Files
meteor/packages/mongo/mongo_driver.js
David Glasser 63a0076305 mongo: provide direct access to node driver db
Specifically, Mongo.Collection objects on the server now have
rawCollection and rawDatabase methods.

You can use MongoInternals.NpmModules.mongodb.version to tell what
version of the mongodb npm module is the backend for HTTP.call.  This
version may change incompatibly from version to version of Meteor; use
at your own risk.  (For example, we expect to upgrade from the 1.4.x
series to the 2.x series in the not-too-distant future.)

Fixes #3640.
2015-03-05 17:12:37 -08:00

1280 lines
44 KiB
JavaScript

/**
* Provide a synchronous Collection API using fibers, backed by
* MongoDB. This is only for use on the server, and mostly identical
* to the client API.
*
* NOTE: the public API methods must be run within a fiber. If you call
* these outside of a fiber they will explode!
*/
var path = Npm.require('path');
var MongoDB = Npm.require('mongodb');
var Fiber = Npm.require('fibers');
var Future = Npm.require(path.join('fibers', 'future'));
MongoInternals = {};
MongoTest = {};
MongoInternals.NpmModules = {
mongodb: {
version: Npm.require('mongodb/package.json').version,
module: MongoDB
}
};
// Older version of what is now available via
// MongoInternals.NpmModules.mongodb.module. It was never documented, but
// people do use it.
// XXX COMPAT WITH 1.0.3.2
MongoInternals.NpmModule = MongoDB;
// This is used to add or remove EJSON from the beginning of everything nested
// inside an EJSON custom type. It should only be called on pure JSON!
var replaceNames = function (filter, thing) {
if (typeof thing === "object") {
if (_.isArray(thing)) {
return _.map(thing, _.bind(replaceNames, null, filter));
}
var ret = {};
_.each(thing, function (value, key) {
ret[filter(key)] = replaceNames(filter, value);
});
return ret;
}
return thing;
};
// Ensure that EJSON.clone keeps a Timestamp as a Timestamp (instead of just
// doing a structural clone).
// XXX how ok is this? what if there are multiple copies of MongoDB loaded?
MongoDB.Timestamp.prototype.clone = function () {
// Timestamps should be immutable.
return this;
};
var makeMongoLegal = function (name) { return "EJSON" + name; };
var unmakeMongoLegal = function (name) { return name.substr(5); };
var replaceMongoAtomWithMeteor = function (document) {
if (document instanceof MongoDB.Binary) {
var buffer = document.value(true);
return new Uint8Array(buffer);
}
if (document instanceof MongoDB.ObjectID) {
return new Mongo.ObjectID(document.toHexString());
}
if (document["EJSON$type"] && document["EJSON$value"]
&& _.size(document) === 2) {
return EJSON.fromJSONValue(replaceNames(unmakeMongoLegal, document));
}
if (document instanceof MongoDB.Timestamp) {
// For now, the Meteor representation of a Mongo timestamp type (not a date!
// this is a weird internal thing used in the oplog!) is the same as the
// Mongo representation. We need to do this explicitly or else we would do a
// structural clone and lose the prototype.
return document;
}
return undefined;
};
var replaceMeteorAtomWithMongo = function (document) {
if (EJSON.isBinary(document)) {
// This does more copies than we'd like, but is necessary because
// MongoDB.BSON only looks like it takes a Uint8Array (and doesn't actually
// serialize it correctly).
return new MongoDB.Binary(new Buffer(document));
}
if (document instanceof Mongo.ObjectID) {
return new MongoDB.ObjectID(document.toHexString());
}
if (document instanceof MongoDB.Timestamp) {
// For now, the Meteor representation of a Mongo timestamp type (not a date!
// this is a weird internal thing used in the oplog!) is the same as the
// Mongo representation. We need to do this explicitly or else we would do a
// structural clone and lose the prototype.
return document;
}
if (EJSON._isCustomType(document)) {
return replaceNames(makeMongoLegal, EJSON.toJSONValue(document));
}
// It is not ordinarily possible to stick dollar-sign keys into mongo
// so we don't bother checking for things that need escaping at this time.
return undefined;
};
var replaceTypes = function (document, atomTransformer) {
if (typeof document !== 'object' || document === null)
return document;
var replacedTopLevelAtom = atomTransformer(document);
if (replacedTopLevelAtom !== undefined)
return replacedTopLevelAtom;
var ret = document;
_.each(document, function (val, key) {
var valReplaced = replaceTypes(val, atomTransformer);
if (val !== valReplaced) {
// Lazy clone. Shallow copy.
if (ret === document)
ret = _.clone(document);
ret[key] = valReplaced;
}
});
return ret;
};
MongoConnection = function (url, options) {
var self = this;
options = options || {};
self._observeMultiplexers = {};
self._onFailoverHook = new Hook;
var mongoOptions = {db: {safe: true}, server: {}, replSet: {}};
// Set autoReconnect to true, unless passed on the URL. Why someone
// would want to set autoReconnect to false, I'm not really sure, but
// keeping this for backwards compatibility for now.
if (!(/[\?&]auto_?[rR]econnect=/.test(url))) {
mongoOptions.server.auto_reconnect = true;
}
// Disable the native parser by default, unless specifically enabled
// in the mongo URL.
// - The native driver can cause errors which normally would be
// thrown, caught, and handled into segfaults that take down the
// whole app.
// - Binary modules don't yet work when you bundle and move the bundle
// to a different platform (aka deploy)
// We should revisit this after binary npm module support lands.
if (!(/[\?&]native_?[pP]arser=/.test(url))) {
mongoOptions.db.native_parser = false;
}
// XXX maybe we should have a better way of allowing users to configure the
// underlying Mongo driver
if (_.has(options, 'poolSize')) {
// If we just set this for "server", replSet will override it. If we just
// set it for replSet, it will be ignored if we're not using a replSet.
mongoOptions.server.poolSize = options.poolSize;
mongoOptions.replSet.poolSize = options.poolSize;
}
self.db = null;
// We keep track of the ReplSet's primary, so that we can trigger hooks when
// it changes. The Node driver's joined callback seems to fire way too
// often, which is why we need to track it ourselves.
self._primary = null;
self._oplogHandle = null;
self._docFetcher = null;
var connectFuture = new Future;
MongoDB.connect(
url,
mongoOptions,
Meteor.bindEnvironment(
function (err, db) {
if (err) {
throw err;
}
// First, figure out what the current primary is, if any.
if (db.serverConfig._state.master)
self._primary = db.serverConfig._state.master.name;
db.serverConfig.on(
'joined', Meteor.bindEnvironment(function (kind, doc) {
if (kind === 'primary') {
if (doc.primary !== self._primary) {
self._primary = doc.primary;
self._onFailoverHook.each(function (callback) {
callback();
return true;
});
}
} else if (doc.me === self._primary) {
// The thing we thought was primary is now something other than
// primary. Forget that we thought it was primary. (This means
// that if a server stops being primary and then starts being
// primary again without another server becoming primary in the
// middle, we'll correctly count it as a failover.)
self._primary = null;
}
}));
// Allow the constructor to return.
connectFuture['return'](db);
},
connectFuture.resolver() // onException
)
);
// Wait for the connection to be successful; throws on failure.
self.db = connectFuture.wait();
if (options.oplogUrl && ! Package['disable-oplog']) {
self._oplogHandle = new OplogHandle(options.oplogUrl, self.db.databaseName);
self._docFetcher = new DocFetcher(self);
}
};
MongoConnection.prototype.close = function() {
var self = this;
if (! self.db)
throw Error("close called before Connection created?");
// XXX probably untested
var oplogHandle = self._oplogHandle;
self._oplogHandle = null;
if (oplogHandle)
oplogHandle.stop();
// Use Future.wrap so that errors get thrown. This happens to
// work even outside a fiber since the 'close' method is not
// actually asynchronous.
Future.wrap(_.bind(self.db.close, self.db))(true).wait();
};
// Returns the Mongo Collection object; may yield.
MongoConnection.prototype.rawCollection = function (collectionName) {
var self = this;
if (! self.db)
throw Error("rawCollection called before Connection created?");
var future = new Future;
self.db.collection(collectionName, future.resolver());
return future.wait();
};
MongoConnection.prototype._createCappedCollection = function (
collectionName, byteSize, maxDocuments) {
var self = this;
if (! self.db)
throw Error("_createCappedCollection called before Connection created?");
var future = new Future();
self.db.createCollection(
collectionName,
{ capped: true, size: byteSize, max: maxDocuments },
future.resolver());
future.wait();
};
// This should be called synchronously with a write, to create a
// transaction on the current write fence, if any. After we can read
// the write, and after observers have been notified (or at least,
// after the observer notifiers have added themselves to the write
// fence), you should call 'committed()' on the object returned.
MongoConnection.prototype._maybeBeginWrite = function () {
var self = this;
var fence = DDPServer._CurrentWriteFence.get();
if (fence)
return fence.beginWrite();
else
return {committed: function () {}};
};
// Internal interface: adds a callback which is called when the Mongo primary
// changes. Returns a stop handle.
MongoConnection.prototype._onFailover = function (callback) {
return this._onFailoverHook.register(callback);
};
//////////// Public API //////////
// The write methods block until the database has confirmed the write (it may
// not be replicated or stable on disk, but one server has confirmed it) if no
// callback is provided. If a callback is provided, then they call the callback
// when the write is confirmed. They return nothing on success, and raise an
// exception on failure.
//
// After making a write (with insert, update, remove), observers are
// notified asynchronously. If you want to receive a callback once all
// of the observer notifications have landed for your write, do the
// writes inside a write fence (set DDPServer._CurrentWriteFence to a new
// _WriteFence, and then set a callback on the write fence.)
//
// Since our execution environment is single-threaded, this is
// well-defined -- a write "has been made" if it's returned, and an
// observer "has been notified" if its callback has returned.
var writeCallback = function (write, refresh, callback) {
return function (err, result) {
if (! err) {
// XXX We don't have to run this on error, right?
refresh();
}
write.committed();
if (callback)
callback(err, result);
else if (err)
throw err;
};
};
var bindEnvironmentForWrite = function (callback) {
return Meteor.bindEnvironment(callback, "Mongo write");
};
MongoConnection.prototype._insert = function (collection_name, document,
callback) {
var self = this;
var sendError = function (e) {
if (callback)
return callback(e);
throw e;
};
if (collection_name === "___meteor_failure_test_collection") {
var e = new Error("Failure test");
e.expected = true;
sendError(e);
return;
}
if (!(LocalCollection._isPlainObject(document) &&
!EJSON._isCustomType(document))) {
sendError(new Error(
"Only plain objects may be inserted into MongoDB"));
return;
}
var write = self._maybeBeginWrite();
var refresh = function () {
Meteor.refresh({collection: collection_name, id: document._id });
};
callback = bindEnvironmentForWrite(writeCallback(write, refresh, callback));
try {
var collection = self.rawCollection(collection_name);
collection.insert(replaceTypes(document, replaceMeteorAtomWithMongo),
{safe: true}, callback);
} catch (e) {
write.committed();
throw e;
}
};
// Cause queries that may be affected by the selector to poll in this write
// fence.
MongoConnection.prototype._refresh = function (collectionName, selector) {
var self = this;
var refreshKey = {collection: collectionName};
// If we know which documents we're removing, don't poll queries that are
// specific to other documents. (Note that multiple notifications here should
// not cause multiple polls, since all our listener is doing is enqueueing a
// poll.)
var specificIds = LocalCollection._idsMatchedBySelector(selector);
if (specificIds) {
_.each(specificIds, function (id) {
Meteor.refresh(_.extend({id: id}, refreshKey));
});
} else {
Meteor.refresh(refreshKey);
}
};
MongoConnection.prototype._remove = function (collection_name, selector,
callback) {
var self = this;
if (collection_name === "___meteor_failure_test_collection") {
var e = new Error("Failure test");
e.expected = true;
if (callback)
return callback(e);
else
throw e;
}
var write = self._maybeBeginWrite();
var refresh = function () {
self._refresh(collection_name, selector);
};
callback = bindEnvironmentForWrite(writeCallback(write, refresh, callback));
try {
var collection = self.rawCollection(collection_name);
collection.remove(replaceTypes(selector, replaceMeteorAtomWithMongo),
{safe: true}, callback);
} catch (e) {
write.committed();
throw e;
}
};
MongoConnection.prototype._dropCollection = function (collectionName, cb) {
var self = this;
var write = self._maybeBeginWrite();
var refresh = function () {
Meteor.refresh({collection: collectionName, id: null,
dropCollection: true});
};
cb = bindEnvironmentForWrite(writeCallback(write, refresh, cb));
try {
var collection = self.rawCollection(collectionName);
collection.drop(cb);
} catch (e) {
write.committed();
throw e;
}
};
MongoConnection.prototype._update = function (collection_name, selector, mod,
options, callback) {
var self = this;
if (! callback && options instanceof Function) {
callback = options;
options = null;
}
if (collection_name === "___meteor_failure_test_collection") {
var e = new Error("Failure test");
e.expected = true;
if (callback)
return callback(e);
else
throw e;
}
// explicit safety check. null and undefined can crash the mongo
// driver. Although the node driver and minimongo do 'support'
// non-object modifier in that they don't crash, they are not
// meaningful operations and do not do anything. Defensively throw an
// error here.
if (!mod || typeof mod !== 'object')
throw new Error("Invalid modifier. Modifier must be an object.");
if (!(LocalCollection._isPlainObject(mod) &&
!EJSON._isCustomType(mod))) {
throw new Error(
"Only plain objects may be used as replacement" +
" documents in MongoDB");
return;
}
if (!options) options = {};
var write = self._maybeBeginWrite();
var refresh = function () {
self._refresh(collection_name, selector);
};
callback = writeCallback(write, refresh, callback);
try {
var collection = self.rawCollection(collection_name);
var mongoOpts = {safe: true};
// explictly enumerate options that minimongo supports
if (options.upsert) mongoOpts.upsert = true;
if (options.multi) mongoOpts.multi = true;
// Lets you get a more more full result from MongoDB. Use with caution:
// might not work with C.upsert (as opposed to C.update({upsert:true}) or
// with simulated upsert.
if (options.fullResult) mongoOpts.fullResult = true;
var mongoSelector = replaceTypes(selector, replaceMeteorAtomWithMongo);
var mongoMod = replaceTypes(mod, replaceMeteorAtomWithMongo);
var isModify = isModificationMod(mongoMod);
var knownId = selector._id || mod._id;
if (options._forbidReplace && ! isModify) {
var e = new Error("Invalid modifier. Replacements are forbidden.");
if (callback) {
return callback(e);
} else {
throw e;
}
}
if (options.upsert && (! knownId) && options.insertedId) {
// XXX If we know we're using Mongo 2.6 (and this isn't a replacement)
// we should be able to just use $setOnInsert instead of this
// simulated upsert thing. (We can't use $setOnInsert with
// replacements because there's nowhere to write it, and $setOnInsert
// can't set _id on Mongo 2.4.)
//
// Also, in the future we could do a real upsert for the mongo id
// generation case, if the the node mongo driver gives us back the id
// of the upserted doc (which our current version does not).
//
// For more context, see
// https://github.com/meteor/meteor/issues/2278#issuecomment-64252706
simulateUpsertWithInsertedId(
collection, mongoSelector, mongoMod,
isModify, options,
// This callback does not need to be bindEnvironment'ed because
// simulateUpsertWithInsertedId() wraps it and then passes it through
// bindEnvironmentForWrite.
function (err, result) {
// If we got here via a upsert() call, then options._returnObject will
// be set and we should return the whole object. Otherwise, we should
// just return the number of affected docs to match the mongo API.
if (result && ! options._returnObject)
callback(err, result.numberAffected);
else
callback(err, result);
}
);
} else {
collection.update(
mongoSelector, mongoMod, mongoOpts,
bindEnvironmentForWrite(function (err, result, extra) {
if (! err) {
if (result && options._returnObject) {
result = { numberAffected: result };
// If this was an upsert() call, and we ended up
// inserting a new doc and we know its id, then
// return that id as well.
if (options.upsert && knownId &&
! extra.updatedExisting)
result.insertedId = knownId;
}
}
callback(err, result);
}));
}
} catch (e) {
write.committed();
throw e;
}
};
var isModificationMod = function (mod) {
var isReplace = false;
var isModify = false;
for (var k in mod) {
if (k.substr(0, 1) === '$') {
isModify = true;
} else {
isReplace = true;
}
}
if (isModify && isReplace) {
throw new Error(
"Update parameter cannot have both modifier and non-modifier fields.");
}
return isModify;
};
var NUM_OPTIMISTIC_TRIES = 3;
// exposed for testing
MongoConnection._isCannotChangeIdError = function (err) {
// First check for what this error looked like in Mongo 2.4. Either of these
// checks should work, but just to be safe...
if (err.code === 13596)
return true;
if (err.err.indexOf("cannot change _id of a document") === 0)
return true;
// Now look for what it looks like in Mongo 2.6. We don't use the error code
// here, because the error code we observed it producing (16837) appears to be
// a far more generic error code based on examining the source.
if (err.err.indexOf("The _id field cannot be changed") === 0)
return true;
return false;
};
var simulateUpsertWithInsertedId = function (collection, selector, mod,
isModify, options, callback) {
// STRATEGY: First try doing a plain update. If it affected 0 documents,
// then without affecting the database, we know we should probably do an
// insert. We then do a *conditional* insert that will fail in the case
// of a race condition. This conditional insert is actually an
// upsert-replace with an _id, which will never successfully update an
// existing document. If this upsert fails with an error saying it
// couldn't change an existing _id, then we know an intervening write has
// caused the query to match something. We go back to step one and repeat.
// Like all "optimistic write" schemes, we rely on the fact that it's
// unlikely our writes will continue to be interfered with under normal
// circumstances (though sufficiently heavy contention with writers
// disagreeing on the existence of an object will cause writes to fail
// in theory).
var newDoc;
// Run this code up front so that it fails fast if someone uses
// a Mongo update operator we don't support.
if (isModify) {
// We've already run replaceTypes/replaceMeteorAtomWithMongo on
// selector and mod. We assume it doesn't matter, as far as
// the behavior of modifiers is concerned, whether `_modify`
// is run on EJSON or on mongo-converted EJSON.
var selectorDoc = LocalCollection._removeDollarOperators(selector);
LocalCollection._modify(selectorDoc, mod, {isInsert: true});
newDoc = selectorDoc;
} else {
newDoc = mod;
}
var insertedId = options.insertedId; // must exist
var mongoOptsForUpdate = {
safe: true,
multi: options.multi
};
var mongoOptsForInsert = {
safe: true,
upsert: true
};
var tries = NUM_OPTIMISTIC_TRIES;
var doUpdate = function () {
tries--;
if (! tries) {
callback(new Error("Upsert failed after " + NUM_OPTIMISTIC_TRIES + " tries."));
} else {
collection.update(selector, mod, mongoOptsForUpdate,
bindEnvironmentForWrite(function (err, result) {
if (err)
callback(err);
else if (result)
callback(null, {
numberAffected: result
});
else
doConditionalInsert();
}));
}
};
var doConditionalInsert = function () {
var replacementWithId = _.extend(
replaceTypes({_id: insertedId}, replaceMeteorAtomWithMongo),
newDoc);
collection.update(selector, replacementWithId, mongoOptsForInsert,
bindEnvironmentForWrite(function (err, result) {
if (err) {
// figure out if this is a
// "cannot change _id of document" error, and
// if so, try doUpdate() again, up to 3 times.
if (MongoConnection._isCannotChangeIdError(err)) {
doUpdate();
} else {
callback(err);
}
} else {
callback(null, {
numberAffected: result,
insertedId: insertedId
});
}
}));
};
doUpdate();
};
_.each(["insert", "update", "remove", "dropCollection"], function (method) {
MongoConnection.prototype[method] = function (/* arguments */) {
var self = this;
return Meteor.wrapAsync(self["_" + method]).apply(self, arguments);
};
});
// XXX MongoConnection.upsert() does not return the id of the inserted document
// unless you set it explicitly in the selector or modifier (as a replacement
// doc).
MongoConnection.prototype.upsert = function (collectionName, selector, mod,
options, callback) {
var self = this;
if (typeof options === "function" && ! callback) {
callback = options;
options = {};
}
return self.update(collectionName, selector, mod,
_.extend({}, options, {
upsert: true,
_returnObject: true
}), callback);
};
MongoConnection.prototype.find = function (collectionName, selector, options) {
var self = this;
if (arguments.length === 1)
selector = {};
return new Cursor(
self, new CursorDescription(collectionName, selector, options));
};
MongoConnection.prototype.findOne = function (collection_name, selector,
options) {
var self = this;
if (arguments.length === 1)
selector = {};
options = options || {};
options.limit = 1;
return self.find(collection_name, selector, options).fetch()[0];
};
// We'll actually design an index API later. For now, we just pass through to
// Mongo's, but make it synchronous.
MongoConnection.prototype._ensureIndex = function (collectionName, index,
options) {
var self = this;
options = _.extend({safe: true}, options);
// We expect this function to be called at startup, not from within a method,
// so we don't interact with the write fence.
var collection = self.rawCollection(collectionName);
var future = new Future;
var indexName = collection.ensureIndex(index, options, future.resolver());
future.wait();
};
MongoConnection.prototype._dropIndex = function (collectionName, index) {
var self = this;
// This function is only used by test code, not within a method, so we don't
// interact with the write fence.
var collection = self.rawCollection(collectionName);
var future = new Future;
var indexName = collection.dropIndex(index, future.resolver());
future.wait();
};
// CURSORS
// There are several classes which relate to cursors:
//
// CursorDescription represents the arguments used to construct a cursor:
// collectionName, selector, and (find) options. Because it is used as a key
// for cursor de-dup, everything in it should either be JSON-stringifiable or
// not affect observeChanges output (eg, options.transform functions are not
// stringifiable but do not affect observeChanges).
//
// SynchronousCursor is a wrapper around a MongoDB cursor
// which includes fully-synchronous versions of forEach, etc.
//
// Cursor is the cursor object returned from find(), which implements the
// documented Mongo.Collection cursor API. It wraps a CursorDescription and a
// SynchronousCursor (lazily: it doesn't contact Mongo until you call a method
// like fetch or forEach on it).
//
// ObserveHandle is the "observe handle" returned from observeChanges. It has a
// reference to an ObserveMultiplexer.
//
// ObserveMultiplexer allows multiple identical ObserveHandles to be driven by a
// single observe driver.
//
// There are two "observe drivers" which drive ObserveMultiplexers:
// - PollingObserveDriver caches the results of a query and reruns it when
// necessary.
// - OplogObserveDriver follows the Mongo operation log to directly observe
// database changes.
// Both implementations follow the same simple interface: when you create them,
// they start sending observeChanges callbacks (and a ready() invocation) to
// their ObserveMultiplexer, and you stop them by calling their stop() method.
CursorDescription = function (collectionName, selector, options) {
var self = this;
self.collectionName = collectionName;
self.selector = Mongo.Collection._rewriteSelector(selector);
self.options = options || {};
};
Cursor = function (mongo, cursorDescription) {
var self = this;
self._mongo = mongo;
self._cursorDescription = cursorDescription;
self._synchronousCursor = null;
};
_.each(['forEach', 'map', 'fetch', 'count'], function (method) {
Cursor.prototype[method] = function () {
var self = this;
// You can only observe a tailable cursor.
if (self._cursorDescription.options.tailable)
throw new Error("Cannot call " + method + " on a tailable cursor");
if (!self._synchronousCursor) {
self._synchronousCursor = self._mongo._createSynchronousCursor(
self._cursorDescription, {
// Make sure that the "self" argument to forEach/map callbacks is the
// Cursor, not the SynchronousCursor.
selfForIteration: self,
useTransform: true
});
}
return self._synchronousCursor[method].apply(
self._synchronousCursor, arguments);
};
});
// Since we don't actually have a "nextObject" interface, there's really no
// reason to have a "rewind" interface. All it did was make multiple calls
// to fetch/map/forEach return nothing the second time.
// XXX COMPAT WITH 0.8.1
Cursor.prototype.rewind = function () {
};
Cursor.prototype.getTransform = function () {
return this._cursorDescription.options.transform;
};
// When you call Meteor.publish() with a function that returns a Cursor, we need
// to transmute it into the equivalent subscription. This is the function that
// does that.
Cursor.prototype._publishCursor = function (sub) {
var self = this;
var collection = self._cursorDescription.collectionName;
return Mongo.Collection._publishCursor(self, sub, collection);
};
// Used to guarantee that publish functions return at most one cursor per
// collection. Private, because we might later have cursors that include
// documents from multiple collections somehow.
Cursor.prototype._getCollectionName = function () {
var self = this;
return self._cursorDescription.collectionName;
}
Cursor.prototype.observe = function (callbacks) {
var self = this;
return LocalCollection._observeFromObserveChanges(self, callbacks);
};
Cursor.prototype.observeChanges = function (callbacks) {
var self = this;
var ordered = LocalCollection._observeChangesCallbacksAreOrdered(callbacks);
return self._mongo._observeChanges(
self._cursorDescription, ordered, callbacks);
};
MongoConnection.prototype._createSynchronousCursor = function(
cursorDescription, options) {
var self = this;
options = _.pick(options || {}, 'selfForIteration', 'useTransform');
var collection = self.rawCollection(cursorDescription.collectionName);
var cursorOptions = cursorDescription.options;
var mongoOptions = {
sort: cursorOptions.sort,
limit: cursorOptions.limit,
skip: cursorOptions.skip
};
// Do we want a tailable cursor (which only works on capped collections)?
if (cursorOptions.tailable) {
// We want a tailable cursor...
mongoOptions.tailable = true;
// ... and for the server to wait a bit if any getMore has no data (rather
// than making us put the relevant sleeps in the client)...
mongoOptions.awaitdata = true;
// ... and to keep querying the server indefinitely rather than just 5 times
// if there's no more data.
mongoOptions.numberOfRetries = -1;
// And if this is on the oplog collection and the cursor specifies a 'ts',
// then set the undocumented oplog replay flag, which does a special scan to
// find the first document (instead of creating an index on ts). This is a
// very hard-coded Mongo flag which only works on the oplog collection and
// only works with the ts field.
if (cursorDescription.collectionName === OPLOG_COLLECTION &&
cursorDescription.selector.ts) {
mongoOptions.oplogReplay = true;
}
}
var dbCursor = collection.find(
replaceTypes(cursorDescription.selector, replaceMeteorAtomWithMongo),
cursorOptions.fields, mongoOptions);
return new SynchronousCursor(dbCursor, cursorDescription, options);
};
var SynchronousCursor = function (dbCursor, cursorDescription, options) {
var self = this;
options = _.pick(options || {}, 'selfForIteration', 'useTransform');
self._dbCursor = dbCursor;
self._cursorDescription = cursorDescription;
// The "self" argument passed to forEach/map callbacks. If we're wrapped
// inside a user-visible Cursor, we want to provide the outer cursor!
self._selfForIteration = options.selfForIteration || self;
if (options.useTransform && cursorDescription.options.transform) {
self._transform = LocalCollection.wrapTransform(
cursorDescription.options.transform);
} else {
self._transform = null;
}
// Need to specify that the callback is the first argument to nextObject,
// since otherwise when we try to call it with no args the driver will
// interpret "undefined" first arg as an options hash and crash.
self._synchronousNextObject = Future.wrap(
dbCursor.nextObject.bind(dbCursor), 0);
self._synchronousCount = Future.wrap(dbCursor.count.bind(dbCursor));
self._visitedIds = new LocalCollection._IdMap;
};
_.extend(SynchronousCursor.prototype, {
_nextObject: function () {
var self = this;
while (true) {
var doc = self._synchronousNextObject().wait();
if (!doc) return null;
doc = replaceTypes(doc, replaceMongoAtomWithMeteor);
if (!self._cursorDescription.options.tailable && _.has(doc, '_id')) {
// Did Mongo give us duplicate documents in the same cursor? If so,
// ignore this one. (Do this before the transform, since transform might
// return some unrelated value.) We don't do this for tailable cursors,
// because we want to maintain O(1) memory usage. And if there isn't _id
// for some reason (maybe it's the oplog), then we don't do this either.
// (Be careful to do this for falsey but existing _id, though.)
if (self._visitedIds.has(doc._id)) continue;
self._visitedIds.set(doc._id, true);
}
if (self._transform)
doc = self._transform(doc);
return doc;
}
},
forEach: function (callback, thisArg) {
var self = this;
// Get back to the beginning.
self._rewind();
// We implement the loop ourself instead of using self._dbCursor.each,
// because "each" will call its callback outside of a fiber which makes it
// much more complex to make this function synchronous.
var index = 0;
while (true) {
var doc = self._nextObject();
if (!doc) return;
callback.call(thisArg, doc, index++, self._selfForIteration);
}
},
// XXX Allow overlapping callback executions if callback yields.
map: function (callback, thisArg) {
var self = this;
var res = [];
self.forEach(function (doc, index) {
res.push(callback.call(thisArg, doc, index, self._selfForIteration));
});
return res;
},
_rewind: function () {
var self = this;
// known to be synchronous
self._dbCursor.rewind();
self._visitedIds = new LocalCollection._IdMap;
},
// Mostly usable for tailable cursors.
close: function () {
var self = this;
self._dbCursor.close();
},
fetch: function () {
var self = this;
return self.map(_.identity);
},
count: function () {
var self = this;
return self._synchronousCount().wait();
},
// This method is NOT wrapped in Cursor.
getRawObjects: function (ordered) {
var self = this;
if (ordered) {
return self.fetch();
} else {
var results = new LocalCollection._IdMap;
self.forEach(function (doc) {
results.set(doc._id, doc);
});
return results;
}
}
});
MongoConnection.prototype.tail = function (cursorDescription, docCallback) {
var self = this;
if (!cursorDescription.options.tailable)
throw new Error("Can only tail a tailable cursor");
var cursor = self._createSynchronousCursor(cursorDescription);
var stopped = false;
var lastTS = undefined;
var loop = function () {
while (true) {
if (stopped)
return;
try {
var doc = cursor._nextObject();
} catch (err) {
// There's no good way to figure out if this was actually an error
// from Mongo. Ah well. But either way, we need to retry the cursor
// (unless the failure was because the observe got stopped).
doc = null;
}
// Since cursor._nextObject can yield, we need to check again to see if
// we've been stopped before calling the callback.
if (stopped)
return;
if (doc) {
// If a tailable cursor contains a "ts" field, use it to recreate the
// cursor on error. ("ts" is a standard that Mongo uses internally for
// the oplog, and there's a special flag that lets you do binary search
// on it instead of needing to use an index.)
lastTS = doc.ts;
docCallback(doc);
} else {
var newSelector = _.clone(cursorDescription.selector);
if (lastTS) {
newSelector.ts = {$gt: lastTS};
}
cursor = self._createSynchronousCursor(new CursorDescription(
cursorDescription.collectionName,
newSelector,
cursorDescription.options));
// Mongo failover takes many seconds. Retry in a bit. (Without this
// setTimeout, we peg the CPU at 100% and never notice the actual
// failover.
Meteor.setTimeout(loop, 100);
break;
}
}
};
Meteor.defer(loop);
return {
stop: function () {
stopped = true;
cursor.close();
}
};
};
MongoConnection.prototype._observeChanges = function (
cursorDescription, ordered, callbacks) {
var self = this;
if (cursorDescription.options.tailable) {
return self._observeChangesTailable(cursorDescription, ordered, callbacks);
}
// You may not filter out _id when observing changes, because the id is a core
// part of the observeChanges API.
if (cursorDescription.options.fields &&
(cursorDescription.options.fields._id === 0 ||
cursorDescription.options.fields._id === false)) {
throw Error("You may not observe a cursor with {fields: {_id: 0}}");
}
var observeKey = JSON.stringify(
_.extend({ordered: ordered}, cursorDescription));
var multiplexer, observeDriver;
var firstHandle = false;
// Find a matching ObserveMultiplexer, or create a new one. This next block is
// guaranteed to not yield (and it doesn't call anything that can observe a
// new query), so no other calls to this function can interleave with it.
Meteor._noYieldsAllowed(function () {
if (_.has(self._observeMultiplexers, observeKey)) {
multiplexer = self._observeMultiplexers[observeKey];
} else {
firstHandle = true;
// Create a new ObserveMultiplexer.
multiplexer = new ObserveMultiplexer({
ordered: ordered,
onStop: function () {
delete self._observeMultiplexers[observeKey];
observeDriver.stop();
}
});
self._observeMultiplexers[observeKey] = multiplexer;
}
});
var observeHandle = new ObserveHandle(multiplexer, callbacks);
if (firstHandle) {
var matcher, sorter;
var canUseOplog = _.all([
function () {
// At a bare minimum, using the oplog requires us to have an oplog, to
// want unordered callbacks, and to not want a callback on the polls
// that won't happen.
return self._oplogHandle && !ordered &&
!callbacks._testOnlyPollCallback;
}, function () {
// We need to be able to compile the selector. Fall back to polling for
// some newfangled $selector that minimongo doesn't support yet.
try {
matcher = new Minimongo.Matcher(cursorDescription.selector);
return true;
} catch (e) {
// XXX make all compilation errors MinimongoError or something
// so that this doesn't ignore unrelated exceptions
return false;
}
}, function () {
// ... and the selector itself needs to support oplog.
return OplogObserveDriver.cursorSupported(cursorDescription, matcher);
}, function () {
// And we need to be able to compile the sort, if any. eg, can't be
// {$natural: 1}.
if (!cursorDescription.options.sort)
return true;
try {
sorter = new Minimongo.Sorter(cursorDescription.options.sort,
{ matcher: matcher });
return true;
} catch (e) {
// XXX make all compilation errors MinimongoError or something
// so that this doesn't ignore unrelated exceptions
return false;
}
}], function (f) { return f(); }); // invoke each function
var driverClass = canUseOplog ? OplogObserveDriver : PollingObserveDriver;
observeDriver = new driverClass({
cursorDescription: cursorDescription,
mongoHandle: self,
multiplexer: multiplexer,
ordered: ordered,
matcher: matcher, // ignored by polling
sorter: sorter, // ignored by polling
_testOnlyPollCallback: callbacks._testOnlyPollCallback
});
// This field is only set for use in tests.
multiplexer._observeDriver = observeDriver;
}
// Blocks until the initial adds have been sent.
multiplexer.addHandleAndSendInitialAdds(observeHandle);
return observeHandle;
};
// Listen for the invalidation messages that will trigger us to poll the
// database for changes. If this selector specifies specific IDs, specify them
// here, so that updates to different specific IDs don't cause us to poll.
// listenCallback is the same kind of (notification, complete) callback passed
// to InvalidationCrossbar.listen.
listenAll = function (cursorDescription, listenCallback) {
var listeners = [];
forEachTrigger(cursorDescription, function (trigger) {
listeners.push(DDPServer._InvalidationCrossbar.listen(
trigger, listenCallback));
});
return {
stop: function () {
_.each(listeners, function (listener) {
listener.stop();
});
}
};
};
forEachTrigger = function (cursorDescription, triggerCallback) {
var key = {collection: cursorDescription.collectionName};
var specificIds = LocalCollection._idsMatchedBySelector(
cursorDescription.selector);
if (specificIds) {
_.each(specificIds, function (id) {
triggerCallback(_.extend({id: id}, key));
});
triggerCallback(_.extend({dropCollection: true, id: null}, key));
} else {
triggerCallback(key);
}
};
// observeChanges for tailable cursors on capped collections.
//
// Some differences from normal cursors:
// - Will never produce anything other than 'added' or 'addedBefore'. If you
// do update a document that has already been produced, this will not notice
// it.
// - If you disconnect and reconnect from Mongo, it will essentially restart
// the query, which will lead to duplicate results. This is pretty bad,
// but if you include a field called 'ts' which is inserted as
// new MongoInternals.MongoTimestamp(0, 0) (which is initialized to the
// current Mongo-style timestamp), we'll be able to find the place to
// restart properly. (This field is specifically understood by Mongo with an
// optimization which allows it to find the right place to start without
// an index on ts. It's how the oplog works.)
// - No callbacks are triggered synchronously with the call (there's no
// differentiation between "initial data" and "later changes"; everything
// that matches the query gets sent asynchronously).
// - De-duplication is not implemented.
// - Does not yet interact with the write fence. Probably, this should work by
// ignoring removes (which don't work on capped collections) and updates
// (which don't affect tailable cursors), and just keeping track of the ID
// of the inserted object, and closing the write fence once you get to that
// ID (or timestamp?). This doesn't work well if the document doesn't match
// the query, though. On the other hand, the write fence can close
// immediately if it does not match the query. So if we trust minimongo
// enough to accurately evaluate the query against the write fence, we
// should be able to do this... Of course, minimongo doesn't even support
// Mongo Timestamps yet.
MongoConnection.prototype._observeChangesTailable = function (
cursorDescription, ordered, callbacks) {
var self = this;
// Tailable cursors only ever call added/addedBefore callbacks, so it's an
// error if you didn't provide them.
if ((ordered && !callbacks.addedBefore) ||
(!ordered && !callbacks.added)) {
throw new Error("Can't observe an " + (ordered ? "ordered" : "unordered")
+ " tailable cursor without a "
+ (ordered ? "addedBefore" : "added") + " callback");
}
return self.tail(cursorDescription, function (doc) {
var id = doc._id;
delete doc._id;
// The ts is an implementation detail. Hide it.
delete doc.ts;
if (ordered) {
callbacks.addedBefore(id, doc, null);
} else {
callbacks.added(id, doc);
}
});
};
// XXX We probably need to find a better way to expose this. Right now
// it's only used by tests, but in fact you need it in normal
// operation to interact with capped collections.
MongoInternals.MongoTimestamp = MongoDB.Timestamp;
MongoInternals.Connection = MongoConnection;