeline_run_status}" != "${pipeline_run_expected_status}"
-        ${error_msg}=    Catenate    Expected pipeline status was ${pipeline_run_expected_status} but pipeline run
-        ...    finished with status=${pipeline_run_status}
-        Fail    ${error_msg}
-    END
+    Verify Run Status   namespace=${namespace}    username=${username}    password=${password}
+    ...    pipeline_run_id=${pipeline_run_id}    pipeline_run_expected_status=${pipeline_run_expected_status}
 
     RETURN    ${pipeline_run_status}
 
@@ -207,5 +269,3 @@ Delete Pipeline And Related Resources
     END
 
     DataSciencePipelinesKfp.Delete Pipeline    ${pipeline_id}
-
-
diff --git a/ods_ci/tests/Resources/CLI/DataSciencePipelines/DataSciencePipelinesUpgradeTesting.resource b/ods_ci/tests/Resources/CLI/DataSciencePipelines/DataSciencePipelinesUpgradeTesting.resource
@@ -0,0 +1,64 @@
+*** Settings ***
+Documentation    Upgrade Testing Keywords
+Resource         DataSciencePipelinesBackend.resource
+Resource         ../../../Resources/Page/ODH/ODHDashboard/ODHDataScienceProject/Projects.resource
+
+
+*** Variables ***
+${PROJECT}=    dsp-upgrade-testing
+${PIPELINE_LONGRUNNING_FILEPATH}=    tests/Resources/Files/pipeline-samples/v2/pip_index_url/take_nap_pipeline_root_compiled.yaml  # robocop: disable:line-too-long
+
+
+*** Keywords ***
+Setup Environment For Upgrade Testing
+    [Documentation]    Creates project dsp-test-upgrade and sets up the resources to test during upgrade:
+    ...    - Creates a pipeline server
+    ...    - Starts a pipeline that will run for 1h
+
+    Create Project And Configure Pipeline Server    ${PROJECT}
+    Start Long Running Pipeline    ${PROJECT}
+
+Verify Resources After Upgrade
+    [Documentation]    Verifies the status of the resources created in ${PROJECT} after the upgrade
+    ...    Deletes ${PROJECT} if all verifications are correct (leaving for debugging purposes if not)
+
+    DataSciencePipelinesBackend.Wait Until Pipeline Server Is Deployed    namespace=${PROJECT}
+
+    Verify Run Status
+    ...    namespace=${PROJECT}    username=${TEST_USER.USERNAME}    password=${TEST_USER.PASSWORD}
+    ...    pipeline_run_id=${DSP_LONGRUNNING_PIPELINE_RUN_ID}    pipeline_run_expected_status=RUNNING
+
+    Projects.Delete Project Via CLI By Display Name    ${PROJECT}
+
+Create Project And Configure Pipeline Server
+    [Documentation]    Creates a data science pipelines project ${namespace}  (deleting existing one if needed),
+    ...    configures a pipeline server using the default configuration and waits until the server is running
+    [Arguments]    ${namespace}
+    Projects.Delete Project Via CLI By Display Name    ${namespace}
+    Projects.Create Data Science Project From CLI    ${namespace}
+    DataSciencePipelinesBackend.Create Pipeline Server    namespace=${namespace}
+    ...    object_storage_access_key=${S3.AWS_ACCESS_KEY_ID}
+    ...    object_storage_secret_key=${S3.AWS_SECRET_ACCESS_KEY}
+    ...    object_storage_endpoint=${S3.BUCKET_2.ENDPOINT}
+    ...    object_storage_region=${S3.BUCKET_2.REGION}
+    ...    object_storage_bucket_name=${S3.BUCKET_2.NAME}
+    ...    dsp_version=v2
+    DataSciencePipelinesBackend.Wait Until Pipeline Server Is Deployed    namespace=${namespace}
+
+Start Long Running Pipeline
+    [Documentation]    Imports and creates a run of a long running pipeline
+    [Arguments]    ${namespace}
+
+    ${pipeline_run_params}=    Create Dictionary    naptime_secs=${3600}
+
+    # robocop:off=unused-variable
+    ${pipeline_id}    ${pipeline_version_id}    ${pipeline_run_id}    ${experiment_id}=
+    ...    DataSciencePipelinesBackend.Import Pipeline And Create Run
+    ...    namespace=${namespace}    username=${TEST_USER.USERNAME}    password=${TEST_USER.PASSWORD}
+    ...    pipeline_name=take-nap
+    ...    pipeline_description=A pipeline that runs for 1h and prints a message
+    ...    pipeline_package_path=${PIPELINE_LONGRUNNING_FILEPATH}
+    ...    pipeline_run_name=take-nap-run
+    ...    pipeline_run_params=${pipeline_run_params}
+
+    Set Global Variable  ${DSP_LONGRUNNING_PIPELINE_RUN_ID}    ${pipeline_run_id}
diff --git a/ods_ci/tests/Resources/CLI/ModelRegistry/samples/istio/components/istio.env b/ods_ci/tests/Resources/CLI/ModelRegistry/samples/istio/components/istio.env
@@ -1,4 +1,4 @@
-AUTH_PROVIDER=opendatahub-auth-provider
+AUTH_PROVIDER=
 DOMAIN=
 ISTIO_INGRESS=ingressgateway
 REST_CREDENTIAL_NAME=modelregistry-sample-rest-credential

diff --git a/ods_ci/tests/Resources/Common.robot b/ods_ci/tests/Resources/Common.robot
@@ -437,13 +437,13 @@ Extract URLs From Text
     RETURN    ${urls}
 
 Run And Verify Command
-    [Documentation]    Run and verify shell command 
-    [Arguments]    ${command}    ${print_to_log}=${TRUE}
+    [Documentation]    Run and verify shell command
+    [Arguments]    ${command}    ${print_to_log}=${TRUE}    ${expected_rc}=${0}
     ${result}=    Run Process    ${command}    shell=yes    stderr=STDOUT
     IF    ${print_to_log}    Log    ${result.stdout}     console=True
-    Should Be True    ${result.rc} == 0
+    Should Be True    ${result.rc} == ${expected_rc}
     RETURN    ${result.stdout}
-
+  
 Run And Watch Command
   [Documentation]    Run any shell command (including args) with optional:
   ...    Timeout: 10 minutes by default.

diff --git a/.../tests/Resources/Files/pipeline-samples/v2/gpu/pytorch/pytorch_verify_gpu_availability.py b/.../tests/Resources/Files/pipeline-samples/v2/gpu/pytorch/pytorch_verify_gpu_availability.py
@@ -22,7 +22,12 @@ def add_gpu_toleration(task: PipelineTask, accelerator_type: str, accelerator_li
     kubernetes.add_toleration(task, key=accelerator_type, operator="Exists", effect="NoSchedule")
 
 
-@dsl.component(base_image=common_base_image, packages_to_install=["torch"], pip_index_urls=["$PIP_INDEX_URL"])
+@dsl.component(
+    base_image=common_base_image,
+    packages_to_install=["torch"],
+    pip_index_urls=["$PIP_INDEX_URL"],
+    pip_trusted_hosts=["$PIP_TRUSTED_HOST"],
+)
 def verify_gpu_availability(gpu_toleration_added: bool):
     import torch  # noqa: PLC0415
 

diff --git a/...urces/Files/pipeline-samples/v2/gpu/pytorch/pytorch_verify_gpu_availability_compiled.yaml b/...urces/Files/pipeline-samples/v2/gpu/pytorch/pytorch_verify_gpu_availability_compiled.yaml
@@ -29,10 +29,10 @@ deploymentSpec:
         - "\nif ! [ -x \"$(command -v pip)\" ]; then\n    python3 -m ensurepip ||\
           \ python3 -m ensurepip --user || apt-get install python3-pip\nfi\n\nPIP_DISABLE_PIP_VERSION_CHECK=1\
           \ python3 -m pip install --quiet --no-warn-script-location --index-url $PIP_INDEX_URL\
-          \ --trusted-host $PIP_INDEX_URL 'kfp==2.9.0' '--no-deps' 'typing-extensions>=3.7.4,<5;\
+          \ --trusted-host $PIP_TRUSTED_HOST 'kfp==2.9.0' '--no-deps' 'typing-extensions>=3.7.4,<5;\
           \ python_version<\"3.9\"'  &&  python3 -m pip install --quiet --no-warn-script-location\
-          \ --index-url $PIP_INDEX_URL --trusted-host $PIP_INDEX_URL 'torch' && \"\
-          $0\" \"$@\"\n"
+          \ --index-url $PIP_INDEX_URL --trusted-host $PIP_TRUSTED_HOST 'torch' &&\
+          \ \"$0\" \"$@\"\n"
         - sh
         - -ec
         - 'program_path=$(mktemp -d)
@@ -72,10 +72,10 @@ deploymentSpec:
         - "\nif ! [ -x \"$(command -v pip)\" ]; then\n    python3 -m ensurepip ||\
           \ python3 -m ensurepip --user || apt-get install python3-pip\nfi\n\nPIP_DISABLE_PIP_VERSION_CHECK=1\
           \ python3 -m pip install --quiet --no-warn-script-location --index-url $PIP_INDEX_URL\
-          \ --trusted-host $PIP_INDEX_URL 'kfp==2.9.0' '--no-deps' 'typing-extensions>=3.7.4,<5;\
+          \ --trusted-host $PIP_TRUSTED_HOST 'kfp==2.9.0' '--no-deps' 'typing-extensions>=3.7.4,<5;\
           \ python_version<\"3.9\"'  &&  python3 -m pip install --quiet --no-warn-script-location\
-          \ --index-url $PIP_INDEX_URL --trusted-host $PIP_INDEX_URL 'torch' && \"\
-          $0\" \"$@\"\n"
+          \ --index-url $PIP_INDEX_URL --trusted-host $PIP_TRUSTED_HOST 'torch' &&\
+          \ \"$0\" \"$@\"\n"
         - sh
         - -ec
         - 'program_path=$(mktemp -d)

diff --git a/ods_ci/tests/Resources/Files/pipeline-samples/v2/pip_index_url/take_nap_pipeline_root.py b/ods_ci/tests/Resources/Files/pipeline-samples/v2/pip_index_url/take_nap_pipeline_root.py
@@ -0,0 +1,42 @@
+from kfp import compiler, dsl, kubernetes
+from kfp.dsl import PipelineTask
+
+common_base_image = (
+    "registry.redhat.io/ubi8/python-39@sha256:3523b184212e1f2243e76d8094ab52b01ea3015471471290d011625e1763af61"
+)
+
+
+def add_pip_index_configuration(task: PipelineTask):
+    kubernetes.use_config_map_as_env(
+        task,
+        config_map_name="ds-pipeline-custom-env-vars",
+        config_map_key_to_env={"pip_index_url": "PIP_INDEX_URL", "pip_trusted_host": "PIP_TRUSTED_HOST"},
+    )
+
+
+@dsl.component(base_image=common_base_image)
+def take_nap(naptime_secs: int) -> str:
+    """Sleeps for secs"""
+    from time import sleep  # noqa: PLC0415
+
+    print(f"Sleeping for {naptime_secs} seconds: Zzzzzz ...")
+    sleep(naptime_secs)
+    return "I'm awake now. Did I snore?"
+
+
+@dsl.component(base_image=common_base_image)
+def wake_up(message: str):
+    """Wakes up from nap printing a message"""
+    print(message)
+
+
+@dsl.pipeline(name="take-nap-pipeline", description="Pipeline that sleeps for 15 mins (900 secs)")
+def take_nap_pipeline(naptime_secs: int = 900):
+    take_nap_task = take_nap(naptime_secs=naptime_secs).set_caching_options(False)
+    add_pip_index_configuration(take_nap_task)
+    wake_up_task = wake_up(message=take_nap_task.output).set_caching_options(False)
+    add_pip_index_configuration(wake_up_task)
+
+
+if __name__ == "__main__":
+    compiler.Compiler().compile(take_nap_pipeline, package_path=__file__.replace(".py", "_compiled.yaml"))