redis-sitesearch

About

redis-sitesearch is a package that indexes websites into RediSearch.

RediSearch is a querying and indexing engine for Redis that supports full-text search. You can read more about it here: https://oss.redislabs.com/redisearch/

redis-sitesearch was originally developed by Redis Labs to provide a search API for our documentation. You can see it live at: https://docs.redislabs.com.

Want to learn more? Join the Redis discord to meet other Redis and RediSearch users.

Installing

The development environment is provided as a Docker container. Running and deploying it all involve using Docker, and this section describes how to do so.

Note: If you want to avoid using Docker, you can try to reverse-engineer the installation steps from the Dockerfile and docker-compose.yml file.

Docker Requirements

This project was built and tested using the following Docker and docker-compose versions:

    Docker version 19.03.13, build 4484c46d9d
    docker-compose version 1.27.4, build 40524192

Ensure you are running at least these versions before you continue.

Environment Variables

The app is configured using environment variables. For local development, you'll need a .env file, from which the app will read environment variables.

An example is included, which you should copy from .env.example to .env from the root directory of the repository:

    cp .env.example .env

Deployment note: If you deploy redis-sitesearch to Google Cloud Platform using the deploy.sh script, the script expects to configure environment variables from a .env.prod file.

Redis Password-protection

This application expects to use a password to authenticate with Redis. The .env.example file includes the required environment variable, REDIS_PASSWORD.

In local development, running docker-compose up will start Redis in password-required mode using the password you set in the REDIS_PASSWORD environment variable, and the app will connect using that password. So, for local development, you shouldn't have to think about the password.

To connect to Redis using redis-cli, use the -a command-line flag or the AUTH command and enter the password from your .env file (or the password you manually set in the REDIS_PASSWORD environment variable, if you didn't use the .env file).

In production, the app will use whatever password you configure in the production environment's REDIS_PASSWORD environment variable.

Virtualenv

If you want code completion to work in an IDE, or you want to run commands outside of Docker (like pytest), create a virtualenv:

    python3.8 -m venv env

Then you can run the following to activate your virtualenv:

    source env/bin/activate

Finally, install the package locally, which also installs dependencies:

    pip install -e '.[dev]'  

Note: This project was built with and is only tested to work on Python 3.8.

Run docker-compose up

After you have the necessary Docker software, "install" the app with docker-compose:

    docker-compose up --build

This command builds the necessary images and brings up the app.

Indexing Documents

As soon as the app finishes starting up, it begins indexing sites configured in the DEV_SITES dictionary in the file sitesearch/config.py.

You should see output like this, indicating that supervisor started the processes:

worker_1   | 2021-01-12 21:17:51,091 INFO success: worker entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
app_1   | 2021-01-12 21:17:51,092 INFO success: app entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)

• The "app" process is the Python search API
• The "worker" process is an RQ worker that will process any indexing tasks it sees

You will also see output from a Redis container.

You might see errors and warnings as the indexing job proceeds, like this:

worker_1   | 2021-01-12 21:18:07 [sitesearch.indexer] ERROR: Document parser error -- Could not find breadcrumbs: https://docs.redislabs.com/latest/
worker_1   | 2021-01-12 21:18:08 [sitesearch.indexer] ERROR: Document parser error -- Skipping 404 page: https://docs.redislabs.com/latest/404.html

This output is normal -- some pages don't have breadcrumbs, and we skip 404 pages. Other times, the scraper encounters PDF or other non-HTML files and gives up trying to index them.

Indexing

There are two ways to run an indexing job: via the API and the command-line.

To index via the API, send a POST request to the /indexer endpoint. With curl, the request looks like this:

$ curl -X POST -H "X-API-KEY: whatever-you-want" -H "Content-Length: 0" "http://localhost:8080/indexer"
{"jobs": ["f602a492-8c21-4675-9f42-3854db3e6572", "3654c138-3160-4516-8d0f-8dd6f930529a", "d6d418f0-5d89-4589-85d6-ffbfddb1eba7", "a7845451-ec07-4c4f-b672-8fc69c7421e9", "b4e7a11f-03de-4ede-b691-2f7b32e88c01", "3a00af31-e5f1-4f7e-8dd3-9f1142d386b6", "6c0695b3-ea1b-421b-b010-d1e84cb197b3"]}

Indexing via the API is asynchronous. The /indexer endpoint creates a Redis Queue job for each site that the app will attempt to index. The response to your POST includes the IDs.

You can make a GET request to the /jobs/ endpoint to get the status of a job:

$ curl -H "X-API-KEY: whatever-you-want" "http://localhost:8080/jobs/d6d418f0-5d89-4589-85d6-ffbfddb1eba7"
{"id": "d6d418f0-5d89-4589-85d6-ffbfddb1eba7", "url": "https://developer.redislabs.com", "status": "queued", "created_at": "2021-05-01T00:13:05.263071", "ended_at": null}

You can also trigger reindexing with the index CLI command, like this:

    $ docker-compose exec app index "https://developer.redis.com"

The index command takes the URL of a site that the app is configured to index. The command indexes that site synchronously, without using RQ.

New Relic

The Python app tries to use New Relic. If you don't specify a valid NEW_RELIC_LICENSE_KEY environment variable in your .env or .env.prod files, the New Relic Agent will log errors. This is ok -- the app will continue to function without New Relic.

app_1   | 2021-01-12 22:36:16,867 (61/NR-Activate-Session/sitesearch) newrelic.core.agent_protocol CRITICAL - Disconnection of the agent has been requested by the data collector for the application where the agent run was None. Please contact New Relic support for further information.

Searching

The app maps localhost port 8080 to the search API. After you start the app with docker-compose up, can search using this URL:

    http://localhost:8080/search?q=<your search terms>

For example:

    $ curl "http://localhost:8080/search?q=redis"

Every response includes the total number of hits and a configurable number of results.

Developing

Assuming you have already brought up the app with docker-compose up per the installation instructions, this section describes things to know about for local development.

Writing Code

Once the application is running, you can write code using a local text editor, and your changes will sync to the container over a volume mount.

The WSGI server running the app is set to reload when code changes, so if you have docker-compose up running (so you can see output), and you change code in the project, you will see output about workers being "reaped."

This is well and good -- they are restarting.

Tests

To run tests, use this command:

    docker-compose run test

This runs pytest within the test container.

You can also run a specific test by passing in the -k parameter and the name of the test:

    docker-compose run test -s -k test_escapes_known_version_numbers

Running Tests Locally

If you have a .env file, you can run pytest locally (after you've activated your virtualenv) instead of through Docker, and the tests will pick up the necessary environment variables from your .env file. But I don't recommend this.

If you try to do this, the .env.example file in this repository configures the app to look at the Docker host for the Redis container, which is redis. To run the app and/or tests outside of Docker, you'll need to change the REDIS_HOST environment variable to "localhost" in your .env file.

Local vs. Docker redis

The Redis container that this project starts runs on port 6379 within Docker. If you want to connect to it from tools like redis-cli on your localhost, you can do so using port 16379.

The app uses port 16379 to avoid conflicting with any Redis instances running on the default port on localhost.

Deploying

This project includes a deployment script for Google Cloud Platform that is specific to Redis Labs' deployment.

If you want to use this script to deploy, follow these steps:

1. Install the Cloud SDK
2. Authenticate and select the appropriate project (ask Education for access!)
3. Make sure you have a .env.prod file -- this is important, and deploying will fail without it
4. Make sure you've committed your local changes
5. Run the deployment script: ./deploy.sh

The deployment script does the following:

1. Builds a new image based on the code in your working directory, which includes the git hash as part of the tag
2. Pushes the image to Google Container Registry
3. Creates a new instance template using the image and the contents of your .env.prod -- all env vars there will be sent to containers run from that template
4. Starts a rolling update in our production regions: US-West (us-west-1a), Mumbai (asia-south-1c), and Zurich (europe-west6-a)

When the rolling update completes, the script will terminate.

However, the rolling update is not technically complete. Each search node will reindex as it comes online, and the new node's health check will only succeed once that completes, in about 20 minutes.

Until the update completes, the past version of the search service remains available.

Questions

If you have any questions, ask the Education team!