Mongo Task Generator

Dynamically split evergreen tasks into subtasks for testing the mongodb/mongo project.

Table of contents

• Mongo Task Generator
  • Table of contents
  • Description
  • Getting Help
    • What's the right channel to ask my question?
    • How can I request a change/report a bug in Mongo Task Generator?
    • What should I include in my ticket or question?
  • Dependencies
  • Installation
  • Usage
  • Documentation
  • Contributor's Guide
    • High Level Architecture
    • Setting up a local development environment
    • linting/formatting
    • Running tests
    • Versioning
    • Code Review
    • Deployment
  • Resources

Description

A tool to dynamically generate and split resmoke tests for the mongodb/mongo project into sub-tasks that can be run in parallel. For example, a task that runs in 2 hours could be split into 4 sub-tasks that can be run in parallel to reduce the wall-clock time of to closer to 30 minutes.

The tool will use Evergreen Test Stats to determine how to divide the tests between sub-tasks. If test stats are not available for any reason, it will fallback to splitting tests so that each sub-tasks contains an equal number of tests.

The tool will generate fuzzer tasks as well. In the case of fuzzer tasks, the generation determines the amount of fuzzer tests to generate. There are two controls for how many fuzzer tests to generate, n_files and n_tasks. They should be specified in the task definition in the evergreen config file. n_files controls how many fuzzer files will be generated in each sub-task, effectively controlling the runtime of each subtask. n_tasks controls how many sub-tasks will be generated. For example, if n_files=3 and n_tasks=5, a total of 15 files will be run in 5 sub-tasks.

Getting Help

What's the right channel to ask my question?

If you have a question about Mongo Task Generator, please reach out on slack in the #server-testing channel, or email us at dev-prod-dag@mongodb.com.

How can I request a change/report a bug in Mongo Task Generator?

Create a DAG ticket in Jira.

What should I include in my ticket or question?

Please include as much information as possible. This can help avoid long information-gathering threads.

Please include the following:

• Motivation for Request: Why is this change being requested? (This help us understand the priority and urgency of the request)
• Context: Is there any background information we should be aware of for this request?
• Description: What would you like investigated or changed?

Dependencies

This project is built to run on linux. It is tested on Ubuntu 18.04.

The evergreen CLI is required.

Installation

The tool can be downloaded from the github releases page. It is currently built on Ubuntu 18.04 and should work on recent x86 linux distributions.

The install can be automated using a tool like wget or curl:

wget https://github.com/mongodb/mongo-task-generator/releases/download/v0.1.0/mongo-task-generator

Usage

To generate the task information, run the mongo-task-generator command at the root of the mongodb repository. The generated configuration will be created in the generated_resmoke_config directory. The evergreen_config.json file in that directory should be passed to evergreen as part of a generate.tasks command call.

mongo-task-generator

USAGE:
    mongo-task-generator --evg-project-file <EVG_PROJECT_FILE> --expansion-file <EXPANSION_FILE> --evg-auth-file <EVG_AUTH_FILE>

OPTIONS:
        --evg-auth-file <EVG_AUTH_FILE>
            File with information on how to authenticate against the evergreen API

        --evg-project-file <EVG_PROJECT_FILE>
            File containing evergreen project configuration

        --expansion-file <EXPANSION_FILE>
            File containing expansions that impact task generation

    -h, --help
            Print help information

Documentation

See Generating tasks for details on how task generation works.

Contributor's Guide

High Level Architecture

Flow chart of generating tasks:

graph TD
    A[Read Evergreen Project Config] -->B(For each task in each build variant)
    B --> C{task is generated?}
    C --> |No| B
    C -->|Yes| D{task is fuzzer?}
    D -->|Yes| E[Generate Fuzzer based on task definition]
    E --> B
    D -->|No| F[Lookup task runtime history from Evergreen]
    F --> G{valid history obtained?}
    G --> |Yes| H[Split task by test runtimes]
    G --> |No| I[Split task by counts]
    H --> J[Generate resmoke suite config]
    I --> J
    J --> |generated task definitions|B
    B --> |done|K[For each Build Variant]
    K --> L[Create task specs and display tasks in Build Variant]
    L --> M[Create shrub config]
    M --> N[Write shrub config to disk]
    N --> O([End])

Setting up a local development environment

Make sure you have a rust development environment installed. See instructions here.

From the root of the repository, run cargo build to build the project.

linting/formatting

cargo fmt
cargo clippy

Running tests

cargo test

Versioning

This project uses semver for versioning.

Please include a description what is added for each new version in CHANGELOG.md.

Code Review

This project uses the Evergreen Commit Queue. Add a PR comment with evergreen merge to trigger a merge.

Deployment

Deployment is done via github actions. Deployment can be triggered via a git tag. To create a deployment, tag the commit you wish to deploy with a git tag and push the tag to the remote repository.

The git tags should be formatted as follows: "v" followed by the 3 number version being deployed. For example, "v0.1.0".

git tag v0.1.0
git push origin --tags

The tagged version will be deployed to the github release for the project.

Resources

• Evergreen generate.tasks documentation
• Evergreen test stats documentation