import json
import numpy as np
import pandas as pd

import pytz
import datetime

import requests
from requests.auth import HTTPBasicAuth

import wallaroo

wl = wallaroo.Client()

workspace = wl.get_workspace(name="metric-retrieval-tutorial", create_if_not_exist=True)
wl.set_current_workspace(workspace)

{'name': 'metric-retrieval-tutorial', 'id': 1713, 'archived': False, 'created_by': 'john.hummel@wallaroo.ai', 'created_at': '2025-08-05T18:41:42.646046+00:00', 'models': [{'name': 'ccfraud-model', 'versions': 2, 'owner_id': '""', 'last_update_time': datetime.datetime(2025, 8, 7, 17, 14, 40, 69527, tzinfo=tzutc()), 'created_at': datetime.datetime(2025, 8, 5, 18, 41, 44, 287180, tzinfo=tzutc())}], 'pipelines': [{'name': 'metrics-retrieval-tutorial-pipeline', 'create_time': datetime.datetime(2025, 8, 5, 18, 41, 44, 982232, tzinfo=tzutc()), 'definition': '[]'}]}

model_name = "ccfraud-model"
model_file_name = "./models/ccfraud.onnx"
ccfraud_model = (wl.upload_model(name=model_name, 
                                 path=model_file_name, 
                                 framework=wallaroo.framework.Framework.ONNX)
                                 .configure(tensor_fields=["tensor"])
                )

# create the pipeline
pipeline_name = "metrics-retrieval-tutorial-pipeline"
pipeline = wl.build_pipeline(pipeline_name)

# add the model as a pipeline step

pipeline.add_model_step(ccfraud_model)

# set the deployment configuration for 0.5 cpu, 1 replica, 1 Gi RAM
deploy_config = wallaroo.DeploymentConfigBuilder().replica_count(1).cpus(0.5).memory("1Gi").build()

# deploy the pipeline
pipeline.deploy(deployment_config=deploy_config, wait_for_status=False)
# saved for later steps
deploy = pipeline._deployment

Deployment initiated for metrics-retrieval-tutorial-pipeline. Please check pipeline status.

# wait until deployment is complete before continuing
import time
time.sleep(15)

while pipeline.status()['status'] != 'Running':
    time.sleep(15)
    print("Waiting for deployment.")
    pipeline.status()['status']
pipeline.status()['status']

'Running'

import time
timeout = time.time() + 60   # 1 minutes from now
while True:
    if time.time() > timeout:
        break
    pipeline.infer_from_file("./data/cc_data_10k.arrow")

pipeline.logs()

Warning: There are more logs available. Please set a larger limit or request a file using export_logs.

pipeline.undeploy()

# set prometheus requirements
pipeline_id = pipeline_name # the name of the pipeline
step = "1m" # the step of the calculation

# this will also format the timezone in the parsing section
timezone = "US/Central"

selected_timezone = pytz.timezone(timezone)

# Define the start and end times
data_start = selected_timezone.localize(datetime.datetime(2025, 8, 10, 9, 0, 0))
data_end = selected_timezone.localize(datetime.datetime(2025, 8, 12, 9, 59, 59))

# this is the URL to get prometheus metrics
query_url = f"{wl.api_endpoint}/v1/metrics/api/v1/query_range"

# Retrieve the token 
headers = wl.auth.auth_header()

# Convert to UTC and get the Unix timestamps
start_timestamp = int(data_start.astimezone(pytz.UTC).timestamp())
end_timestamp = int(data_end.astimezone(pytz.UTC).timestamp())    

query_rps = f'sum by (pipeline_name) (rate(latency_histogram_ns_count{{pipeline_name="{pipeline_id}"}}[{step}]))'
#request parameters
params_rps = {
    'query': query_rps,
    'start': start_timestamp,
    'end': end_timestamp,
    'step': step
}

response_rps = requests.get(query_url, headers=headers, params=params_rps)

if response_rps.status_code == 200:
    print("Requests Per Second Data:")
    # only show non-zero results
    display(list(filter(lambda x: x[1] != '0', response_rps.json()['data']['result'][0]['values'])))
else:
    print("Failed to fetch RPS data:", response_rps.status_code, response_rps.text)

Requests Per Second Data:

[[1754936520, '0.0858'], [1754936580, '0.34545454545454546']]

query_inference_rate = f'sum by (pipeline_name) (rate(tensor_throughput_batch_count{{pipeline_name="{pipeline_id}"}}[{step}]))'

# inference rte
params_inference_rate = {
    'query': query_inference_rate,
    'start': start_timestamp,
    'end': end_timestamp,
    'step': step
}

response_inference_rate = requests.get(query_url, headers=headers, params=params_inference_rate)

if response_inference_rate.status_code == 200:
    print("Cluster Inference Rate Data:")
    # filter out non-zero
    display(list(filter(lambda x: x[1] != '0', response_rps.json()['data']['result'][0]['values'])))
else:
    print("Failed to fetch Inference Rate data:", response_inference_rate.status_code, response_inference_rate.text)

Cluster Inference Rate Data:

[[1754936520, '0.0858'], [1754936580, '0.34545454545454546']]