mongopatch

MongoDB patching tool. Update and log mongodb documents.

npm install -g mongopatch

!!WARNING!!

Module has broken tests and probably functionality on MongoDB versions 2.6.* and 3.4.*. Versions 3.0.* and 3.2.* seems to be fine. See travis build for more details.

Writing patches

Patches are written as separate modules, exposing a single patching function.

module.exports = function(patch) {
	// Specify which patching system version to use for this patch (required)
	patch.version('0.1.0');

	// Update all users that match the provided query.
	// The query is optional, if not provided all the documents
	// in the collection are processed.
	patch.update('users', { name: 'e-conomic' }, function(document, callback) {
		// The callback function should be called with the update to apply,
		// this can be any valid mongodb update query.
		callback(null, { $set: { email: '[email protected]', associates: 'unknown' } });
	});

	// Register an after callback, to be run after each update.
	patch.after(function(update, callback) {
		var isValid = update.after.email === '[email protected]';

		// Call the callback function with an error to abort the patching process.
		// Use this to guard against corrupted updates.
		callback(isValid ? null : new Error('Update failed'));
	});
}

The after callback gets an options map, containing the before and after documents, a modfifed flag (telling if there any changes between the two documents) and a diff object (the diff between the two documents).

Another example where we process all users.

function shouldUpdate(document) {
	// ...
}

function update(document) {
	// ...
}

function isValid(document) {
	// ...
}

module.exports = function(patch) {
	patch.version('0.1.0');

	// All users are processed, since no filter query provided.
	patch.update('users', function(document, callback) {
		if(!shouldUpdate(document)) {
			// Calling the callback with no arguments, skips the document in the update process.
			return callback();
		}

		update(document);

		if(!isValid(document)) {
			// Validate document before performing the actual update in the database.
			// Passing an error as first argument, aborts the patching process,
			// and can leave the database in inconsistent state.
			return callback(new Error('Invalid document'));
		}

		// Apply the update, by overwritting the whole document
		callback(null, document);
	});
}

It's also possible to register setup and teardown hooks, executed before and after the patch is run. The teardown callback gets called with an additional stats object, containg accumulated details about the patch.

patch.setup(function(callback) {
	// Pass an error object as first argument to callback, to terminate execution.
	callback();
});

patch.teardown(function(stats, callback) {
	// Stats contains details about execution time, number of modified documents and average speed.
	callback();
});

Runing patches

Run patches using the mongopatch command-line tool. Basic usage:

mongopatch path/to/patch.js --db "..." --dry-run --log-db "..."

Available options (too see a full list of options, run mongopatch without any arguments).

• db: MongoDB connection string (e.g. user:password@localhost:27017/development or development).
• log-db: MongoDB connection string for the log database. When provided a version of the document is stored before and after the update.
• dry-run: Do not perform any changes in the database. Changes are performed on copy of the documents and stored in the log db (if available).
• parallel: Run the patch with given parallelism. It may run the patch faster.
• update: Run the patch with one of the available update modes: dummy, query or document (default).
• diff-object: Use objects instead of arrays in document diffs.

Update option

Three update modes are available. query and document both perform real updates on the database. dummy mode is the same as specifying the --dry-run option. Note also that --dry-run overrides --update.

When performing updates on real data, external changes may occur, modifying the documents as they are being processed. When mongopatch is started it fetches all the documents matching the provided query. These are loaded in batches and there can be a significant amount of time between, when a document is loaded and when the actual update is performed. To prevent external updates from conflicting with patching two strategies are employed.

The query mode uses the document's _id property and the query, originally provided to the patch.update method, as the criteria for finding and modifying the document (findAndModify MongoDB command). This means if the document has been changed externally, so that it no longer satisfies the query, it will be skipped. Other external changes to the document aren't considered.

The document mode, on the other hand, uses the query and the whole document as the criteria. External changes to document will prevent the document from being patched (an exception to this is addition of properties to the root of the document which pass the query). If that occurs, the document is fetched again using the _id property and the original query (similiar when in query mode), and run through the worker function again (the function passed to patch.update). If the worker function returns a modifier, the whole proccess is repeated with the new document and modifier. This has the consequence, that the worker function can be called with the same document multiple times in arbitrary order. This could affect patches with some form of state (e.g. counting number of documents by incrementing a counter every time the worker function has been called).

Runing updates in query mode:

mongopatch path/to/patch.js --db "..." --log-db "..." --update query

diff-object option

Usually changes to arrays are also stored as arrays in the diff (truncated if necessary to avoid empty entries in the diff array). But in cases where changed objects contain numeric keys, the current algorithm uses arrays for the diff. This will allocate an array that is at least as big as the key, and can cause memory and cpu problems.

When diff-object is enabled only objects are used in the diff. This means also changes to arrays are stored as objects, which is more effective, but might be harder to query in the log database.

CLI

The tool has a simple command-line interface, where it is possible to track progress and accumulated changes done to the documents. The diff shows how many times a property has been added, updated or removed between the original and the updated documents (note that all array changes are grouped).

When running on a live database, where external changes can occur, the progress indicator may be incorrect, as documents can be added or removed. Also skipping documents in patch.update causes the progress to fall behind.

Log database

When a log database is available, a collection is created for every patch run. A document in the patch collection, contains data about the applied update. The before key points to the original document, after to the updated document, modified is a boolean flag telling if there were any changes and diff the difference between the before and after document (if modified is false, this is going to be an empty object). It also includes additional meta data.

{
	"before": {
		"_id": ObjectId("507d2a650ea37a02000001ae"),
		"name": "e-conomic",
		"associates": "debitoor"
	},
	"after": {
		"_id": ObjectId("507d2a650ea37a02000001ae"),
		"name": "e-conomic",
		"associates": "unknown",
		"email": "[email protected]"
	},
	"modified": true,
	"skipped": false,
	"diff": {																	// diff is a nested object, where leafs can have one of the three values added, updated, removed
		"associates": "updated",
		"email": "added"
	},
	"createdAt": ISODate("2013-12-17T15:28:14.737Z"),							// when was the log document created
	"collection": "development.users",											// full collection name
	"modifier": "{ \"$set\": { \"email\": \"[email protected]\" } }",		// stringified modifier (passed to the callback in patch.update)
	"query": "{ \"name\": \"e-conomic\" }"										// stringified query (passed to patch.update function)
}

In some cases if an error occures during the patching, an error object is added to the log document, containing the error message and stack.

License

MIT