Skip to main content
Version: 0.8.1

Search with AWS Lambda

In this tutorial, we will index and search about 20 million log entries (7 GB decompressed) located on AWS S3 with Quickwit Lambda.

Concretely, we will deploy an AWS CloudFormation stack with the Quickwit Lambdas, and two buckets: one staging for hosting gzipped newline-delimited JSON files to be indexed and one for hosting the index data. The staging bucket is optional as Quickwit indexer can read data from any S3 files it has access to.

Tutorial stack overview


Install AWS CDK

We will use AWS CDK for our infrastructure automation script. Install it using npm:

npm install -g aws-cdk 

You also need AWS credentials to be properly configured in your shell. One way is using the [credentials file](

Finally, clone the Quickwit repository:
git clone
cd tutorials/simple-lambda-stack

Setup python environment

We use python 3.10 to define the AWS CloudFormation stack we need to deploy, and a python CLI to invoke Lambdas. Let's install those few packages (boto3, aws-cdk-lib, click, pyyaml).

# Install pipenv if needed.
pip install --user pipenv
pipenv shell
pipenv install

Download Quickwit Lambdas

mkdir -p cdk.out
wget -P cdk.out
wget -P cdk.out

Bootstrap and deploy

Configure the AWS region and account id where you want to deploy the stack:

export CDK_ACCOUNT=123456789
# us-east-1 is where the Quickwit public dataset bucket is located
export CDK_REGION=us-east-1

If this region/account pair was not bootstrapped by CDK yet, run:

cdk bootstrap aws://$CDK_ACCOUNT/$CDK_REGION

This initializes some basic resources to host artifacts such as Lambda packages. We can now deploy the stack:

cdk deploy -a cdk/

Index the HDFS logs dataset

Here is an example of a log entry of the dataset:

"timestamp": 1460530013,
"severity_text": "INFO",
"body": "PacketResponder: BP-108841162-, type=HAS_DOWNSTREAM_IN_PIPELINE terminating",
"resource": {
"service": "datanode/01"
"attributes": {
"class": "org.apache.hadoop.hdfs.server.datanode.DataNode"
"tenant_id": 58

If you have 5 minutes ahead of you, you can index the whole dataset which is available on our public S3 bucket.

python index s3://quickwit-datasets-public/hdfs-logs-multitenants.json.gz

Once the indexing Lambda is started, you cannot stop it manually. A keyboard interrupt will stop the client but not the function. If you try running it again, you might get a TooManyRequestsException error. That's because the concurrency of the indexer is set to 1 to avoid race conditions on the metastore.

If you don't want to wait, just index the 10,000 documents dataset:

python index s3://quickwit-datasets-public/hdfs-logs-multitenants-10000.json

Execute search queries

Let's start with a query on the field severity_text and look for errors: severity_text:ERROR:

python search '{"query":"severity_text:ERROR"}'

It should respond in under 1 second and return 10 hits out of 345 if you indexed the whole dataset. If you index the first 10,000 documents, you won't have any hits, try to query INFO logs instead.

Let's now run a more advanced query: a date histogram with a term aggregation on the `severity_text`` field:

python search '{ "query": "*", "max_hits": 0, "aggs": { "events": { "date_histogram": { "field": "timestamp", "fixed_interval": "30d" }, "aggs": { "log_level": { "terms": { "size": 10, "field": "severity_text", "order": { "_count": "desc" } } } } } } }'

It should respond under 2 seconds and return the top log levels per 30 days.

Cleaning up

First, you have to delete the files created on your S3 buckets. Once done, you can delete the stack.

cdk destroy -a cdk/
rm -rf cdk.out

Congratz! You finished this tutorial! You can level up with the following tutorials to discover all Quickwit features.

Next steps