Post

MongoDB 4.4 Change Streams and Experimental Pre-Image 'Support'

As of MongoDB 6.0, Change Streams Support Pre- and Post-Image Retrieval, DDL operations, and more!

Warning - Not for Production Use (in MongoDB 4.4)

MongoDB’s source code is available (through an SSPL license) and the Core Server project in MongoDB’s JIRA is publicly accessible, which is where I found this information.

Until officially announced as stable/official the methods described herein should not be considered as “production ready”.

This post is for informational purposes and though at the time of writing it I am a MongoDB Inc. employee this should not be considered an official communication.

Introduction

A feature that has long been requested since Change Streams were introduced in MongoDB 3.6 is the ability to support returning the n-1 state (or pre-image) of a document when it is changed.

In SERVER-36941: Option to provide “before image” with change streams this request is captured, but at the time of writing this ticket is still in an Open state which would imply no progress has been made.

Interestingly enough, there actually has been some progress due to efforts surrounding the support of Realm Sync, namely in the following tickets:

Pre-Image Support

“Support” for this feature is only available in MongoDB 4.4+, so first we must ensure we are running a compatible version.

Next, as change streams are only supported in replica sets or sharded clusters our cluster cannot be a standalone instance.

Before a pre-image can be returned in a change stream support for the feature must be enabled at the collection level.

Create a New Collection with Pre-Image Support

To enable pre-image support on a new (non-existent) collection, the recordPreImages flag needs to be set when calling the create command or db.createCollection() shell method:

1
2
3
4
5
6
db.coll1.drop();
// create command
db.runCommand({ create: "coll1", recordPreImages: true });
// or
// using the createCollection() helper
db.createCollection("coll1", { recordPreImages: true });

Updating an Existing Collection with Pre-Image Support

If the collection already exists, the recordPreImages flag can be set using the collMod command:

1
2
3
4
db.coll1.drop();
db.coll1.insert({ _id: 1, created_at: new Date() });
// update the collection's metadata using the collMod command
db.runCommand({ collMod: "coll1", recordPreImages: true });

Pre-Image Support in Change Streams

First it’s important to understand that a change stream is actually an Aggregation Pipeline Stage, even if it doesn’t appear as such in the documentation.

This can be easily verified by checking the source code for the Mongo.prototype.watch implementation which coincides with db.collection.watch() shell method.

We’ll be using the $changeStream pipeline stage directly to review the impact of pre-image support.

First we’ll being by opening a change stream cursor against our modified collection:

1
2
3
4
5
6
db.coll1.drop();
db.createCollection("coll1", { recordPreImages: true });
db.coll1.insert({ _id: 1, created_at: new Date() });
var cursor = db.coll1.aggregate([{
    $changeStream: { fullDocumentBeforeChange: "whenAvailable" } }
]);

Note the options include a fullDocumentBeforeChange field, which can accept one of three (3) fullDocumentBeforeChange mode values:

  • off: Disables support for the fullDocumentBeforeChange field
  • whenAvailable: Only includes a fullDocumentBeforeChange document if it’s available, but won’t fail if it’s not present
  • required: Require the fullDocumentBeforeChange document, and errors out if it’s not available

Since we have a change stream cursor open, we can update our test document and iterate the cursor to see the change this option produces:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
db.coll1.update({ _id: 1 }, { $set: { updated_at: new Date() } })
cursor.next()
/*
{
	"operationType" : "update",
	"fullDocumentBeforeChange" : {
		"_id" : 1,
		"created_at" : ISODate("2020-07-15T17:41:32.043Z")
	},
	"ns" : {
		"db" : "test",
		"coll" : "coll1"
	},
	"documentKey" : {
		"_id" : 1
	},
	"updateDescription" : {
		"updatedFields" : {
			"updated_at" : ISODate("2020-07-15T17:41:32.056Z")
		},
		"removedFields" : [ ]
	}
}
*/

The output contains a fullDocumentBeforeChange field which includes the full document prior to the changes the update operation would apply. As this was the first update to this document and the field was being added for the first time the value here may not be apparent, but running the operation again produces result that contains our previous created_at value along with our updated created_at value in the updateDescription:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
db.coll1.update({ _id: 1 }, { $set: { updated_at: new Date() } })
cursor.next()
/*
{
	"operationType" : "update",
	"fullDocumentBeforeChange" : {
		"_id" : 1,
		"created_at" : ISODate("2020-07-15T17:41:32.043Z"),
		"updated_at" : ISODate("2020-07-15T17:41:32.056Z")
	},
	"ns" : {
		"db" : "test",
		"coll" : "coll1"
	},
	"documentKey" : {
		"_id" : 1
	},
	"updateDescription" : {
		"updatedFields" : {
			"updated_at" : ISODate("2020-07-15T17:44:29.494Z")
		},
		"removedFields" : [ ]
	}
}
*/

If the fullDocumentBeforeChange mode was set to required and the collection wasn’t created with the recordPreImages flag set, the change stream cursor will error out when iterated.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
db.coll2.drop();
db.createCollection("coll2");
db.coll2.insert({ _id: 2, created_at: new Date() });
var cursor = db.coll2.aggregate([{
    $changeStream: { fullDocumentBeforeChange: "required" } }
]);
db.coll2.update({ _id: 2 }, { $set: { updated_at: new Date() } })
cursor.next()
/*
2020-07-16T07:03:37.657-0400 E  QUERY    [js] Error: command failed: {
	"ok" : 0,
	"errmsg" : "Change stream was configured to require a pre-image for all update, delete and replace events, but no pre-image optime was recorded for event: {_id: {_data: \"825F103409000000042B022C0100296E5A10044912D5BB665545B48DDAD38FCD774270461E5F6964002B040004\", _typeBits: BinData(0, \"40\")}, operationType: \"update\", clusterTime: Timestamp(1594897417, 4), ns: {db: \"test\", coll: \"coll2\"}, documentKey: {_id: 2}, updateDescription: {updatedFields: {updated_at: 2020-07-16T11:03:37.611Z}, removedFields: []}}",
	"code" : 51770,
}
*/

Pre-Image and { fullDocument: "updateLookup" }

The change stream cursors we’ve been opening have not been specifying a fullDocument options, which results in the default value of { fullDocument: "default" } being used (or prior to SPEC-909 a value of none).

When we set { fullDocument: "updateLookup" }, the cursor will look up the most current majority-committed version of the updated document and include a fullDocument field with the document lookup in addition to the updateDescription delta.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
db.coll3.drop();
db.createCollection("coll3", { recordPreImages: true });
db.coll3.insert({ _id: 3, name: "Alex", role: "TSE", created_at: new Date() });
var cursor = db.coll3.aggregate([{
    $changeStream: { fullDocument: "updateLookup", fullDocumentBeforeChange: "whenAvailable" } }
]);
db.coll3.update({ _id: 3 }, { $set: { updated_at: new Date() } });
cursor.next()
/*
{
    "_id" : {
        "_data" : "825F103644000000042B022C0100296E5A100452A662230D6B4C08B0AF844900F3A335461E5F6964002B060004",
        "_typeBits" : BinData(0, "QA==")
    },
    "operationType" : "update",
    "clusterTime" : Timestamp(1594897988, 4),
    "fullDocument" : {
        "_id" : 3.0,
        "name" : "Alex",
        "role" : "TSE",
        "created_at" : ISODate("2020-07-16T11:13:08.180+0000"),
        "updated_at" : ISODate("2020-07-16T11:13:08.268+0000")
    },
    "fullDocumentBeforeChange" : {
        "_id" : 3.0,
        "name" : "Alex",
        "role" : "TSE",
        "created_at" : ISODate("2020-07-16T11:13:08.180+0000")
    },
    "ns" : {
        "db" : "test",
        "coll" : "coll3"
    },
    "documentKey" : {
        "_id" : 3.0
    },
    "updateDescription" : {
        "updatedFields" : {
            "updated_at" : ISODate("2020-07-16T11:13:08.268+0000")
        },
        "removedFields" : [

        ]
    }
}
*/

By including both fullDocument and fullDocumentBeforeChange options, the cursor now returns three fields that represent the state of the document before the change, after the change as well as a description of the changes in the form of the updateDescription field.

How are Pre-Images Stored

Since change streams are messages in the oplog, which are themselves BSON documents they must adhere to the BSON Document Size limit of 16 megabytes.

Pre-image details aren’t stored directly within the oplog entry that describes the update or delete (op type of u or d).

When recordPreImages has been enabled for a collection, prior to an oplog entry for an update or delete, a noop (op: "n") is inserted into the oplog with current document state. The subsequent operation (the actual update or delete) will contain a preImageOpTime which contains the optime of the oplog entry containing this noop.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
db.getSiblingDB("local").getCollection("oplog.rs").find({ ns: "test.coll3" });
/*
{
  "op": "n",
  "ns": "test.coll3",
  "ui": UUID("d2c8288d-2748-4abf-b5f0-37de24e3128e"),
  "o": {
    "_id": 3,
    "name": "Alex",
    "role": "TSE",
    "created_at": ISODate("2020-07-17T13:09:52.612Z")
  },
  "ts": Timestamp(1594991392, 4),
  "t": NumberLong(4),
  "wall": ISODate("2020-07-17T13:09:52.649Z"),
  "v": NumberLong(2)
}
{
  "op": "u",
  "ns": "test.coll3",
  "ui": UUID("d2c8288d-2748-4abf-b5f0-37de24e3128e"),
  "o": {
    "$v": 1,
    "$set": {
      "updated_at": ISODate("2020-07-17T13:09:52.648Z")
    }
  },
  "o2": {
    "_id": 3
  },
  "preImageOpTime": {
    "ts": Timestamp(1594991392, 4),
    "t": NumberLong(4)
  },
  "ts": Timestamp(1594991392, 5),
  "t": NumberLong(4),
  "wall": ISODate("2020-07-17T13:09:52.649Z"),
  "v": NumberLong(2)
}
*/

What Could Go Wrong?

Though enabling pre-images won’t bloat the oplog entry to the point where it may exceed the BSON max size, when the change stream cursor resolves resulting document the size cannot exceed the BSON max size.

For example:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
function randomString(length) {
   var result           = '';
   var characters       = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789';
   var charactersLength = characters.length;
   for ( var i = 0; i < length; i++ ) {
      result += characters.charAt(Math.floor(Math.random() * charactersLength));
   }
   return result;
}

// BSON max size less 512 bytes
var size = db.serverBuildInfo().maxBsonObjectSize - 512;

db.coll4.drop();
db.createCollection("coll4", { recordPreImages: true });
db.coll4.insert({ _id: 4, created_at: new Date(), junk: randomString(size) });
var cursor = db.coll4.aggregate([{
    $changeStream: { fullDocument: "updateLookup", fullDocumentBeforeChange: "off" } }
]);
db.coll4.update({ _id: 4 }, { $set: { updated_at: new Date() } });
cursor.next()

The above example has the fullDocumentBeforeChange disabled, which is the current default behavior of change streams. Though the result would be large (almost 16MB), iterating the change stream cursor would produce a result.

Modifying the cursor to now request a fullDocumentBeforeChange (either required or whenAvailable) would now raise an error when the cursor is iterated.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
// ...
db.coll4.drop();
db.createCollection("coll4", { recordPreImages: true });
db.coll4.insert({ _id: 4, created_at: new Date(), junk: randomString(size) });
var cursor = db.coll4.aggregate([{
    $changeStream: { fullDocument: "updateLookup", fullDocumentBeforeChange: "required" } }
]);
db.coll4.update({ _id: 4 }, { $set: { updated_at: new Date() } });
cursor.next();
/*
2020-07-16T07:40:38.307-0400 E  QUERY    [js] Error: command failed: {
	"errmsg" : "BSONObj size: 33553900 (0x1FFFDEC) is invalid. Size must be between 0 and 16793600(16MB) First element: _id: { _data: \"825F103CB5000000032B022C0100296E5A100484010E04E239427B8CBCA810283D05A5461E5F6964002B080004\", _typeBits: BinData(0, 40) }",
	"code" : 10334,
	"codeName" : "BSONObjectTooLarge",
}
*/

I personally look forward to pre-images being officially supported in MongoDB, however in the meantime be advised that this may not be ideal for production just yet.

Happy Coding!

This post is licensed under CC BY 4.0 by the author.

Comments powered by Disqus.