clone aotc and get metrics from gcs and set pythonpath

.github/CODEOWNERS

@@ -16,3 +16,5 @@ dags/sparsity_diffusion_devx/project_bite* @RissyRan @parambole @jiangjy1982 @ai
 dags/sparsity_diffusion_devx/configs/project_bite* @RissyRan @parambole @jiangjy1982 @aireenmei @michelle-yooh @jiya-zhang
 
 dags/inference @yeandy @vipannalla @morgandu @mailvijayasingh @sixiang-google @joezijunzhou @singh-mitali
+
+dags/map_reproducibility @crankshaw-google @polydier1

.github/requirements.txt

@@ -10,3 +10,4 @@ tensorflow-cpu
 kubernetes
 pyarrow
 apache-airflow-providers-google
+dacite

dags/map_reproducibility/nemo_gpt3.py

@@ -15,21 +15,29 @@
 """DAGs to run Aotc reproducibility benchmarks."""
 
 import datetime
+import sys
+import tempfile
+
 from airflow import models
 from airflow.decorators import task
 from airflow.hooks.subprocess import SubprocessHook
 from dags import composer_env
-from dags.map_reproducibility.aotc_reproducibility import get_metrics_cmds
-from dags.map_reproducibility.aotc_reproducibility import set_variables_cmds
-from dags.map_reproducibility.aotc_reproducibility import configure_project_and_cluster
-from dags.map_reproducibility.aotc_reproducibility import install_helm_cmds
-from dags.map_reproducibility.aotc_reproducibility import namespace_cmds
-from dags.map_reproducibility.aotc_reproducibility import wait_for_jobs_cmds
-from dags.map_reproducibility.aotc_reproducibility import copy_bucket_cmds
-from dags.map_reproducibility.aotc_reproducibility import cleanup_cmds
-from dags.map_reproducibility.aotc_reproducibility import git_cookie_authdaemon
-from dags.map_reproducibility.aotc_reproducibility import clone_gob
-from dags.map_reproducibility.aotc_reproducibility import helm_install_cmds
+from dags.map_reproducibility.utils import get_metrics_cmds
+from dags.map_reproducibility.utils import set_variables_cmds
+from dags.map_reproducibility.utils import configure_project_and_cluster
+from dags.map_reproducibility.utils import install_helm_cmds
+from dags.map_reproducibility.utils import namespace_cmds
+from dags.map_reproducibility.utils import wait_for_jobs_cmds
+from dags.map_reproducibility.utils import copy_bucket_cmds
+from dags.map_reproducibility.utils import cleanup_cmds
+from dags.map_reproducibility.utils import git_cookie_authdaemon
+from dags.map_reproducibility.utils import clone_gob
+from dags.map_reproducibility.utils import helm_install_cmds
+from dags.map_reproducibility.utils import get_metrics_from_gcs
+from dags.map_reproducibility.utils import get_aotc_repo
+from dags.map_reproducibility.utils import extract_bucket_file_name
+from dags.map_reproducibility.utils import extract_python_path
+
 
 # Run once a day at 2 pm UTC (6 am PST)
 SCHEDULED_TIME = "0 14 * * *" if composer_env.is_prod_env() else None
@@ -50,29 +58,40 @@ def run_aotc_workload():
       "export JOB_NAME=gpt3-xlml-$NOW-175b-nemo",
   )
 
-  hook = SubprocessHook()
-  result = hook.run_command(
-      [
-          "bash",
-          "-c",
-          ";".join(
-              set_variables_cmds()
-              + configure_project_and_cluster()
-              + git_cookie_authdaemon()
-              + clone_gob()
-              + gpu_recipe_cmd
-              + install_helm_cmds()
-              + namespace_cmds()
-              + workload_cmds
-              + helm_install_cmds()
-              + wait_for_jobs_cmds()
-              + copy_bucket_cmds()
-              + get_metrics_cmds()
-              + cleanup_cmds()
-          ),
-      ],
-  )
-  assert result.exit_code == 0, f"Command failed with code {result.exit_code}"
+  with tempfile.TemporaryDirectory() as tmpdir:
+    hook = SubprocessHook()
+    result = hook.run_command(
+        [
+            "bash",
+            "-c",
+            ";".join(
+                set_variables_cmds()
+                + configure_project_and_cluster()
+                + git_cookie_authdaemon()
+                + clone_gob()
+                + gpu_recipe_cmd
+                + install_helm_cmds()
+                + namespace_cmds()
+                + workload_cmds
+                + helm_install_cmds()
+                + wait_for_jobs_cmds()
+                + copy_bucket_cmds()
+                + get_metrics_cmds()
+                + cleanup_cmds()
+                + get_aotc_repo()
+            ),
+        ],
+        cwd=tmpdir,
+    )
+    assert result.exit_code == 0, f"Command failed with code {result.exit_code}"
+
+    # Extract COMPLETE_JOB_NAME from the output
+    bucket_name, file_name, python_path = extract_bucket_file_name(
+        result.output
+    )
+    get_metrics_from_gcs(bucket_name, file_name)
+
+    sys.path.append(python_path)
 
 
 with models.DAG(

dags/map_reproducibility/aotc_reproducibility.py → dags/map_reproducibility/utils.py

@@ -14,7 +14,8 @@
 
 "Bash helper commands for AOTC artifacts"
 
-import os
+import re
+from google.cloud import storage
 
 
 def set_variables_cmds():
@@ -39,12 +40,16 @@ def configure_project_and_cluster():
 
 
 # This is required to get auth to access
-# internal GoB repo
 def git_cookie_authdaemon():
   auth_cmds = (
       "git clone https://gerrit.googlesource.com/gcompute-tools",
       "echo 'trying to run git-cookie-authdaemon'",
-      "./gcompute-tools/git-cookie-authdaemon",
+      # Check if the daemon is already running
+      "if pgrep -f git-cookie-authdaemon; then "
+      "  echo 'git-cookie-authdaemon is already running'; "
+      "else "
+      "  ./gcompute-tools/git-cookie-authdaemon || echo 'Error running git-cookie-authdaemon'; "  # Run if not running
+      "fi",
   )
   return auth_cmds
 
@@ -56,6 +61,7 @@ def clone_gob():
       "reproducible-benchmark-recipes",
       "cd reproducible-benchmark-recipes/projects",
       "cd gpu-recipes",
+      "pwd",
   )
   return gob_clone_cmds
 
@@ -101,9 +107,6 @@ def helm_install_cmds():
 
 def wait_for_jobs_cmds():
   wait_for_job = (
-      "echo 'will wait for job to start running'",
-      "kubectl wait --for=condition=running job/$JOB_NAME"
-      " --namespace=default --timeout=10m",
       "echo 'will wait for jobs to finish'",
       "kubectl wait --for=condition=complete "
       "job/$JOB_NAME --namespace=default --timeout=100m",
@@ -113,10 +116,9 @@ def wait_for_jobs_cmds():
 
 def copy_bucket_cmds():
   copy_bucket_contents = (
-      "COMPLETE_JOB_NAME=$(gcloud storage ls "
+      "export COMPLETE_JOB_NAME=$(gcloud storage ls "
       "gs://$BUCKET_NAME/nemo-experiments/ | grep $JOB_NAME)",
-      "echo 'copying from' ",
-      "echo $COMPLETE_JOB_NAME",
+      'echo "COMPLETE_JOB_NAME ${COMPLETE_JOB_NAME}"',
       "cd $REPO_ROOT/src/utils/training_metrics",
       "gcloud storage cp ${COMPLETE_JOB_NAME}"
       "dllogger/rank-0/dllogger.json .",
@@ -127,21 +129,98 @@ def copy_bucket_cmds():
 def get_metrics_cmds():
   # TODO(gunjanj007): get these parameters from the recipe
   get_metrics = (
+      "METRICS_FILE=$COMPLETE_JOB_NAME/metrics.txt",
       "python3 process_training_results.py --file"
       " dllogger.json --batch_size 2048 "
       "--num_accelerators 256 "
       "--precision fp8  "
       "--model_type gpt3-175b "
-      "--accelerator_type h100 ",
+      "--accelerator_type h100 | "
+      "gsutil cp - $METRICS_FILE",
+      'echo "METRICS_FILE=${METRICS_FILE}"',
   )
   return get_metrics
 
 
+def get_aotc_repo():
+  gob_clone_cmds = (
+      "echo 'trying to clone GoB aotc repo'",
+      "git clone https://cmcs-perf-tooling-internal.googlesource.com/"
+      "benchmark-automation",
+      "cd benchmark-automation/aotc/src",
+      "export PYTHONPATH=$PWD",
+      'echo "PYTHONPATH=$PYTHONPATH and METRICS_FILE=$METRICS_FILE"',
+  )
+  return gob_clone_cmds
+
+
 def cleanup_cmds():
   cleanup = (
+      "cd $REPO_ROOT",
+      "cd ../../..",
       "kubectl get pods "
       "--no-headers=true | awk '{print $1}' "
       "| grep $JOB_NAME |  xargs kubectl delete pods",
       "helm uninstall $JOB_NAME",
   )
   return cleanup
+
+
+def get_metrics_from_gcs(bucket_name, file_name):
+  # Initialize GCS and BigQuery clients
+  storage_client = storage.Client()
+
+  # Get the bucket and file
+  bucket = storage_client.bucket(bucket_name)
+  blob = bucket.blob(file_name)
+
+  # Download the file content
+  metrics_output = blob.download_as_string().decode("utf-8")
+
+  # Parse the metrics (adjust based on your file format)
+  lines = metrics_output.splitlines()
+  average_step_time = float(lines[0].split(": ")[1])
+  tflops_per_accelerator = float(lines[1].split(": ")[1])
+  mfu = float(lines[2].split(": ")[1])
+
+  print(f"Average Step Time: {average_step_time}")
+  print(f"TFLOPS/Accelerator: {tflops_per_accelerator}")
+  print(f"MFU: {mfu}")
+
+
+def extract_bucket_file_name(last_line):
+  metrics_file = None
+
+  # We match here because subprocesshook only outputs the last line.
+  match = re.search(r"PYTHONPATH=(.*?)\s+METRICS_FILE=(.*)", last_line)
+  if match:
+    python_path = match.group(1)
+    metrics_file = match.group(2)
+    print(f"PYTHONPATH in python: {python_path}")
+    print(f"METRICS_FILE: {metrics_file}")
+  else:
+    print("Error: Could not extract PYTHONPATH and METRICS_FILE")
+  print(f"Metrics file name: {metrics_file}")
+  if metrics_file:
+    # Extract bucket_name and file_name
+    bucket_name = re.search(r"gs://([^/]+)/", metrics_file).group(1)
+    file_name = re.search(r"gs://[^/]+/(.+)", metrics_file).group(1)
+
+    print(f"Bucket name: {bucket_name}")
+    print(f"File name: {file_name}")
+  else:
+    print("Metrics file not found in the output.")
+
+  return bucket_name, file_name, python_path
+
+
+def extract_python_path(bash_result_output):
+  python_path = None
+  for line in bash_result_output.splitlines():
+    if line.startswith("PYTHONPATH"):
+      python_path = line.split("=", 1)[1]
+      break
+
+    print(f"Pyhon path name: {python_path}")
+
+  return python_path