When considering a service in NServiceBus at what point do you start questioning how many messages handled by a service is too much and start to break these into a new service?
Consider the following: I have a sales service which can currently be broken into a few distinct business components, these are sales order validation, sales order processing, purchase order validation and purchase order processing.
There are currently about 20 message handlers and 2 sagas used within this service. My concern is that during high volume traffic from my website this can cause an initial spike in the messages to jump into the hundreds. Considering that the messages need to be processed in the order they are taken off the queue this can cause a delay for the last in the queue ( depending on what processing each message does).
When separating concerns within a service into smaller business components I find this makes things a little easier. Sure, it's a logical separation, but it seems to provide a layer of clarity and understanding. To me it seems it seems an easier option to do this than creating new services where in the end the more services I have the more maintenance I need to do.
Does anyone have any similar concerns to this?
I think you have actually answered you own question :)
As soon as the message volume reaches a point where the lag becomes an issue you could look to instance your endpoint. You do not necessarily need to reduce the number of handlers. You could simply install the service a number of times and have specific message types sent to the relevant endpoint by mapping.
So it becomes a matter of a simple instance installation and some config changes. So you can then either split messages on sending so that messages from a particular source end up on a particular endpoint (maybe priority) or on message type.
I happened to do the same thing on a previous project (not using NServiecBus though) where we needed document conversion messages coming from the UI to be processed ASAP. We simply installed the conversion service again with its own set of queues and changed the UI configuration to send the conversion messages to the new endpoint. The background conversion messages were still going to the previous endpoint. So here the source determined the separation.
Related
I am using an Axon Event Tracking processor. Sometimes events take longer that 10 seconds to process.
This seems to cause the message to be processed again and this appears in the log "Releasing claim of token X/0 failed. It was owned by another node."
If I up the number of segments it does not log this BUT the event is still processed twice so I think this might be misleading. (I think I was mistaken about this)
I have tried adjusting the fetchDelay, cleanupDelay and tokenClaimInterval. None of which has fixed this. Is there a property or something that I am missing?
Edit
The scenario taking longer than 10 seconds is making a HTTP request to an external service.
I'm using axon 4.1.2 with all default configuration when using with Spring auto configuration. I cannot see the Releasing claim on token and preparing for retry in [timeout]s log.
I was having this issue with a single segment and 2 instances of the application. I realised I hadn't increased the number of segments like I thought I had.
After further investigation I have discovered that adding an additional segment seems to have stopped this. Even if I have for example 2 segments and 6 applications it still doesn't reappear, however I'm not sure how this is different to my original scenario of 1 segment and 2 application?
I didn't realise it would be possible for multiple threads to grab the same tracking token and process the same event. It sounds like the best action would be to put an idem-potency check before the HTTP call?
The Releasing claim of token [event-processor-name]/[segment-id] failed. It was owned by another node. message can only occur in three scenarios:
You are performing a merge operation of two segments which fails because the given thread doesn't own both segments.
The main event processing loop of the TrackingEventProcessor is stopped, but releasing the token claim fails because the token is already claimed by another thread.
The main event processing loop has caught an Exception, making it retry with a exponential back-off, and it tries to release the claim (which might fail with the given message).
I am guessing it's not options 1 and 2, so that would leave us with option 3. This should also mean you are seeing other WARN level messages, like:
Releasing claim on token and preparing for retry in [timeout]s
Would you be able to share whether that's the case? That way we can pinpoint a little better what the exact problem is you are encountering.
By the way, very likely you have several processes (event handling threads of the TrackingEventProcessor) stealing the TrackingToken from one another. As they're stealing an un-updated token, both (or more) will handled the same event. Hence why you see the event handler being invoked twice.
Obviously undesirable behavior and something we should resolve for you. I would like to ask you to provide answers to my comments under the question, as right now I have to little to go on. Let us figure this out #Dan!
Update
Thanks for updating your question #dan, that's very helpful.
From what you've shared, I am fairly confident that both instances are stealing the token from one another. This does depend though on whether both are using the same database for the token_entry table (although I am assuming they are).
If they are using the same table, then they should "nicely" share their work, unless one of them takes to long. If it takes to long, the token will be claimed by another process. This other process in this case is the thread of the TEP of your other application instance. The "claim timeout" is defaulted to 10 seconds, which also corresponds with the long running event handling process.
This claimTimeout is adjustable though, by invoking the Builder of the JpaTokenStore/JdbcTokenStore (depending on which you are using / auto wiring) and calling the JpaTokenStore.Builder#claimTimeout(TemporalAmount) method. And, I think this would be required on your end, giving the fact you have a long running operation.
There are of course different ways of tackling this. Like, making sure the TEP is only ran on a single instance (not really fault tolerant though), or offloading this long running operation to a schedule task which is triggered by the event.
But, I think we've found the issue at least, so I'd suggest to tweak the claimTimeout and see if the problem persists.
Let us know if this resolves the problem on your end #dan!
I am building a project from scratch using event-sourcing with Java and Cassandra.
My apps we be based on microservices and in some use cases information will be processed asynchronously. I was wondering what part a Message Queue (such as Rabbit, Active MQ Artemis, Kafka, etc) would play to improve the technology stack in this environment and if I understand the scenarios if I won't use it.
I would start with separating messaging infrastructure like RabbitMQ from event streaming/storing/processing like Kafka. These are two different things made for two (or more) different purposes.
Concerning the event sourcing, you have to have a place where you must store events. This storage must be append-only and support fast reads of unstructured data based on an identity. One example of such persistence is the EventStore.
Event sourcing goes together with CQRS, which means you have to project your changes (event) to another store, which you can query. This is done by projecting events to that store, this is where events get processed to change the domain object state. It is important to understand that using message infrastructure for projections is generally a bad idea. This is due to the nature of messaging and two-phase commit issue.
If you look at how events get persisted, you can see that they get saved to the store as one transaction. If you then need to publish events, this will be another transaction. Since you are dealing with two different pieces of infrastructure, things can get broken.
The messaging issue as such is that messages are usually guaranteed to be delivered "at least once" and the order of messages is usually not guaranteed. Also, when your message consumer fails and NACKs the message, it will be redelivered but usually a bit later, again breaking the sequence.
The ordering and duplication concerns, whoever, do not apply to event streaming servers like Kafka. Also, the EventStore will guarantee once only event delivery in order if you use catch-up subscription.
In my experience, messages are used to send commands and to implement event-driven architecture to connect independent services in a reactive way. Event stores, at the other hand, are used to persist events and only events that get there are then projected to the query store and also get published to the message bus.
Make sure you are clear on the distinction between send(command) and publish(event). Udi Dahan touches on that topic in his essay on busses and brokers.
In most cases where you are event sourcing, you do not want to be reconstructing state from published events. If you need state, then query the technical authority/book of record for the history, and reconstruct the state from the history.
On the other hand, event driven activity off of a message queue should be fine. When a single event (plus the subscriber's state) has everything you need, then running off of the bus is fine.
In some cases, you might do both. For example, if you were updating cached views, you'd subscribe to various BobChanged events to know when your cached data was stale; to rebuild a stale view, you would reload a representation of the history and transform it into an updated view.
In the world of event-sourcing applications, message queues usually allow you to implement publish-subscribe pattern style of communication between producers and consumers. Also, they usually help you with delivery guarantees: which messages were delivered to which subscribers and which ones were not.
But they don't store all messages indefinitely. You need to have an event store to do any kind of event sourcing.
The question is not 'to queue or not to queue', but it is more like:
can this thing store huge volume of events indefinitely?
does it have publish-subscribe capabilities?
does it provide at-least-once delivery guarantees?
So, you should use something like Kafka or EventStore to have all that out-of-the-box. Alternatively, you can combine event store with message queue manually, but this is going to be more involved.
We are using Micro services architecture where top services are used for exposing REST API's to end user and backend services does the work of querying database.
When we get 1 user request we make ~30k requests to backend service. We are using RxJava for top service so all 30K requests gets executed in parallel.
We are using haproxy to distribute the load between backend services.
However when we get 3-5 user requests we are getting network connection Exceptions, No Route to Host Exception, Socket connection Exception.
What are the best practices for this kind of use case?
Well you ended up with the classical microservice mayhem. It's completely irrelevant what technologies you employ - the problem lays within the way you applied the concept of microservices!
It is natural in this architecture, that services call each other (preferably that should happen asynchronously!!). Since I know only little about your service APIs I'll have to make some assumptions about what went wrong in your backend:
I assume that a user makes a request to one service. This service will now (obviously synchronously) query another service and receive these 30k records you described. Since you probably have to know more about these records you now have to make another request per record to a third service/endpoint to aggregate all the information your frontend requires!
This shows me that you probably got the whole thing with bounded contexts wrong! So much for the analytical part. Now to the solution:
Your API should return all the information along with the query that enumerates them! Sometimes that could seem like a contradiction to the kind of isolation and authority over data/state that the microservices pattern specifies - but it is not feasible to isolate data/state in one service only because that leads to the problem you currently have - all other services HAVE to query that data every time to be able to return correct data to the frontend! However it is possible to duplicate it as long as the authority over the data/state is clear!
Let me illustrate that with an example: Let's assume you have a classical shop system. Articles are grouped. Now you would probably write two microservices - one that handles articles and one that handles groups! And you would be right to do so! You might have already decided that the group-service will hold the relation to the articles assigned to a group! Now if the frontend wants to show all items in a group - what happens: The group service receives the request and returns 30'000 Article numbers in a beautiful JSON array that the frontend receives. This is where it all goes south: The frontend now has to query the article-service for every article it received from the group-service!!! Aaand your're screwed!
Now there are multiple ways to solve this problem: One is (as previously mentioned) to duplicate article information to the group-service: So every time an article is assigned to a group using the group-service, it has to read all the information for that article form the article-service and store it to be able to return it with the get-me-all-the-articles-in-group-x query. This is fairly simple but keep in mind that you will need to update this information when it changes in the article-service or you'll be serving stale data from the group-service. Event-Sourcing can be a very powerful tool in this use case and I suggest you read up on it! You can also use simple messages sent from one service (in this case the article-service) to a message bus of your preference and make the group-service listen and react to these messages.
Another very simple quick-and-dirty solution to your problem could also be just to provide a new REST endpoint on the articles services that takes an array of article-ids and returns the information to all of them which would be much quicker. This could probably solve your problem very quickly.
A good rule of thumb in a backend with microservices is to aspire for a constant number of these cross-service calls which means your number of calls that go across service boundaries should never be directly related to the amount of data that was requested! We closely monitory what service calls are made because of a given request that comes through our API to keep track of what services calls what other services and where our performance bottlenecks will arise or have been caused. Whenever we detect that a service makes many (there is no fixed threshold but everytime I see >4 I start asking questions!) calls to other services we investigate why and how this could be fixed! There are some great metrics tools out there that can help you with tracing requests across service boundaries!
Let me know if this was helpful or not, and whatever solution you implemented!
I'm in the process of building a consumer service for a Kafka topic. Each message contains a url to which my service will make an http request. Each message / url is completely independent from other messages / urls.
The problem I'm worried about is how to handle long-running requests. It's possible for some http requests to take 50+ minutes before a response is returned. During that time, I do not want to hold up any other messages.
What is the best way to parallelize this operation?
I know that Kafka's approach to parallelism is to create partitions. However, from what I've read, it seems that you need to define the number of partitions up front when I really want an infinite or dynamic number of partitions (ideally each message gets its own partition created on the fly)
As an example, let's say I create 1,000 partitions. If 1,001+ messages are produced to my topic, the first 1,000 requests will be made but every message after that will be queued up until the previous request in that partition finishes.
I've thought about making the http requests asynchronous but then I seem to run into a problem when determining what offset to commit.
For instance, on a single partition I can have a consumer read the first message and make an async request. It provides a callback function which commits that offset to Kafka. While that request is waiting, my consumer reads the next message and makes another async request. If that request finishes before the first it will commit that offset. Now, what happens if the first request fails for some reason or my consumer process dies? If I've already committed a higher offset, it sounds like this means my first message will never get reprocessed, which is not what I want.
I'm clearly missing something when it comes to long-running, asynchronous message processing using Kafka. Has anyone experienced a similar issue or have thoughts on how to best solve this? Thanks in advance for taking the time to read this.
You should look at Apache Storm for the processing portion of your consumer and leave the message storage and retrieval to Kafka. What you've described is a very common use case in Big Data (although the 50+ minute thing is a bit extreme). In short, you'll have a small number of partitions for your topic and let Storm stream processing scale the number of components ("bolts" in Storm-speak) that would actual make the http requests. A single spout (the kind of storm component that reads data from an external source) could read the messages from the Kafka topic and stream them to the processing bolts.
I've posted an open source example of how to write a Storm/Kafka application on github.
Some follow-on thoughts to this answer:
1) While I think Storm is the correct platform approach to take, there's no reason you couldn't roll your own by writing a Runnable that performs the http call and then write some more code to make a single Kafka consumer read messages and process them with multiply-threaded instances of your runnable. The management code required is a bit interesting, but probably easier to write than what it takes to learn Storm from scratch. So you'd scale by adding more instances of the Runnable on more threads.
2) Whether you use Storm or your own multi-threaded solution, you'll still have the problem of how to manage the offset in Kafka. The short answer there is that you'll have to do your own complex offset management. Not only will you have to persist the offset of the last message you read from Kafka, but you'll have to persist and manage the list of in-flight messages currently being processed. In this way, if your app goes down, you know what messages were being processed and you can retrieve and re-process them when you start back up. The base Kafka offset persistence doesn't support this more complex need, but it's only there as a convenience for the simpler use cases anyway. You can persist your offsets info anywhere you like (Zookeeper, file system or any data base).
I'm having difficulty conceptualising a requirement I have into something that will fit into our nascent SOA/EDA
We have a component I'll call the Data Downloader. This is a facade for an external data provider that has both high latency and a cost associated with every request. I want to take this component and create a re-usable service out of it with a clear contract definition. It is up to me to decide how that contract should work, however its responsibilities are two-fold:
Maintain the parameter list (called a Download Definition) for an upcoming scheduled download
Manage the technical details of the communication to the external service
Basically, it manages the 'how' of the communication. The 'what' and the 'when' are the responsibilities of two other components:
The 'what' is managed by 'Clients' who are responsible for
determining the parameters for the download.
The 'when' is managed by a dedicated scheduling component. Because of the cost associated with the downloads we'd like to batch the requests intraday.
Hopefully this sequence diagram explains the responsibilities of the services:
Because each of the responsibilities are split out in three different components, we get all sorts of potential race conditions with async messaging. For instance when the Scheduler tells the Downloader to do its work, because the 'Append to Download Definition' command is asynchronous, there is no guarantee that the pending requests from Client A have actually been serviced. But this all screams high-coupling to me; why should the Scheduler necessarily know about any 'prerequisite' client requests that need to have been actioned before it can invoke a download?
Some potential solutions we've toyed with:
Make the 'Append to Download Definition' command a blocking request/response operation. But this then breaks the perf. and scalability benefits of having an EDA
Build something in the Downloader to ensure that it only runs when there are no pending commands in its incoming request queue. But that then introduces a dependency on the underlying messaging infrastructure which I don't like either.
Makes me think I'm thinking about this problem in a completely backward way. Or is this just a classic case of someone trying to fit a synchronous RPC requirement into an async event-driven architecture?
The thing I like most about EDA and SOA, is that it almost completely eliminates the notion of race condition. As long as your events are associated with some association key (e.g. downloadId), the problem you describe can be addressed with several solutions of different complexities - depending on your needs. I'm not sure I totally understand the described use-case but I will try my best
Out of the top of my head:
DataDownloader maintains a list of received Download Definitions and a list of triggered downloads. When a definition is received it is checked against the triggers list to see if the associated download has already been triggered, and if it was, execute the download. When a TriggerDownloadCommand is recieved, the definitions list is checked against a definition with the associated downloadId.
For more complex situation, consider using the Saga pattern, which is implemented by some 3rd party messaging infrastructures. With some simple configuration, it will handle both messages, and initiate the actual download when the required condition is satisfied. This is more appropriate for distributed systems, where an in-memory collection is out of the question.
You can also configure your scheduler (or the trigger command handler) to retry when an error is signaled (e.g. by an exception), in order to avoid that race condition, and ultimately give up after a specified timeout.
Does this help?