QGate-Sln-MLRun

The Quality Gate for solution MLRun (and Iguazio). The main aims of the project are:

• independent quality test (function, integration, performance, vulnerability, acceptance, ... tests)
• deeper quality checks before full rollout/use in company environments
• identification of possible compatibility issues (if any)
• external and independent test coverage
• community support
• etc.

The tests use these key components, MLRun solution see GIT mlrun, sample meta-data model see GIT qgate-model and this project.

Test scenarios

The quality gate covers these test scenarios (✅ done, ✔ in-progress, ❌ planned):

• 01 - Project
  • ✅ TS101: Create project(s)
  • ✅ TS102: Delete project(s)
• 02 - Feature set
  • ✅ TS201: Create feature set(s)
  • ✅ TS202: Create feature set(s) & Ingest from DataFrame source (one step)
  • ✅ TS203: Create feature set(s) & Ingest from CSV source (one step)
  • ✅ TS204: Create feature set(s) & Ingest from Parquet source (one step)
  • ✅ TS205: Create feature set(s) & Ingest from SQL source (one step)
  • ✔ TS206: Create feature set(s) & Ingest from Kafka source (one step)
  • ✔ TS207: Create feature set(s) & Ingest from HTTP source (one step)
• 03 - Ingest data
  • ✅ TS301: Ingest data (Preview mode)
  • ✅ TS302: Ingest data to feature set(s) from DataFrame source
  • ✅ TS303: Ingest data to feature set(s) from CSV source
  • ✅ TS304: Ingest data to feature set(s) from Parquet source
  • ✅ TS305: Ingest data to feature set(s) from SQL source
  • ✔ TS306: Ingest data to feature set(s) from Kafka source
  • ✔ TS307: Ingest data to feature set(s) from HTTP source
• 04 - Ingest data & pipeline
  • ✅ TS401: Ingest data & pipeline (Preview mode)
  • ✅ TS402: Ingest data & pipeline to feature set(s) from DataFrame source
  • ✅ TS403: Ingest data & pipeline to feature set(s) from CSV source
  • ✅ TS404: Ingest data & pipeline to feature set(s) from Parquet source
  • ✅ TS405: Ingest data & pipeline to feature set(s) from SQL source
  • ✔ TS406: Ingest data & pipeline to feature set(s) from Kafka source
  • ❌ TS407: Ingest data & pipeline to feature set(s) from HTTP source
• 05 - Feature vector
  • ✅ TS501: Create feature vector(s)
• 06 - Get data from vector
  • ✅ TS601: Get data from off-line feature vector(s)
  • ✅ TS602: Get data from on-line feature vector(s)
• 07 - Pipeline
  • ✅ TS701: Simple pipeline(s)
  • ✅ TS702: Complex pipeline(s)
  • ✅ TS703: Complex pipeline(s), mass operation
• 08 - Build model
  • ✅ TS801: Build CART model
  • ❌ TS802: Build XGBoost model
  • ❌ TS803: Build DNN model
• 09 - Serve model
  • ✅ TS901: Serving score from CART
  • ❌ TS902: Serving score from XGBoost
  • ❌ TS903: Serving score from DNN
• 10 - Model monitoring/drifting
  • ❌ TS1001: Real-time monitoring
  • ❌ TS1002: Batch monitoring
• 11 - Performance tests
  • ❌ TS1101: Simple pipeline
  • ❌ TS1102: Complex pipeline(s)
  • ❌ TS11xx: TBD.

NOTE: Each test scenario contains addition specific test cases (e.g. with different targets for feature sets, etc.).

Test inputs/outputs

The quality gate tests these inputs/outputs (✅ done, ✔ in-progress, ❌ planned):

• Outputs (targets)
  • ✅ RedisTarget, ✅ SQLTarget/MySQL, ✔ SQLTarget/Postgres, ✅ KafkaTarget
  • ✅ ParquetTarget, ✅ CSVTarget
  • ✅ File system, ❌ S3, ❌ BlobStorage
• Inputs (sources)
  • ✅ Pandas/DataFrame, ✅ SQLSource/MySQL, ❌ SQLSource/Postgres, ❌ KafkaSource
  • ✅ ParquetSource, ✅ CSVSource
  • ✅ File system, ❌ S3, ❌ BlobStorage

The current supported sources/targets in MLRun.

Sample of outputs

The PART reports in original form, see:

• all DONE - HTML, TXT
• with ERRors - HTML, TXT

Usage

You can easy use this solution in four steps:

1. Download content of these two GIT repositories to your local environment
   • qgate-sln-mlrun
   • qgate-model
2. Update file qgate-sln-mlrun.env from qgate-model
   • Update variables for MLRun/Iguazio, see MLRUN_DBPATH, V3IO_USERNAME, V3IO_ACCESS_KEY, V3IO_API
     • setting of V3IO_* is needed only in case of Iguazio installation (not for pure free MLRun)
   • Update variables for QGate, see QGATE_* (basic description directly in *.env)
     • detail setup configuration
3. Run from qgate-sln-mlrun
   • python main.py
4. See outputs (location is based on QGATE_OUTPUT in configuration)
   • './output/qgt-mlrun-<date> <sequence>.html'
   • './output/qgt-mlrun-<date> <sequence>.txt'

Precondition: You have available MLRun or Iguazio solution (MLRun is part of that), see official installation steps, or directly installation for Desktop Docker.

Tested with

The project was tested with these MLRun versions (see change log):

• MLRun (in Kubernates or Desktop Docker)
  • ❌ MLRun 1.8.0 (plan Q1/2025)
  • ✔ MLRun 1.7.0 (? 1.7.1 ?)
  • ✅ MLRun 1.6.4, 1.6.3, 1.6.2, 1.6.1, 1.6.0
  • ✅ MLRun 1.5.2, 1.5.1, 1.5.0
  • ✅ MLRun 1.4.1
  • ✅ MLRun 1.3.0
• Iguazio (k8s, on-prem, VM on VMware)
  • ✅ Iguazio 3.5.3 (with MLRun 1.4.1)
  • ✅ Iguazio 3.5.1 (with MLRun 1.3.0)

NOTE: Current state, only the last MLRun/Iguazio versions are tested (the backward compatibility is based on MLRun/Iguazio, see).

Others

• To-Do, the list of expected/future improvements, see
• Applied limits, the list of applied limits/issues, see
• How can you test the solution?, you have to focus on Linux env. or Windows with WSL2 (see step by step tutorial)
• MLRun/Iguazio, the key changes in a nutshell (customer view), see
• MLRun local installation, see the hack