The MongoDB API Tester

This repository is used to perform correctness testing against a MongoDB API endpoint and analyze the results. It also contains instructions for how to run performance tests.

Performance Testing

• Use these intructions to run YCSB
• Use these instructions to run Socialite

Correctness Testing

There are two main components to the correctness tests:

• The test runner, distributed as a Docker image that checks out the MongoDB repository and runs API correctness tests
• The results analyzer, a Python script that loads the raw output from the test into MongoDB, analyzes it to classify failures as either unsupported features or potential bugs, and outputs a CSV file for easy review.

Test Runner Docker

The Docker image in this repository can run API acceptance tests against a service implementing the MongoDB API. Everything you need to run these tests is built into it.

There are only 3 things you need to run this project:

• Docker
• Git (or you can just download this repo as a zip file from Github),
• A bash shell.

The tests this harness runs are the subset of official MongoDB correctness tests that treat the system under test as a black box, without relying on fixtures or assumptions about the server's internal state.

We built 6 test suites that make sense in this DBaaS context and validate most of the features of the MongoDB 4.0, 4.2, 4.4, or 5.x API.

Recommended infrastructure

MongoDB Atlas

Recommended cluster configuration (larger instances will run faster but not impact results):

• Dedicated MongoDB Atlas M50 Cluster with 3000 IOPS, 16000 max connections, 32GB RAM, and 8 vCPUs.

AWS DocumentDB

Recommended cluster configuration (larger instances will run faster but not impact results):

• DocumentDB db.r5.large or xlarge with 3 instances.
• EC2 m5.large running AWS Cloud9 to easily connect to DocumentDB and manage the tests.
• Be sure to properly set up your VPC and security groups to allow the testing service to connect.

Note: If you provision a server other than Amazon Linux or Ubuntu, you will have to adapt the setup scripts accordingly to install Docker and Git.

Cosmos DB

Recommended cluster configuration:

• Standard Cosmos DB deployment with their MongoDB Imitation API.

Instructions to run the tests

Note: Changes were introducted in MongoDB 5.0 that are incompatible with older versions of the testing harness. Please use the pre-5.0 directory for running tests against MongoDB Atlas v3.6, v4.0, v4.2, and v4.4. The post-5.0 directory can be used for running tests against MongoDB Atlas v5.0 and later.

On MongoDB Atlas

• Clone this repo
• Build the image - it's a bit long the first time (2-3 minutes) so go get some coffee, it's on us!
• Use the pre-5.0 directory for running the 3.6, 4.0, 4.2, and 4.4 test suites.
• Use the post-5.0 directory for running the 5.0 test suite and beyond.
• Version should either be 4.0, 4.2, 4.4, or 5.x depending on the suite you plan on running.

./0_docker-build.sh <version>

• Create a MongoDB Atlas Cluster v4.0, 4.2, 4.4, or 5.x. Then create an admin user and whitelist your public IP address. Find some help here.
• Collect the MongoDB Atlas connection string for the next command.
• Run the 6 test suites.

./1_docker-run.sh 'mongodb+srv://<USER>:<PASSWORD>@<DEPLOYMENT NAME>.tsnei.mongodb.net' <version>

• You can monitor by looking at the results folder.
• Or you can docker ps -a and docker logs -f <CONTAINER_NAME> to check what is currently running.
• All the results (JSON + STDOUT) are in the results-<version> folder.

On Amazon DocumentDB

• Set up a DocumentDB Cluster and an EC2 server using the instructions provided here.

• Step 1 will help you install the DocumentDB Cluster and setup the security aspect.

• Step 2 will help you setup an EC2 server in the same VPC because it's currently not possible to connect to DocumentDB from the outside of the VPC and AWS.

• Step 3 is optional as we are going to use the MongoDB Shell from our Docker image but you can still go through step 3 to validate your setup and that you are allowed to connect from this EC2 server.

  • The cluster and EC2 client should be in the same Security Group.
  • The client should be able to communicate over TCP port 27017 with the cluster.

• Note 1: If you are unable to connect, it could be that the security group is still not allowing inbound connections to port 27017. For help with that and other connecting troubleshooting, check out AWS' guide here.

• Note 2: If you are running a cluster with more than one node (i.e. a 3 node replica set) you must assign subnets for each of the other nodes in your cluster.

• Once you are sure that you can connect from this EC2 server to DocumentDB execute this script on the EC2 client:

For the Amazon Linux image:

sudo yum update -y
sudo yum install docker git -y
sudo service docker start
sudo usermod -a -G docker ec2-user

For the Ubuntu image:

sudo apt-get update
sudo apt-get upgrade -y
sudo apt-get install docker git -y
sudo service docker start
sudo usermod -a -G docker ubuntu

• For that last command to take effect, you need to log out and log in again to your EC2 server.
• Then run this script.

git clone https://github.com/mongodb-developer/service-tests.git
cd ./service-tests
./0_docker-build.sh <version>
./1_docker-run.sh 'mongodb://<USER>:<PASSWORD>@db-name.cluster-xxxx.eu-west-1.docdb.amazonaws.com:27017/?ssl=true&replicaSet=rs0' <version>

Notes:

1. Please copy and paste the DocumentDB URI from the AWS interface but confirm that it contains the same flags and none extra.
2. The AWS certificate which is required to access any Document DB cluster is built into the Docker Image (see the Dockerfile for more details). There is no need to pass the "ssl_ca_certs" argument found in the connection string AWS provides.

• When it's over, you can collect the logs and the JSON files in the results-<version> folder.

On Cosmos DB

• Follow the instructions to deploy the "Azure Cosmos DB API for MongoDB" here.

• Clone this repository locally

• Navigate to the service-tests directory

• Build your Docker image using ./0_docker-build.sh <version>

• Run the tests using ./1_docker-run.sh '<connection string>' <version> (Note - the connection string can be found in the Cosmos DB portal under 'Settings' -> 'Connection String'. Please truncate the string after the port number. It should look something like: 'mongodb://accountname:[email protected]:10255/')

• Note : Running the comparison tests against Cosmos DB is significantly more costly than Atlas or DocumentDB. Please keep this in mind if you are budget constrained.

Online results

Available at:

• https://www.isdocumentdbreallymongodb.com/
• https://www.iscosmosdbreallymongodb.com/

Results Breakdown from periodic test runs

Amazon DocumentDB Results Over Time

Amazon DocumentDB v4.0 with MongoDB v5.2 Tests ─ January 2022

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	9	0	6	0
JSON Schema	21	2	0	19	0
Change Streams	25	3	0	22	0
Aggregation	317	88	0	229	0
Core	998	361	0	637	0
Transactions	49	21	0	28	0
TOTAL	1425	484	0	941	0
PERCENTAGES	-	33.96%	0%	66.04%	0%

Amazon DocumentDB v4.0 with MongoDB v5.0 Tests ─ November 2021

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	9	0	6	0
JSON Schema	25	2	0	23	0
Change Streams	22	2	0	20	0
Aggregation	300	84	0	216	0
Core	1001	363	0	638	0
Transactions	52	22	0	30	0
TOTAL	1415	482	0	933	0
PERCENTAGES	-	34.06%	0%	65.94%	0%

AWS DocumentDB v4.0 with MongoDB v5.0 Tests ─ July 2021

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	9	0	6	0
JSON Schema	25	2	0	23	0
Change Streams	22	2	0	20	0
Aggregation	295	76	0	219	0
Core	991	358	0	634	0
Transactions	52	22	0	30	0
TOTAL	1400	469	0	932	0
PERCENTAGES	-	33.43%	0%	66.57%	0%

AWS DocumentDB v4.0 with MongoDB v4.4 Tests ─ May 18, 2021

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	9	0	6	0
Json Schema	26	2	0	24	0
Change Streams	22	2	0	20	0
Aggregation	236	82	0	154	0
Core	891	348	0	543	0
Transactions	49	25	0	24	0
TOTAL	1239	468	0	771	0
PERCENTAGES	-	37.77%	0%	62.23%	0%

AWS DocumentDB v4.0 with MongoDB v4.0 Tests ─ Nov 13, 2020

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	13	7	0	6	0
Json Schema	24	0	0	24	0
Change Streams	20	0	0	20	0
Aggregation	166	76	0	90	0
Core	849	348	0	501	0
Transactions	32	12	0	20	0
TOTAL	1104	443	0	661	0
PERCENTAGES	-	40.13%	0%	59.87%	0%

Azure Cosmos DB Results Over Time

Azure Cosmos DB v4.2 with MongoDB v6.0 Tests ─ August 2022

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
JSON Schema	21	2	0	19	0
Change Streams	38	8	0	30	0
Aggregation	328	109	0	219	0
Core	1030	346	0	684	0
Transactions	49	21	0	28	0
TOTAL	1480	496	0	984	0
PERCENTAGES	-	33.51%	0%	66.49%	0%

Azure Cosmos DB v4.2 with MongoDB v5.2 Tests ─ March 2022

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
JSON Schema	21	2	0	19	0
Change Streams	25	3	0	22	0
Aggregation	317	95	0	222	0
Core	998	338	0	660	0
Transactions	49	21	0	28	0
TOTAL	1425	474	0	950	0
PERCENTAGES	-	33.29%	0%	66.71%	0%

Azure Cosmos DB v4.2 with MongoDB v4.2 Tests ─ March 2022

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
JSON Schema	26	2	0	24	0
Change Streams	24	2	0	22	0
Aggregation	209	83	0	126	0
Core	885	335	0	550	0
Transactions	39	20	0	19	0
TOTAL	1198	452	0	746	0
PERCENTAGES	-	37.73%	0%	62.27%	0%

Azure Cosmos DB v4.0 with MongoDB v5.2 Tests ─ January 2022

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
JSON Schema	21	2	0	19	0
Change Streams	25	3	0	22	0
Aggregation	316	100	0	216	0
Core	998	338	0	660	0
Transactions	49	21	0	28	0
TOTAL	1424	474	0	950	0
PERCENTAGES	-	33.29%	0%	66.71%	0%

Azure Cosmos DB v4.0 with MongoDB v5.0 Tests ─ July 2021

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
JSON Schema	25	2	0	23	0
Change Streams	22	2	0	20	0
Aggregation	295	35	0	260	0
Core	991	330	0	661	0
Transactions	52	21	0	31	0
TOTAL	1400	400	0	1000	0
PERCENTAGES	-	28.57%	0%	71.43%	0%

Azure CosmosDB v4.0 with MongoDB v4.4 Tests ─ Mar 4, 2021

Tests Suite	Number of tests	Succeeded	Skipped	Failed	Errored
Decimal	15	10	0	5	0
Json Schema	26	2	0	24	0
Change Streams	22	2	0	20	0
Aggregation	236	88	0	148	0
Core	891	303	0	588	0
Transactions	49	4	0	45	0
TOTAL	1239	409	0	830	0
PERCENTAGES	-	33.01%	0%	66.99%	0%

Pro tips

• You just need to build the image once then you can run the tests as much as you want.
• If you re-run the tests with the script 1_docker-run.sh, the results folder is reset so please save it before if you want to keep the previous results.
• If you want to run the test suites individually, you can. Read the 1_docker-run.sh and run the docker command manually with the correct URI.

Test Results Analyzer

Once you have finished running the test suite, the next step is to analyze and understand the failures from the run.

The goal of this analysis is to allow us to quickly see what core MongoDB features are and are not supported by a given implementation. The analysis attempts to categorize failures into UNSUPPORTED and FURTHER_INVESTIGATION to help reduce the amount of debugging and allow us to quickly focus on more interesting failures.

There are about 1000 correctness tests that are run, and looking at the each failure one by one would take too much time. The analyzer lets you quickly sift through the various tests, categorize a large set of them, and then do further analysis as needed. At this time, the analyzer only supports classifying DocumentDB errors, as it relies on the specific error messages that service emits.

Requirements

1. Python 3.6+
2. A MongoDB instance: The results are analyzed in MongoDB. If you're testing Atlas and another service side-by-side, you can use the Atlas instance you tested, or you can just use a free tier instance. Note: This code will not work against DocumentDB (as of 2019-01-22), as it uses aggregation features that DocumentDB does not implement.
3. pymongo package: This driver is needed for connecting to MongoDB instance.
4. dnspython package: You will need this if using Atlas and mongodb+srv connection strings.

Steps to Run

After running the docker tests for a given target platform the results will be stored in the ./results directory. The Python program can then be run by executing the following:

$ pip install -r requirements.txt
$ python analyze.py --mdburl "mongodb://MDB_URL" --platform EMULATION_PLATFORM_NAME

The MongoDB URL (--mdburl) should point to the Atlas instance you have configured to perform the data analysis with. The Platform (--platform) should be a string representing the MongoDB-compatible platform that has been tested. e.g. one of atlas, cosmosdb, foundationdb, or documentdb. There are no restrictions on what that string is but you should make it meaningful to avoid any confusion.

The --mdburl and --platform flag are required; optional parameters are:

1. --drop: Drop the previous database set by --db, default: False
2. --version: Database version run against, default: v4.0
3. --db: Database to store the results in, default: results
4. --coll: Collection to store results in, default: correctness
5. --run: Eun number, if we want to keep multiple results of the same platform, default: 1
6. --csv: CSV file name to be output, default: ./results.csv
7. --csvfilter: Filter to be applied to CSV processing, default: {}
8. --rdir: Results directory to use, default: ./results-4.0

The results.csv will be generated for all data in the database (assuming an open filter). This CSV can be imported to the spreadsheet of your choice and further analysis can be done. The CSV file will contain the file, the suite, platform, version, run, status, reason, description.

Further analysis

After digesting the results and finding the more interesting failures, we can look deeper by running those specific tests against the service with the following steps:

1. Clone MongoDB repository and check out version

The version checked out is the default version we tested with.

$ git clone --branch=v5.0 https://github.com/mongodb/mongo.git
$ cd ./mongo

2. Identify and run test

Choose an 'interesting' test to run, and execute it using the following command:

$ /path/to/mongo PLATFORM_URL path/to/test.js

The /path/to/mongo is the shell used to execute the javascript, the PLATFORM_URL is the emulation platform that the failure was recorded on, the path/to/test.js is the test file that failed.

The mongo shell will kick off the script and the results will be written to stdout, with a stack trace (if there is a failure). From there you can investigate the test file, copy it, change it, etc.

Should you find anything glaringly problematic with the tests, please reach out and let us know.

TODO

1. Document Layer (FoundationDB): Run correctness results through analyze.py
2. Automate more of the analysis, it is very simplistic right now

Authors & Contact

Test runner

• Craig Homa [email protected] - Market Intelligence Analyst
• Maxime BEUGNET [email protected] - Senior Developer Advocate @ MongoDB
• Greg McKeon - Former MongoDB employee - Competitive Analyst

Results analyzer

• Shawn McCarthy - Former MongoDB employee - Developer Advocate

LICENSE

This repository is published under the Server Side Public License (SSPL) v1. See individual files (LICENSE) for details.