I have a data model which is sort of like this simplified drawing:
alt text http://dl.dropbox.com/u/545670/thedatamodel.png
It's a little weird, but the idea is that the app manages multiple accounts/identities a person may have into a single messaging system. Each account is associated with one user on the system, and each message could potentially be seen/sent-to multiple accounts (but they have a globally unique ID hence the messageID property which is used on import to fetch message objects that may have already been downloaded and imported by a prior session).
The app is used from a per-account point of view - what I mean is that you choose which account you want to use, then you see the messages and stuff from that account's point of view in your window. So I have the messages attached to the account so that I can easily get the messages that should be shown using a fetch like this:
fetch.fetchPredicate = [NSPredicate predicateWithFormat:#"%# IN accounts", theAccount];
fetch.sortDescriptors = [NSArray arrayWithObject:[[NSSortDescriptor alloc] initWithKey:#"date" ascending:NO]];
fetch.fetchLimit = 20;
This seems like the right way to set this up in that the messages are shared between accounts and if a message is marked as read by one, I want it seen as being read by the other and so on.
Anyway, after all this setup, the big problem is that memory usage seems to get a little crazy. When I setup a test case where it's importing hundreds of messages into the system, and periodically re-fetching (using the fetch mentioned above) and showing them in a list (only the last 20 are referenced by the list), memory just gets crazy. 60MB.. 70MB... 100MB.. etc.
I tracked it down to the many-to-many relation between Account and Message. Even with garbage collection on, the managed objects are still being referenced strongly by the account's messages relationship property. I know this because I put a log in the finalize of my Message instance and never see it - but if I periodically reset the context or do refreshObject:mergeChanges: on the account object, I see the finalize messages and memory usage stays pretty consistent (although still growing somewhat, but considering I'm importing stuff, that's to be expected). The problem is that I can't really reset the context or the account object all the time because that really messes up observers that are observing other attributes of the account object!
I might just be modeling this wrong or thinking about it wrong, but I keep reading over and over that it's important to think of Core Data as an object graph and not a database. I think I've done that here, but it seems to be causing trouble. What should I do?
Use the Object Graph instrument. It'll tell you all of the ownerships keeping an object alive.
Have you read the section of the docs on this topic?
Related
So, I'm working on a CQRS/ES project in which we are having some doubts about how to handle trivial problems that would be easy to handle in other architectures
My scenario is the following:
I have a customer CRUD REST API and each customer has unique document(number), so when I'm registering a new customer I have to verify if there is another customer with that document to avoid duplicity, but when it comes to a CQRS/ES architecture where we have eventual consistency, I found out that this kind of validations can be very hard to address.
It is important to notice that my problem is not across microservices, but between the command application and the query application of the same microservice.
Also we are using eventstore.
My current solution:
So what I do today is, in my command application, before saving the CustomerCreated event, I ask the query application (using PostgreSQL) if there is a customer with that document, and if not, I allow the event to go on. But that doesn't guarantee 100%, right? Because my query can be desynchronized, so I cannot trust it 100%. That's when my second validation kicks in, when my query application is processing the events and saving them to my PostgreSQL, I check again if there is a customer with that document and if there is, I reject that event and emit a compensating event to undo/cancel/inactivate the customer with the duplicated document, therefore finishing that customer stream on eventstore.
Altough this works, there are 2 things that bother me here, the first thing is my command application relying on the query application, so if my query application is down, my command is affected (today I just return false on my validation if query is down but still...) and second thing is, should a query/read model really be able to emit events? And if so, what is the correct way of doing it? Should the command have some kind of API for that? Or should the query emit the event directly to eventstore using some common shared library? And if I have more than one view/read? Which one should I choose to handle this?
Really hope someone could shine a light into these questions and help me this these matters.
For reference, you may want to be reviewing what Greg Young has written about Set Validation.
I ask the query application (using PostgreSQL) if there is a customer with that document, and if not, I allow the event to go on. But that doesn't guarantee 100%, right?
That's exactly right - your read model is stale copy, and may not have all of the information collected by the write model.
That's when my second validation kicks in, when my query application is processing the events and saving them to my PostgreSQL, I check again if there is a customer with that document and if there is, I reject that event and emit a compensating event to undo/cancel/inactivate the customer with the duplicated document, therefore finishing that customer stream on eventstore.
This spelling doesn't quite match the usual designs. The more common implementation is that, if we detect a problem when reading data, we send a command message to the write model, telling it to straighten things out.
This is commonly referred to as a process manager, but you can think of it as the automation of a human supervisor of the system. Conceptually, a process manager is an event sourced collection of messages to be sent to the command model.
You might also want to consider whether you are modeling your domain correctly. If documents are supposed to be unique, then maybe the command model should be using the document number as a key in the book of record, rather than using the customer. Or perhaps the document id should be a function of the customer data, rather than being an arbitrary input.
as far as I know, eventstore doesn't have transactions across different streams
Right - one of the things you really need to be thinking about in general is where your stream boundaries lie. If set validation has significant business value, then you really need to be thinking about getting the entire set into a single stream (or by finding a way to constrain uniqueness without using a set).
How should I send a command message to the write model? via API? via a message broker like Kafka?
That's plumbing; it doesn't really matter how you do it, so long as you are sure that the command runs within its own transaction/unit of work.
So what I do today is, in my command application, before saving the CustomerCreated event, I ask the query application (using PostgreSQL) if there is a customer with that document, and if not, I allow the event to go on. But that doesn't guarantee 100%, right? Because my query can be desynchronized, so I cannot trust it 100%.
No, you cannot safely rely on the query side, which is eventually consistent, to prevent the system to step into an invalid state.
You have two options:
You permit the system to enter in a temporary, pending state and then, eventually, you will bring it into a valid permanent state; for this you could allow the command to pass, yield CustomerRegistered event and using a Saga/Process manager you verify against a uniquely-indexed-by-document-collection and issue a compensating command (not event!), i.e. UnregisterCustomer.
Instead of sending a command, you create&start a Saga/Process that preallocates the document in a uniquely-indexed-by-document-collection and if successfully then send the RegisterCustomer command. You can model the Saga as an entity.
So, in both solution you use a Saga/Process manager. In order for the system to be resilient you should make sure that RegisterCustomer command is idempotent (so you can resend it if the Saga fails/is restarted)
You've butted up against a fairly common problem. I think the other answer by VoicOfUnreason is worth reading. I just wanted to make you aware of a few more options.
A simple approach I have used in the past is to create a lookup table. Your command tries to register the key in a unique constraint table. If it can reserve the key the command can go ahead.
Depending on the nature of the data and the domain you could let this 'problem' occur and raise additional events to mark it. If it is something that's important to the business/the way the application works then you can deal with it either manually or at the time via compensating commands. if the latter then it would make sense to use a process manager.
In some (rare) cases where speed/capacity is less of an issue then you could consider old-fashioned locking and transactions. Admittedly these are much better suited to CRUD style implementations but they can be used in CQRS/ES.
I have more detail on this in my blog post: How to Handle Set Based Consistency Validation in CQRS
I hope you find it helpful.
I'm developing small CQRS+ES framework and develop applications with it. In my system, I should log some action of the client and use it for analytics, statistics and maybe in the future do something in domain with it. For example, client (on web) download some resource(s) and I need save date, time, type (download, partial,...), from region or country (maybe IP), etc. after that in some view client can see count of download or some complex report. I'm not sure how to implement this feather.
First solution creates analytic context and some aggregate, in each client action send some command like IncreaseDownloadCounter(resourced) them handle the command and raise domain event's and updating view, but in this scenario first download occurred and after that, I send command so this is not really command and on other side version conflict increase.
The second solution is raising event, from client side and update the view model base on it, but in this type of handling my event not store in event store because it's not raise by command and never change any domain context. If is store it in event store, no aggregate to handle it after fetch for some other use.
Third solution is raising event, from client side and I store it on other database may be for each type of event have special table, but in this manner of event handle I have multiple event storage with different schema and difficult on recreating view models and trace events for recreating contexts states so in future if I add some domain for use this type of event's it's difficult to use events.
What is the best approach and solution for this scenario?
First solution creates analytic context and some aggregate
Unquestionably the wrong answer; the event has already happened, so it is too late for the domain model to complain.
What you have is a stream of events. Putting them in the same event store that you use for your aggregate event streams is fine. Putting them in a separate store is also fine. So you are going to need some other constraint to make a good choice.
Typically, reads vastly outnumber writes, so one concern might be that these events are going to saturate the domain store. That might push you towards storing these events separately from your data model (prior art: we typically keep the business data in our persistent book of record, but the sequence of http requests received by the server is typically written instead to a log...)
If you are supporting an operational view, push on the requirement that the state be recovered after a restart. You might be able to get by with building your view off of an in memory model of the event counts, and use something more practical for the representations of the events.
Thanks for your complete answer, so I should create something like the ES schema without some field (aggregate name or type, version, etc.) and collect client event in that repository, some offline process read and update read model or create command to do something on domain space.
Something like that, yes. If the view for the client doesn't actually require any validation by your model at all, then building the read model from the externally provided events is fine.
Are you recommending save some claim or authorization token of the user and sender app for validation in another process?
Maybe, maybe not. The token describes the authority of the event; our own event handler is the authority for the command(s) that is/are derived from the events. It's an interesting question that probably requires more context -- I'd suggest you open a new question on that point.
I am designing some events that will be raised when actions are performed or data changes in a system. These events will likely be consumed by many different services and will be serialized as XML, although more broadly my question also applies to the design of more modern funky things like Webhooks.
I'm specifically thinking about how to describe changes with an event and am having difficulty choosing between different implementations. Let me illustrate my quandry.
Imagine a customer is created, and a simple event is raised.
<CustomerCreated>
<CustomerId>1234</CustomerId>
<FullName>Bob</FullName>
<AccountLevel>Silver</AccountLevel>
</CustomerCreated>
Now let's say Bob spends lots of money and becomes a gold customer, or indeed any other property changes (e.g.: he now prefers to be known as Robert). I could raise an event like this.
<CustomerModified>
<CustomerId>1234</CustomerId>
<FullName>Bob</FullName>
<AccountLevel>Gold</AccountLevel>
</CustomerModified>
This is nice because the schema of the Created and Modified events are the same and any subscriber receives the complete current state of the entity. However it is difficult for any receiver to determine which properties have changed without tracking state themselves.
I then thought about an event like this.
<CustomerModified>
<CustomerId>1234</CustomerId>
<AccountLevel>Gold</AccountLevel>
</CustomerModified>
This is more compact and only contains the properties that have changed, but comes with the downside that the receiver must apply the changes and reassemble the current state of the entity if they need it. Also, the schemas of the Created and Modified events must be different now; CustomerId is required but all other properties are optional.
Then I came up with this.
<CustomerModified>
<CustomerId>1234</CustomerId>
<Before>
<FullName>Bob</FullName>
<AccountLevel>Silver</AccountLevel>
</Before>
<After>
<FullName>Bob</FullName>
<AccountLevel>Gold</AccountLevel>
</After>
</CustomerModified>
This covers all bases as it contains the full current state, plus a receiver can figure out what has changed. The Before and After elements have the exact same schema type as the Created event. However, it is incredibly verbose.
I've struggled to find any good examples of events; are there any other patterns I should consider?
You tagged the question as "Event Sourcing", but your question seems to be more about Event-Driven SOA.
I agree with #Matt's answer--"CustomerModified" is not granular enough to capture intent if there are multiple business reasons why a Customer would change.
However, I would back up even further and ask you to consider why you are storing Customer information in a local service, when it seems that you (presumably) already have a source of truth for customer. The starting point for consuming Customer information should be getting it from the source when it's needed. Storing a copy of information that can be queried reliably from the source may very well be an unnecessary optimization (and complication).
Even if you do need to store Customer data locally (and there are certainly valid reasons for need to do so), consider passing only the data necessary to construct a query of the source of truth (the service emitting the event):
<SomeInterestingCustomerStateChange>
<CustomerId>1234</CustomerId>
</SomeInterestingCustomerStateChange>
So these event types can be as granular as necessary, e.g. "CustomerAddressChanged" or simply "CustomerChanged", and it is up to the consumer to query for the information it needs based on the event type.
There is not a "one-size-fits-all" solution--sometimes it does make more sense to pass the relevant data with the event. Again, I agree with #Matt's answer if this is the direction you need to move in.
Edit Based on Comment
I would agree that using an ESB to query is generally not a good idea. Some people use an ESB this way, but IMHO it's a bad practice.
Your original question and your comments to this answer and to Matt's talk about only including fields that have changed. This would definitely be problematic in many languages, where you would have to somehow distinguish between a property being empty/null and a property not being included in the event. If the event is getting serialized/de-serialized from/to a static type, it will be painful (if not impossible) to know the difference between "First Name is being set to NULL" and "First Name is missing because it didn't change".
Based on your comment that this is about synchronization of systems, my recommendation would be to send the full set of data on each change (assuming signal+query is not an option). That leaves the interpretation of the data up to each consuming system, and limits the responsibility of the publisher to emitting a more generic event, i.e. "Customer 1234 has been modified to X state". This event seems more broadly useful than the other options, and if other systems receive this event, they can interpret it as they see fit. They can dump/rewrite their own data for Customer 1234, or they can compare it to what they have and update only what changed. Sending only what changed seems more specific to a single consumer or a specific type of consumer.
All that said, I don't think any of your proposed solutions are "right" or "wrong". You know best what will work for your unique situation.
Events should be used to describe intent as well as details, for example, you could have a CustomerRegistered event with all the details for the customer that was registered. Then later in the stream a CustomerMadeGoldAccount event that only really needs to capture the customer Id of the customer who's account was changed to gold.
It's up to the consumers of the events to build up the current state of the system that they are interested in.
This allows only the most pertinent information to be stored in each event, imagine having hundreds of properties for a customer, if every command that changed a single property had to raise an event with all the properties before and after, this gets unwieldy pretty quickly. It's also difficult to determine why the change occurred if you just publish a generic CustomerModified event, which is often a question that is asked about the current state of an entity.
Only capturing data relevant to the event means that the command that issues the event only needs to have enough data about the entity to validate the command can be executed, it doesn't need to even read the whole customer entity.
Subscribers of the events also only need to build up a state for things that they are interested in, e.g. perhaps an 'account level' widget is listening to these events, all it needs to keep around is the customer ids and account levels so that it can display what account level the customer is at.
Instead of trying to convey everything through payload xmls' fields, you can distinguish between different operations based on -
1. Different endpoint URLs depending on the operation(this is preferred)
2. Have an opcode(operation code) as an element in the xml file which tells which operation is to used to handle the incoming request.(more nearer to your examples)
There are a few enterprise patterns applicable to your business case - messaging and its variants, and if your system is extensible then Enterprise Service Bus should be used. An ESB allows reliable handling of events and processing.
So I'm currently diving the CQRS architecture along with the EventStore "pattern".
It opens applications to a new dimension of scalability and flexibility as well as testing.
However I'm still stuck on how to properly handle data migration.
Here is a concrete use case:
Let's say I want to manage a blog with articles and comments.
On the write side, I'm using MySQL, and on the read side ElasticSearch, now every time a I process a Command, I persist the data on the write side, dispatch an Event to persist the data on the read side.
Now lets say I've some sort of ViewModel called ArticleSummary which contains an id, and a title.
I've a new feature request, to include the article tags to my ArticleSummary, I would add some dictionary to my model to include the tags.
Given the tags did already exist in my write layer, I would need to update or use a new "table" to properly use the new included data.
I'm aware of the EventLog Replay strategy which consists in replaying all the events to "update" all the ViewModel, but, seriously, is it viable when we do have a billion of rows?
Is there any proven strategies? Any feedbacks?
I'm aware of the EventLog Replay strategy which consists in replaying
all the events to "update" all the ViewModel, but, seriously, is it
viable when we do have a billion of rows?
I would say "yes" :)
You are going to write a handler for the new summary feature that would update your query side anyway. So you already have the code. Writing special once-off migration code may not buy you all that much. I would go with migration code when you have to do an initial update of, say, a new system that requires some data transformation once off, but in this case your infrastructure would exist.
You would need to send only the relevant events to the new handler so you also wouldn't replay everything.
In any event, if you have a billion rows of data your servers would probably be able to handle the load :)
Im currently using the NEventStore by JOliver.
When we started, we were replaying our entire store back through our denormalizers/event handlers when the application started up.
We were initially keeping all our data in memory but knew this approach wouldn't be viable in the long term.
The approach we use currently is that we can replay an individual denormalizer, which makes things a lot faster since you aren't unnecessarily replaying events through denomalizers that haven't changed.
The trick we found though was that we needed another representation of our commits so we could query all the events that we handled by event type - a query that cannot be performed against the normal store.
I have a message model and I want it to have several receivers, possibly a lot of them.
I would also like to be able to tell for each receiver if the message was viewed or not (read/unread). Also I would like a receiver to be able to delete the message.
The two possible solutions are the following, for each I have a Message model an User model.
For the first (using the ideas presented here http://www.google.com/events/io/2009/sessions/BuildingScalableComplexApps.html)
I have a MessageReceivers class which has a ListProperty containing the users that will receive the message and set the parent to the message. I query of this with messages = db.GqlQuery('SELECT __key__ FROM MessageReceivers WHERE receivers = :1', user) and the do a db.get([ key.parent() for key in messages ]).
The problem I have which this is that I'm not sure how to store the state of the message: whether it is read or not and a subsequent issue whether the user has new messages. An additional issue would be the overhead of deleting a message (would have to remove user from receivers list property)
For the second: I have a MessageReceiver for each receiver it has links to message and to user and also stores the state (read/unread).
Which of this two approached do you consider that it has a better performance? And in the case of the first do you have any suggestion on handling the status of the message.
I've implement first option in production. Drawback is that ListProperty is limited to 2500 entries if you use custom index. Shameless plug: See my blog bost http://bravenewmethod.wordpress.com/2011/03/23/developing-on-google-app-engine-for-production/
Read state storing. I did this by implementing an entity that stored unread messages up to few months back and then just assumed older ones read. Even simpler is to query the messages in date order, and store the last known message timestamp in entity and assume all older as read. I don't recommended keeping long history in entity with huge list property, because reading and storing such entities can get really slow.
Message deletion is expensive, no way around that.
If you need to store state per message, your best option is to write one entity per recipient, with read state (and anything else, such as flags, etcetera), rather than using the index relation pattern.