RHOAIENG-69301: fix disconnected env job submission (main/3.5)

tests/e2e/mnist_raycluster_sdk_oauth_test.py

@@ -142,13 +142,10 @@ def assert_jobsubmit_withoutLogin(self, cluster):
             # API endpoint is directly under the hostname
             api_url = dashboard_url + "/api/jobs/"
 
+        job_spec = get_mnist_job_submission_spec()
         jobdata = {
-            "entrypoint": "python mnist.py",
-            "runtime_env": {
-                "working_dir": "./tests/e2e/",
-                "pip": "./tests/e2e/mnist_pip_requirements.txt",
-                "env_vars": get_setup_env_variables(),
-            },
+            "entrypoint": job_spec["entrypoint"],
+            "runtime_env": job_spec["runtime_env"],
         }
 
         # Try to submit a job without authentication
@@ -286,13 +283,11 @@ def assert_jobsubmit_withlogin(self, cluster):
                 "Verified: No jobs exist from the previous unauthenticated submission attempt."
             )
 
+        job_spec = get_mnist_job_submission_spec()
+        print(f"Submitting job: {job_spec['entrypoint']}")
         submission_id = client.submit_job(
-            entrypoint="python mnist.py",
-            runtime_env={
-                "working_dir": "./tests/e2e/",
-                "pip": "./tests/e2e/mnist_pip_requirements.txt",
-                "env_vars": get_setup_env_variables(),
-            },
+            entrypoint=job_spec["entrypoint"],
+            runtime_env=job_spec["runtime_env"],
             entrypoint_num_cpus=1,
         )
         print(f"Submitted job with ID: {submission_id}")
@@ -318,9 +313,10 @@ def assert_jobsubmit_withlogin(self, cluster):
         client.delete_job(submission_id)
 
     def assert_job_completion(self, status):
-        if status == "SUCCEEDED":
-            print(f"Job has completed: '{status}'")
+        status_value = getattr(status, "value", status)
+        if status_value == "SUCCEEDED":
+            print(f"Job has completed: '{status_value}'")
             assert True
         else:
-            print(f"Job has completed: '{status}'")
+            print(f"Job has completed: '{status_value}'")
             assert False

tests/e2e/support.py

@@ -157,6 +157,130 @@ def get_setup_env_variables(**kwargs):
     return env_vars
 
 
+def _env_flag_enabled(name):
+    return os.environ.get(name, "").strip().lower() in ("1", "true", "yes")
+
+
+def _env_flag_disabled(name):
+    return os.environ.get(name, "").strip().lower() in ("0", "false", "no", "off")
+
+
+def _disconnected_cluster_signals():
+    """Return True when the cluster is likely disconnected / air-gapped."""
+    if _env_flag_enabled("DISCONNECTED_CLUSTER") or _env_flag_enabled(
+        "IS_DISCONNECTED_CLUSTER"
+    ):
+        return True
+    try:
+        server = (run_oc_command(["whoami", "--show-server=true"]) or "").lower()
+        if "-dis-" in server or "disconnected" in server:
+            return True
+    except Exception:
+        pass
+    return False
+
+
+def _mnist_prerequisites_met():
+    """
+    Return True when full MNIST can run (pip packages + dataset reachable).
+
+    Connected labs may use public PyPI and MNIST mirrors with no extra env.
+    Disconnected labs need internal PIP_INDEX_URL and AWS_DEFAULT_ENDPOINT (MinIO).
+    """
+    if not _disconnected_cluster_signals():
+        return True
+
+    pip_url = (os.environ.get("PIP_INDEX_URL") or "").strip()
+    aws_endpoint = (os.environ.get("AWS_DEFAULT_ENDPOINT") or "").strip()
+    pip_ok = bool(pip_url) and "pypi.org" not in pip_url
+    aws_ok = bool(aws_endpoint)
+    if pip_ok and aws_ok:
+        print(
+            "Disconnected cluster with PIP mirror and S3 endpoint configured; "
+            "using full MNIST job"
+        )
+        return True
+    return False
+
+
+def use_smoke_job():
+    """
+    Use a lightweight Ray job when full MNIST is not viable.
+
+    Detection order (first match wins):
+      1. USE_SMOKE_JOB / UPGRADE_USE_SMOKE_JOB=true|false (explicit override)
+      2. Full MNIST prerequisites met (connected, or disconnected with mirrors)
+      3. DISCONNECTED_CLUSTER / IS_DISCONNECTED_CLUSTER env (Jenkins)
+      4. API server URL heuristic (-dis- / disconnected), last resort
+
+    ImageDigestMirrorSet / ICSP are intentionally not used: many connected
+    OpenShift clusters mirror container registries without blocking pip/PyPI.
+    """
+    for name in ("USE_SMOKE_JOB", "UPGRADE_USE_SMOKE_JOB"):
+        if _env_flag_enabled(name):
+            print(f"{name} enabled; using smoke job (no pip install)")
+            return True
+        if _env_flag_disabled(name):
+            print(f"{name} disabled; using full MNIST job")
+            return False
+
+    if _mnist_prerequisites_met():
+        return False
+
+    if _env_flag_enabled("DISCONNECTED_CLUSTER") or _env_flag_enabled(
+        "IS_DISCONNECTED_CLUSTER"
+    ):
+        print(
+            "Disconnected cluster env set without PIP/S3 mirrors; "
+            "using smoke job (no pip install)"
+        )
+        return True
+
+    try:
+        server = (run_oc_command(["whoami", "--show-server=true"]) or "").lower()
+        if "-dis-" in server or "disconnected" in server:
+            print(
+                "Detected disconnected cluster from API server URL; "
+                "using smoke job (no pip install)"
+            )
+            return True
+    except Exception:
+        pass
+
+    return False
+
+
+def use_upgrade_smoke_job():
+    """Backward-compatible alias for upgrade tests."""
+    return use_smoke_job()
+
+
+def get_mnist_job_submission_spec(**kwargs):
+    """Return entrypoint and runtime_env for tier1 / upgrade MNIST job submission tests."""
+    env_vars = get_setup_env_variables(**kwargs)
+    if use_smoke_job():
+        return {
+            "entrypoint": "python upgrade_job_smoke.py",
+            "runtime_env": {
+                "working_dir": "./tests/e2e/",
+                "env_vars": env_vars,
+            },
+        }
+    return {
+        "entrypoint": "python mnist.py",
+        "runtime_env": {
+            "working_dir": "./tests/e2e/",
+            "pip": "./tests/e2e/mnist_pip_requirements.txt",
+            "env_vars": env_vars,
+        },
+    }
+
+
+def get_upgrade_job_submission_spec():
+    """Backward-compatible alias for post-upgrade job submission tests."""
+    return get_mnist_job_submission_spec()
+
+
 def random_choice():
     alphabet = string.ascii_lowercase + string.digits
     return "".join(random.choices(alphabet, k=5))
@@ -694,18 +818,12 @@ def assert_get_cluster_and_jobsubmit(
         client = cluster.job_client
 
     # Submit a job and get the submission ID
-    env_vars = (
-        get_setup_env_variables(ACCELERATOR=accelerator)
-        if accelerator
-        else get_setup_env_variables()
-    )
+    spec_kwargs = {"ACCELERATOR": accelerator} if accelerator else {}
+    job_spec = get_mnist_job_submission_spec(**spec_kwargs)
+    print(f"Submitting job: {job_spec['entrypoint']}")
     submission_id = client.submit_job(
-        entrypoint="python mnist.py",
-        runtime_env={
-            "working_dir": "./tests/e2e/",
-            "pip": "./tests/e2e/mnist_pip_requirements.txt",
-            "env_vars": env_vars,
-        },
+        entrypoint=job_spec["entrypoint"],
+        runtime_env=job_spec["runtime_env"],
         entrypoint_num_cpus=1 if number_of_gpus is None else None,
         entrypoint_num_gpus=number_of_gpus,
     )

tests/e2e/upgrade_job_smoke.py

@@ -0,0 +1,16 @@
+# Copyright 2024 IBM, Red Hat
+#
+# Minimal Ray job for upgrade qualification on disconnected clusters.
+# Validates job submission and execution without pip installs or external datasets.
+
+import sys
+
+
+def main() -> int:
+    print("upgrade-job-smoke: job started")
+    print("upgrade-job-smoke: job finished successfully")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

tests/upgrade/01_raycluster_sdk_upgrade_test.py

@@ -174,13 +174,10 @@ def assert_jobsubmit_withoutLogin(self, cluster):
             # API endpoint is directly under the hostname
             api_url = dashboard_url + "/api/jobs/"
 
+        job_spec = get_upgrade_job_submission_spec()
         jobdata = {
-            "entrypoint": "python mnist.py",
-            "runtime_env": {
-                "working_dir": "./tests/e2e/",
-                "pip": "./tests/e2e/mnist_pip_requirements.txt",
-                "env_vars": get_setup_env_variables(),
-            },
+            "entrypoint": job_spec["entrypoint"],
+            "runtime_env": job_spec["runtime_env"],
         }
 
         # Try to submit a job without authentication
@@ -282,14 +279,12 @@ def assert_jobsubmit_withlogin(self, cluster):
         header = {"Authorization": f"Bearer {auth_token}"}
         client = RayJobClient(address=ray_dashboard, headers=header, verify=False)
 
-        # Submit the job
+        job_spec = get_upgrade_job_submission_spec()
+        print(f"Submitting upgrade job: {job_spec['entrypoint']}")
+
         submission_id = client.submit_job(
-            entrypoint="python mnist.py",
-            runtime_env={
-                "working_dir": "./tests/e2e/",
-                "pip": "./tests/e2e/mnist_pip_requirements.txt",
-                "env_vars": get_setup_env_variables(),
-            },
+            entrypoint=job_spec["entrypoint"],
+            runtime_env=job_spec["runtime_env"],
         )
         print(f"Submitted job with ID: {submission_id}")
         done = False
@@ -314,9 +309,10 @@ def assert_jobsubmit_withlogin(self, cluster):
         client.delete_job(submission_id)
 
     def assert_job_completion(self, status):
-        if status == "SUCCEEDED":
-            print(f"Job has completed: '{status}'")
+        status_value = getattr(status, "value", status)
+        if status_value == "SUCCEEDED":
+            print(f"Job has completed: '{status_value}'")
             assert True
         else:
-            print(f"Job has completed: '{status}'")
+            print(f"Job has completed: '{status_value}'")
             assert False

tests/upgrade/README.md

@@ -181,6 +181,23 @@ Post-upgrade job tests use `is_byoidc_cluster_detected()` from `tests/e2e/suppor
 
 On **htpasswd/LDAP** QE clusters, job tests use `oc whoami --show-token=true` with `OCP_ADMIN_USER_*` — not Keycloak password grant.
 
+## Disconnected clusters (post-upgrade job submission)
+
+Full MNIST installs torch via pip and may fetch datasets from the public internet. On disconnected clusters that fails unless mirrors are configured.
+
+Post-upgrade and tier1 MNIST job tests use `get_mnist_job_submission_spec()` from `tests/e2e/support.py`:
+
+| Job | When |
+|-----|------|
+| **`upgrade_job_smoke.py`** (no pip) | Disconnected / mirror-only cluster without PyPI + S3 env |
+| **`mnist.py`** (full pip + training) | Connected cluster, or disconnected with `PIP_INDEX_URL` (non-pypi.org) **and** `AWS_DEFAULT_ENDPOINT` set |
+
+**Detection order:** `USE_SMOKE_JOB` / `UPGRADE_USE_SMOKE_JOB` override → MNIST prerequisites → `DISCONNECTED_CLUSTER` / `IS_DISCONNECTED_CLUSTER` (Jenkins) → API URL `-dis-` heuristic (last resort).
+
+Registry image mirrors (`ImageDigestMirrorSet` / `ICSP`) are **not** used for detection — connected OpenShift clusters often have them without blocking pip.
+
+Set `USE_SMOKE_JOB=false` or `UPGRADE_USE_SMOKE_JOB=false` to force full MNIST when mirrors are configured. Set `DISCONNECTED_CLUSTER=true` in the test env file for explicit lab config (aligns with Jenkins `IS_DISCONNECTED_CLUSTER`).
+
 ---
 
 ## Migration script