Skip to main content
Version: Next

Pre-migration testing

To ensure a smooth and successful migration from MongoDB, we offer several methods through which you can test your application with FerretDB.

Operation modes

We offer multiple operation modes which help facilitate the testing of your application by enabling FerretDB to act as a proxy. For more details, refer to the operation modes.

Manual and automated testing with diff-normal mode

For details on how to install FerretDB, refer to the quickstart guide.

You can manually test your application or use integration tests, among other methods. Afterward, you can inspect the differential output for errors or inconsistencies between responses that require your attention.

As an example, let us say that your application performs some complex query and you'd like to test it in diff-normal mode. You would do the following:

  1. Start FerretDB in diff-normal mode.

    This can be achieved by setting the --mode flag or FERRETDB_MODE environment variable to diff-normal. By default, FerretDB starts in normal mode (--mode=normal/FERRETDB_MODE=normal). For more details, see operation modes.

    Ensure to specify --listen-addr and --proxy-addr flags or set the FERRETDB_LISTEN_ADDR and FERRETDB_PROXY_ADDR environment variables. Specify the address of your MongoDB instance for --proxy-addr flag or FERRETDB_PROXY_ADDR environment variable. See docs for more details. For example:

    ferretdb --mode=diff-normal \
    --proxy-addr=<mongodb-URI> \
    --listen-addr=<ferretdb-listen-address> \

    The --listen-addr flag or the FERRERDB_LISTEN_ADDR environment variable is set to by default.

  2. Run mongosh to connect to the --listen-addr and then insert some documents.

  3. Run a query to fetch the first post from each author sorted by date and author.

    Please note that due to running in diff-normal mode, any error returned from FerretDB will be transmitted to the client, allowing us to promptly identify the issue. In the majority of cases, this does not necessitate additional scrutiny of the diff output. Nevertheless, if FerretDB does not handle the error, additional inspection becomes necessary.

    # run mongosh
    $ mongosh
    // insert some documents
    title: 'title A',
    body: 'some content',
    author: 'Bob',
    date: ISODate('2023-08-29T10:33:23.134Z')
    title: 'another title',
    body: 'some content',
    author: 'Bob',
    date: ISODate('2023-08-28T10:33:23.134Z')
    title: 'title B',
    body: 'some content',
    author: 'Alice',
    date: ISODate('2023-08-20T10:33:23.134Z')
    title: 'some other title',
    body: 'some content',
    author: 'Alice',
    date: ISODate('2023-08-21T10:33:23.134Z')

    // run the query
    { $sort: { date: 1, author: 1 } },
    $group: {
    _id: '$author',
    firstPost: { $first: '$date' }
    // the below error is returned to the client:
    // MongoServerError: $group accumulator "$first" is not implemented yet

Manual and automated testing with diff-proxy mode

Continuing with the same example above, we can further examine the diff output while in diff-proxy mode.

  1. Run FerretDB in diff-proxy mode. This can again be achieved by using the --mode flag or by setting the FERRETDB_MODE environment variable to diff-proxy.

  2. Follow the same instructions as the one for diff-normal above to run FerretDB in diff-proxy mode and re-run the query.

    { $sort: { date: 1, author: 1 } },
    $group: {
    _id: '$author',
    firstPost: { $first: '$date' }
    // the query was handled by MongoDB, so the following documents are returned:
    // { _id: 'Alice', firstPost: ISODate("2023-08-20T10:33:23.134Z") }
    // { _id: 'Bob', firstPost: ISODate("2023-08-28T10:33:23.134Z") }

In the diff output below, however, we have discovered that the query cannot be serviced by our application because the $first accumulator operator is not implemented in FerretDB.

Header diff:
--- res header
+++ proxy header
@@ -1 +1 @@
-length: 140, id: 8, response_to: 19, opcode: OP_MSG
+length: 306, id: 128, response_to: 19, opcode: OP_MSG

Body diff:
--- res body
+++ proxy body
@@ -7,6 +7,25 @@
"Document": {
- "ok": 0.0,
- "errmsg": "$group accumulator \"$first\" is not implemented yet",
- "code": 238,
- "codeName": "NotImplemented",
+ "cursor": {
+ "firstBatch": [
+ {
+ "_id": "Bob",
+ "firstPost": 2023-08-28T10:33:23.134Z,
+ },
+ {
+ "_id": "Alice",
+ "firstPost": 2023-08-20T10:33:23.134Z,
+ },
+ ],
+ "id": int64(0),
+ "ns": "test.posts",
+ },
+ "ok": 1.0,

Response metrics

Metrics are captured and written to standard output (stdout) upon exiting in Debug builds. This is a useful way to quickly determine what commands are not implemented for the client requests sent by your application. In the metrics provided below, we can observe that the $first accumulator operator was invoked 18 times within the aggregate command, and the findAndModify command was executed 6 times with a fields projection document. Both of these operations resulted in result="NotImplemented". To address this issue, it's essential to carefully inspect any result that lacks an ok value.

# HELP ferretdb_client_requests_total Total number of requests.
# TYPE ferretdb_client_requests_total counter
ferretdb_client_requests_total{command="aggregate",opcode="OP_MSG"} 105
ferretdb_client_requests_total{command="find",opcode="OP_MSG"} 398
ferretdb_client_requests_total{command="findAndModify",opcode="OP_MSG"} 6
ferretdb_client_requests_total{command="hello",opcode="OP_MSG"} 4
ferretdb_client_requests_total{command="insert",opcode="OP_MSG"} 10
ferretdb_client_requests_total{command="ismaster",opcode="OP_MSG"} 17
ferretdb_client_requests_total{command="unknown",opcode="OP_QUERY"} 28
ferretdb_client_requests_total{command="update",opcode="OP_MSG"} 59
# HELP ferretdb_client_responses_total Total number of responses.
# TYPE ferretdb_client_responses_total counter
ferretdb_client_responses_total{argument="$first (accumulator)",command="aggregate",opcode="OP_MSG",result="NotImplemented"} 18
ferretdb_client_responses_total{argument="fields",command="findAndModify",opcode="OP_MSG",result="NotImplemented"} 6
ferretdb_client_responses_total{argument="unknown",command="aggregate",opcode="OP_MSG",result="ok"} 87
ferretdb_client_responses_total{argument="unknown",command="find",opcode="OP_MSG",result="ok"} 398
ferretdb_client_responses_total{argument="unknown",command="hello",opcode="OP_MSG",result="ok"} 4
ferretdb_client_responses_total{argument="unknown",command="insert",opcode="OP_MSG",result="ok"} 10
ferretdb_client_responses_total{argument="unknown",command="ismaster",opcode="OP_MSG",result="ok"} 17
ferretdb_client_responses_total{argument="unknown",command="unknown",opcode="OP_REPLY",result="ok"} 28
ferretdb_client_responses_total{argument="unknown",command="update",opcode="OP_MSG",result="ok"} 59

Other tools

We also have a fork of the Amazon DocumentDB Compatibility Tool here. The tool examines files to identify queries that use unsupported operators in FerretDB. Please note that this tool is not highly accurate and may generate inaccurate reports, as it does not parse query syntax with contextual information about the originating command. For example, an unsupported operator might appear within a find or aggregate command, which the tool does not differentiate. Note that we also mark operators as unsupported if they are not supported in all commands, which could result in false negatives.

Running the tool to check FerretDB compatibility:

# clone the repository and run the compat-tool
$ git clone && cd amazon-documentdb-tools/compat-tool
$ python3 --directory=/path/to/myapp --version=FerretDB