Atlas Search Index Latency Test

In this toolkit, you can test the approximate latency of Lucene index update.

In contrast to regular indexes (Wired Tiger Indexes) in MongoDB, Lucene search indexes are asynchronously updated. Therefore, you might see stale data, if you immediately query just after you've written.

Where It was Tested

It's been tested succesfully on Amazon Linux 2, with the node version v17.1.0 and npm version 8.1.2.

Installing

After you've successfully installed node and npm on the machine, clone this repository and run the following command to install dependent libraries
- npm install

Preparing Atlas Cluster and Search Index

Provision an M30 Atlas Cluster in the region closest to the location where the script is going to be executed.
Load Sample Data into the cluster.
Create a dynamic index on the collection movies in the sample_mflix database. Result should be similar to below.

Modifying Connection String

Change the connectionString variable in the latency_test.js accordingly.

How the Test Works

In every step script generates a random string and updates a field in the collection and that field is also in the search index.
- Depends on the variable numberOfUpdatesPerIteration , 1 or more documents can be updated.
- _id fields of the updated documents were stored temporarily.
After updateMany successfully executed, $search query executed with the phrase search to match generated random string.
- $search query is executed until it returns all the updated documents in the updateMany stage.

Running the Test

Locate the folder where the script in place
Run the following:
- node latency_test.js

Observing Results

After you run you will see the similar output as below.

How to Customize This Script

Change the db/collection name accordingly in the config part of the script
Change the index name and field to be updated properly.
You can update multiple fields rather than one field as in this script and re-evaluate the latency values
This is only one single thread therefore you can run multiple node process concurrently and re-evaluate the latency values
If you increase number of updates per iteration, you may see higher latencies.
- e.g. var numberOfUpdatesPerIteration=1000

afsungur/mongodb-atlas-fts-latency-test