I'm working on a filtered live search module with Meteor.js.
Usecase & problem:
A user wants to do a search through all the users to find friends. But I cannot afford for each user to ask the complete users collection. The user filter the search using checkboxes. I'd like to subscribe to the matched users. What is the best way to do it ?
I guess it would be better to create the query client-side, then send it the the method to get back the desired set of users. But, I wonder : when the filtering criteria changes, does the new subscription erase all of the old one ? Because, if I do a first search which return me [usr1, usr3, usr5], and after that a search that return me [usr2, usr4], the best would be to keep the first set and simply add the new one to it on the client-side suscribed collection.
And, in addition, if then I do a third research wich should return me [usr1, usr3, usr2, usr4], the autorunned subscription would not send me anything as I already have the whole result set in my collection.
The goal is to spare processing and data transfer from the server.
I have some ideas, but I haven't coded enough of it yet to share it in a easily comprehensive way.
How would you advice me to do to be the more relevant possible in term of time and performance saving ?
Thanks you all.
David
It depends on your application, but you'll probably send a non-empty string to a publisher which uses that string to search the users collection for matching names. For example:
Meteor.publish('usersByName', function(search) {
check(search, String);
// make sure the user is logged in and that search is sufficiently long
if (!(this.userId && search.length > 2))
return [];
// search by case insensitive regular expression
var selector = {username: new RegExp(search, 'i')};
// only publish the necessary fields
var options = {fields: {username: 1}};
return Meteor.users.find(selector, options);
});
Also see common mistakes for why we limit the fields.
performance
Meteor is clever enough to keep track of the current document set that each client has for each publisher. When the publisher reruns, it knows to only send the difference between the sets. So the situation you described above is already taken care of for you.
If you were subscribed for users: 1,2,3
Then you restarted the subscription for users 2,3,4
The server would send a removed message for 1 and an added message for 4.
Note this will not happen if you stopped the subscription prior to rerunning it.
To my knowledge, there isn't a way to avoid removed messages when modifying the parameters for a single subscription. I can think of two possible (but tricky) alternatives:
Accumulate the intersection of all prior search queries and use that when subscribing. For example, if a user searched for {height: 5} and then searched for {eyes: 'blue'} you could subscribe with {height: 5, eyes: 'blue'}. This may be hard to implement on the client, but it should accomplish what you want with the minimum network traffic.
Accumulate active subscriptions. Rather than modifying the existing subscription each time the user modifies the search, start a new subscription for the new set of documents, and push the subscription handle to an array. When the template is destroyed, you'll need to iterate through all of the handles and call stop() on them. This should work, but it will consume more resources (both network and server memory + CPU).
Before attempting either of these solutions, I'd recommend benchmarking the worst case scenario without using them. My main concern is that without fairly tight controls, you could end up publishing the entire users collection after successive searches.
If you want to go easy on your server, you'll want to send as little data to the client as possible. That means every document you send to the client that is NOT a friend is waste. So let's eliminate all that waste.
Collect your filters (eg filters = {sex: 'Male', state: 'Oregon'}). Then call a method to search based on your filter (eg Users.find(filters). Additionally, you can run your own proprietary ranking algorithm to determine the % chance that a person is a friend. Maybe base it off of distance from ip address (or from phone GPS history), mutual friends, etc. This will pay dividends in efficiency in a bit. Index things like GPS coords or other highly unique attributes, maybe try out composite indexes. But remember more indexes means slower writes.
Now you've got a cursor with all possible friends, ranked from most likely to least likely.
Next, change your subscription to match those friends, but put a limit:20 on there. Also, only send over the fields you need. That way, if a user wants to skip this step, you only wasted sending 20 partial docs over the wire. Then, have an infinite scroll or 'load more' button the user can click. When they load more, it's an additive subscription, so it's not resending duplicate info. Discover Meteor describes this pattern in great detail, so I won't.
After a few clicks/scrolls, the user won't find any more friends (because you were smart & sorted them) so they will stop trying & move on to the next step. If you returned 200 possible friends & they stop trying after 60, you just saved 140 docs from going through the pipeline. There's your efficiency.
Related
Context
I have a database with a collection of documents using this schema (shortened schema because some data is irrelevant to my problem):
{
title: string;
order: number;
...
...
...
modificationsHistory: HistoryEntry[];
items: ListRow[];
finalItems: ListRow[];
...
...
...
}
These documents can easily reach 100 or 200 kB, depending on the amount of items and finalItems that they hold. It's also very important that they are updated as fast as possible, with the smallest bandwidth usage possible.
This is inside a web application context, using Angular 9 and #angular/fire 6.0.0.
Problems
When the end user edits one item inside the object's item array, like editing just a property, reflecting that inside the database requires me to send the entire object, because firestore's update method doesn't support array indexes inside the field path, the only operations that can be done on arrays are adding or deleting an element as described inside documentation.
However, updating an element of the items array by sending the entire document creates poor performances for anyone without a good connection, which is the case for a lot of my users.
Second issue is that having everything in realtime inside one document makes collaboration hard in my case, because some of these elements can be edited by multiple users at the same time, which creates two issues:
Some write operations may fail due to too much contention on the document if two updates are made in the same second.
The updates are not atomic as we're sending the entire document at once, as it doesn't use transactions to avoid using bandwidth even more.
Solutions I already tried
Subcollections
Description
This was a very simple solution: create a subcollection for items, finalItems and modificationsHistory arrays, making them easy to edit as they now have their own ID so it's easy to reach them to update them.
Why it didn't work
Having a list with 10 finalItems, 30 items and 50 entries inside modificationsHistory means that I need to have a total of 4 listeners opened for one element to be listened entirely. Considering the fact that a user can have many of these elements opened at once, having several dozens of documents being listened creates an equally bad performance situation, probably even worse in a full user case.
It also means that if I want to update a big element with 100 items and I want to update half of them, it'll cost me one write operation per item, not to mention the amount of read operations needed to check permissions, etc, probably 3 per write so 150 read + 50 write just to update 50 items in an array.
Cloud Function to update the document
const {
applyPatch
} = require('fast-json-patch');
function applyOffsets(data, entries) {
entries.forEach(customEntry => {
const explodedPath = customEntry.path.split('/');
explodedPath.shift();
let pointer = data;
for (let fragment of explodedPath.slice(0, -1)) {
pointer = pointer[fragment];
}
pointer[explodedPath[explodedPath.length - 1]] += customEntry.offset;
});
return data;
}
exports.updateList = functions.runWith(runtimeOpts).https.onCall((data, context) => {
const listRef = firestore.collection('lists').doc(data.uid);
return firestore.runTransaction(transaction => {
return transaction.get(listRef).then(listDoc => {
const list = listDoc.data();
try {
const [standard, custom] = JSON.parse(data.diff).reduce((acc, entry) => {
if (entry.custom) {
acc[1].push(entry);
} else {
acc[0].push(entry);
}
return acc;
}, [
[],
[]
]);
applyPatch(list, standard);
applyOffsets(list, custom);
transaction.set(listRef, list);
} catch (e) {
console.log(data.diff);
}
});
});
});
Description
Using a diff library, I was making a diff between previous document and the new updated one, and sending this diff to a GCF that was operating the update using the transaction API.
Benefits of this approach being that since transaction happens inside GCF, it's super fast and doesn't consume too much bandwidth, plus the update only requires a diff to be sent, not the entire document anymore.
Why it didn't work
In reality, the cloud function was really slow and some updates were taking over 2 seconds to be made, they could also fail due to contention, without firestore connector knowing it, so no possibility to ensure data integrity in this case.
I will be edited accordingly to add more solutions if I find other stuff to try
Question
I feel like I'm missing something, like if firestore had something I just didn't know at all that could solve my use case, but I can't figure out what it is, maybe my previously tested solutions were badly implemented or I missed something important. What did I miss? Is it even possible to achieve what I want to do? I am open to data remodeling, query changes, anything, as it's mostly for learning purpose.
You should be able to reduce the bandwidth required to update your documents by using Maps instead of Arrays to store your data. This would allow you to send only the item that is being updated using its key.
I don't know how involved this would be for you to change, but it sounds like less work than the other options.
You said that it's not impossible for your documents to reach 200kb individually. It would be good to keep in mind that Firestore limits document size to 1mb. If you plan on supporting documents beyond that, you will need to find a way to fragment the data.
Regarding your contention issues... You might consider a system that "locks" the document and prevents it from receiving updates while another user is attempting to save. You could use a simple message system built with websockets or Firebase FCM to do this. A client would subscribe to the document's channel, and publish when they are attempting an update. Other clients would then receive a notice that the document is being updated and have to wait before they can save their own changes.
Also, I don't know what the contents of modificationsHistory look like, but that sounds to me like the type of data that you might keep in a subcollection instead.
Of the solutions you tried, the subcollection seems like the most scalable to me. You could look into the possibility of not using onSnapshot listeners and instead create your own event system to notify clients of changes. I suppose it could work similar to the "locking" system I mentioned above. A client sends an event when it updates an item belonging to a document. Other clients subscribed to that document's channel will know to check the database for the newest version.
Your diff-approach appeared mostly sensible, details aside.
You should store items inline, but defer modificationsHistory into a sub collection. For the entire root document, record which elements of modificationsHistory have been merged yet (by timestamp should suffice), and all elements not merged yet, you have to re-apply individually on each client, querying with aforementioned timestamp.
Each entry in modificationsHistory should not describe a single diff, but whenever possible a set of diffs.
Apply changes from modificationsHistory collections onto items in batch, deferred via GCF. You may defer this arbitrarily far, and you may want to exclude modifications performed only in the last few seconds, to account for not established consistency in Firestore. There is no risk of contention, that way.
Cleanup from the modificationsHistory collection has to be deferred even further, until you can be sure that no client has still access to an older revision of the root document. Especially if you consider that the client is not strictly required to update the root document when the listener is triggered.
You may need to reconstruct the patch stack on the client side if modificationsHistory changes in unexpected ways due to eventual consistency constraints. E.g. if you have a total order in the set of patches, you need to re-apply the patch stack from base image if the collection unexpectedly suddenly contains "older" patches unknown to the client before.
All in all, you should be able avoid frequent updates all together, and limit this solely to inserts into to modificationsHistory sub-collection. With bandwidth requirements not exceeding the cost of fetching the entire document once, plus streaming the collection of not-yet-applied patches. No contention expected.
You can tweak for how long clients may ignore hard updates to the root document, and how many changes they may batch client-side before submitting a new diff. Latter is also a tradeof with regard to how many documents another client has to fetch initially, with regard to max-documents-per-query limits.
If you require other information which are likely to suffer from contention, like list of users currently having a specific document open, that should go into sub-collections as well.
Should the latency for seeing changes by other users eventually turn out to be unacceptable, you may opt for an additional, real-time capable data channel for distribution of patches on a specific document. ActiveMQ or some other message broker operated on dedicated resources, running independently from FireStore.
The scenario is: some user sending messages to some group of people.
I was thinking to create one ROW for that specific conversation into one CLASS. WHERE in that ROW contains information such "sender name", "receiver " and addition I have column (PFRelation) which connects this specific row to another class where all messages from the user to the receiver would be saved(vice-versa) into.
So this action will happen every time the user starts a new conversation.
The benefit from this prospective :
Privacy because the only convo that is being saved are only from the user and the receiver group.
Downside of this prospective:
We all know that parse only provide 30reqs/s for free which means that 1 min =1800 reqs. So every time I create a new class to keep track of the convo. Am I using a lot of requests ?
I am looking suggestions and thoughts for the ideal way before I implement this messenger library.
It sounds like you have come up with something that is similar to what I have used before to implement messaging in an app with Parse as a backend. It's also important to think about how your UI will be querying for data. In general, it's most important to ensure that it is very easy and fast to read data. For most social apps, the following quote from Facebook's engineering team on Haystack is particularly relevant.
Haystack is an object store that we designed for sharing photos on
Facebook where data is written once, read often, never modified, and
rarely deleted.
The crucial piece of information here is written once, read often, never modified, and rarely deleted. No matter what approach you decide to take, keep that in mind while engineering your solution. The approach that I have used before to implement a messaging system using Parse is described below.
Overview
Each row (object) of the Message class corresponds with an individual text, picture, or video message that was posted. Each Message belongs to a Group. A Group can be as small as 2 User (private conversation) or grow as large as you like.
The RecentMessage class is the solution I came up with to deal with quickly and easily populating the UI. Each RecentMessage object corresponds to each Group that a given User may belong. Each User in a Group will have their own RecentMessage object which is kept up to date using beforeSave/afterSave cloud code triggers. Whenever a new Message is created, in the afterSave trigger we want to update all of the RecentMessage objects that belong to the Group.
You will most likely have a table in your app which displays all of the conversations that the user is part of. This is easily achieved by querying for all of that user's RecentMessage objects which already contains all of the Group information needed to load the rest of the messages when selected and also contains the most recent message's data (hence the name) to display in the table. Alternatively, RecentMessage could contain a pointer to the most recent Message, however I decided that copying the data was a beneficial tradeoff since it streamlines future queries.
Message
group (pointer to group which message is part of)
user (pointer to user who created it)
text (string)
picture (optional file)
video (optional file)
RecentMessage
group (group pointer)
user (user pointer)
lastMessage (string containing the text of most recent Message)
lastUser (pointer to the User who posted the most recent Message)
Group
members (array of user pointers)
name or whatever other info you want
Security/Privacy
Security and privacy are imperative when creating messaging functionality in your app. Make sure to read through the Parse Engineering security blog posts, and take your time to let it all soak in: Part I, Part II, Part III, Part IV, Part V.
Most important in our case is Part III which describes ACLs, or Access Control Lists. Group objects will have an ACL which corresponds to all of its member User. RecentMessage objects will have a restricted read/write ACL to its owner User. Message objects will inherit the ACL of the Group to which they belong, allowing all of the Group members to read. I recommend disabling the write ACL in the afterSave trigger so messages cannot be modified.
General Remarks
With regards to Parse and the request limit, you need to accept that fact that you will very quickly surpass the 30 req/s free tier. As a general rule of thumb, it's much better to focus on building the best possible user experience than to focus too much on scalability. By and large, issues of scalability rarely come into play because most apps fail. Not saying that to be discouraging — just something to keep in mind to prevent you from falling into the trap of over-engineering at the cost of time :)
we have a button in a web game for the users to collect reward. That should only be clicked once, and upon receiving the request, we'll mark it collected in DB.
we've already blocked the buttons in the client from repeated clicking. But that won't help if people resend the package multiple times to our server in short period of time.
what I want is a method to block this from server side.
we're using Playframework 2 (2.0.3-RC2) for server side and so far it's stateless, I'm tempted to use a Set to guard like this:
if processingSet has userId then BadRequest
else put userId in processingSet and handle request
after that remove userId from that Set
but then I'd have to face problem like Updating Scala collections thread-safely and still fail to block the user once we have more than one server behind load balancing.
one possibility I'm thinking about is to have a table in DB in place of the processingSet above, but that would incur 1+ DB operation per request, are there any better solution~?
thanks~
Additional DB operation is relatively 'cheap' solution in that case. You should use it if you'e planning to save the buttons state permanently.
If the button is disabled only for some period of time (for an example until the game is over) you can also consider using the cache API however keep in mind that's not dedicated for solutions which should be stored for long time (it should not be considered as DB alternative).
Given that you're using Mongo and so don't have transactions spanning separate collections, I think you can probably implement this guard using an atomic operation - namely "Update if current", which is effectively CompareAndSwap.
Assuming you've got a collection like "rewards" which has a "collected" attribute, you can update the collected flag to true only if it is currently false and if that operation doesn't fail you can proceed to apply the reward knowing that for any other requests the same operation will fail.
I'm implementing a RESTful API which exposes Orders as a resource and supports pagination through the resultset:
GET /orders?start=1&end=30
where the orders to paginate are sorted by ordered_at timestamp, descending. This is basically approach #1 from the SO question Pagination in a REST web application.
If the user requests the second page of orders (GET /orders?start=31&end=60), the server simply re-queries the orders table, sorts by ordered_at DESC again and returns the records in positions 31 to 60.
The problem I have is: what happens if the resultset changes (e.g. a new order is added) while the user is viewing the records? In the case of a new order being added, the user would see the old order #30 in first position on the second page of results (because the same order is now #31). Worse, in the case of a deletion, the user sees the old order #32 in first position on the second page (#31) and wouldn't see the old order #31 (now #30) at all.
I can't see a solution to this without somehow making the RESTful server stateful (urg) or building some pagination intelligence into each client... What are some established techniques for dealing with this?
For completeness: my back-end is implemented in Scala/Spray/Squeryl/Postgres; I'm building two front-end clients, one in backbone.js and the other in Python Django.
The way I'd do it, is to make the indices from old to new. So they never change. And then when querying without any start parameter, return the newest page. Also the response should contain an index indicating what elements are contained, so you can calculate the indices you need to request for the next older page. While this is not exactly what you want, it seems like the easiest and cleanest solution to me.
Initial request: GET /orders?count=30 returns:
{
"start"=1039;
"count"=30;
...//data
}
From this the consumer calculates that he wants to request:
Next requests: GET /orders?start=1009&count=30 which then returns:
{
"start"=1009;
"count"=30;
...//data
}
Instead of raw indices you could also return a link to the next page:
{
"next"="/orders?start=1009&count=30";
}
This approach breaks if items get inserted or deleted in the middle. In that case you should use some auto incrementing persistent value instead of an index.
The sad truth is that all the sites I see have pagination "broken" in that sense, so there must not be an easy way to achieve that.
A quick workaround could be reversing the ordering, so the position of the items is absolute and unchanging with new additions. From your front page you can give the latest indices to ensure consistent navigation from up there.
Pros: same url gives the same results
Cons: there's no evident way to get the latest elements... Maybe you could use negative indices and redirect the result page to the absolute indices.
With a RESTFUL API, Application state should be in the client. Here the application state should some sort of time stamp or version number telling when you started looking at the data. On the server side, you will need some form of audit trail, which is properly server data, as it does not depend on whether there have been clients and what they have done. At the very least, it should know when the data last changed. No contradiction with REST here.
You could add a version parameter to your get. When the client first requires a page, it normally does not send a version. The server replies contains one. For instance, if there are links in the reply to next/other pages, those links contains &version=... The client should send the version when requiring another page.
When the server recieves some request with a version, it should at least know whether the data have changed since the client started looking and, dependending of what sort of audit trail you have, how they have changed. If they have not, it answer normally, transmitting the same version number. If they have, it may at least tell the client. And depending how much it knows on how the data have changed, it may taylor the reply accordingly.
Just as an example, suppose you get a request with start, end, version, and that you know that since version was up to date, 3 rows coming before start have been deleted. You might send a redirect with start-3, end-3, new version.
WebSockets can do this. You can use something like pusher.com to catch realtime changes to your database and pass the changes to the client. You can then bind different pusher events to work with models and collections.
Just Going to throw it out there. Please feel free to tell me if it's completely wrong and why so.
This approach is trying to use a left_off variable to sort through without using offsets.
Consider you need to make your result Ordered by timestamp order_at DESC.
So when I ask for first result set
it's
SELECT * FROM Orders ORDER BY order_at DESC LIMIT 25;
right?
This is the case when you ask for the first page (in terms of URL probably the request that doesn't have any
yoursomething.com/orders?limit=25&left_off=$timestamp
Then When receiving your data set. just grab the timestamp of last viewed item. 2015-12-21 13:00:49
Now to Request next 25 items go to: yoursomething.com/orders?limit=25&left_off=2015-12-21 13:00:49 (to lastly viewed timestamp)
In Sql you would just make the same query and say where timestamp is equal or less than $left_off
SELECT * FROM (SELECT * FROM Orders ORDER BY order_at DESC) as a
WHERE a.order_at < '2015-12-21 13:00:49' LIMIT 25;
You should get a next 25 items from the last seen item.
For those who sees this answer. Please comment if this approach is relevant or even possible in the first place. Thank you.
The repository in the CommonDomain only exposes the "GetById()". So what to do if my Handler needs a list of Customers for example?
On face value of your question, if you needed to perform operations on multiple aggregates, you would just provide the ID's of each aggregate in your command (which the client would obtain from the query side), then you get each aggregate from the repository.
However, looking at one of your comments in response to another answer I see what you are actually referring to is set based validation.
This very question has raised quite a lot debate about how to do this, and Greg Young has written an blog post on it.
The classic question is 'how do I check that the username hasn't already been used when processing my 'CreateUserCommand'. I believe the suggested approach is to assume that the client has already done this check by asking the query side before issuing the command. When the user aggregate is created the UserCreatedEvent will be raised and handled by the query side. Here, the insert query will fail (either because of a check or unique constraint in the DB), and a compensating command would be issued, which would delete the newly created aggregate and perhaps email the user telling them the username is already taken.
The main point is, you assume that the client has done the check. I know this is approach is difficult to grasp at first - but it's the nature of eventual consistency.
Also you might want to read this other question which is similar, and contains some wise words from Udi Dahan.
In the classic event sourcing model, queries like get all customers would be carried out by a separate query handler which listens to all events in the domain and builds a query model to satisfy the relevant questions.
If you need to query customers by last name, for instance, you could listen to all customer created and customer name change events and just update one table of last-name to customer-id pairs. You could hold other information relevant to the UI that is showing the data, or you could simply hold IDs and go to the repository for the relevant customers in order to work further with them.
You don't need list of customers in your handler. Each aggregate MUST be processed in its own transaction. If you want to show this list to user - just build appropriate view.
Your command needs to contain the id of the aggregate root it should operate on.
This id will be looked up by the client sending the command using a view in your readmodel. This view will be populated with data from the events that your AR emits.