What I am trying to accomplish on higher level:
I have a function that does I/O and generate messages. I have multiple subscriber clients that can subscribe or leave at any time. When a new client subscribes, it should get x number of previous output before streaming new messages (much like unix "tail -f").
My idea was to send-off the messages to an agent, which is a ring buffer. New clients will read the agent and then add-watch to the agent. Problem is, how can I ensure no new message arrive between reading and add-watch?
Next idea was to create 2 refs, one for a list of clients, one for the ring buffer. I can then add clients or post message in transactions. Problem is, when I add clients, I have to read the ring buffer and send it to the client (I/O). This is side effect in transaction that may be retried.
Last idea is to use locks, but that can't be the only way?
Related
I understand that standard SQS uses "at least once" delivery while FIFO messages are delivered exactly once. I'm trying to weigh standard queues vs FIFO for my application, and one factor is how long it takes for the duplicated message to arrive.
I intend to consume messages from SQS then post the data I received to an idempotent third-party API. I understand that with standard SQS, there's always a risk of me overwriting more recent data with the old duplicated data.
For example:
Message A arrives, I post it onwards.
Message A duplicate arrives, I post it onwards.
Message B arrives, I post it onwards.
All fine ✓
On the other hand:
Message A arrives, I post it onwards.
Message B arrives, I post it onwards.
Message A duplicate arrives - I post it and overwrite the latest data, which was B! ✖
I want to measure this risk, i.e. I want to know how long the duplicate message should take to arrive. Will the duplicate message take roughly the same amount of time to arrive, as the original message?
Maybe it's useful to understand how message duplication occurs. As far as I know this isn't documented in the official docs, but instead it's my mental model of how it works. This is an educated guess.
Whenever you send a message to SQS (SendMessage API), this message arrives at the SQS webservice endpoint, which is one of probably thousands of servers. This endpoint receives your message, duplicates it one or more times and stores these duplicates on more than one SQS server. After it has received confirmation from at least two SQS servers, it acknowledges to the client that the message has been received.
When you call the ReceiveMessage API only a subset of the SQS servers that handle your queue are queried for messages. When a message is returned, these servers communicate to their peers, that this message is currently in-flight and the visibility timeout starts. This doesn't happen instantaneously, as it's a distributed system. While this ReceiveMessage call takes place another consumer might also do a ReceiveMessage call and happen to query one of the servers that have a replica of the message, before it's marked as in-flight. That server hands out the message and now you have to consumers working on it.
This is just one scenario, which is the result of this being a distributed system.
There are a couple of edge cases that can happen as the result of network issues, e.g. when the SQS response to the initial SendMessage gets lost and the client thinks the message didn't arrive and sends it again - poof, you got another duplicate.
The point being: things fail in weird and complex ways. That makes measuring the risk of a delayed message difficult. If your use case can't handle duplicate and out of order messages, you should go for FIFO, but that will inherently limit your throughput. Alternatives are based on distributed locking mechanisms and keeping track of which messages you have already processed, which are complex tools to solve a complex problem.
Currently we want to pull down an entire FIFO queue, and process the contents, and if any issues, release messages back into the queue.
The problem is, that currently AWS only gives us 10 messages, and won't give us 10 more (which is the way you get bulk messages in SQS, multiple 10 max message requests) until we delete or release the first 10.
We need to get more than 10 though. Is this not possible? We understand we can set the group_id to a random string, and that allows processing more, but then the order isn't guaranteed, which defeats the purpose of FIFO.
I managed to reproduce your results -- I could retrieve 10 messages, but then running the same command again would not return another set of messages.
The relevant documentation seems to be:
While messages with a particular MessageGroupId are invisible, no more messages belonging to the same MessageGroupId are returned until the visibility timeout expires. You can still receive messages with another MessageGroupId as long as it is also visible.
I suspect (just a theory!) that this is to preserve the ordering of messages... If a client asked for a set of messages and they are still being processed, there is the chance that the messages might be returned to the queue. Therefore, no further messages are provided until the original messages are deleted or pass their visibility timeout.
This is only a behaviour of FIFO queues.
It seems that you will need to receive and delete all messages to be able to access them all. I would suggest:
Receive one (or more) message.
Process it. If everything worked, delete the message.
If there were problems, push the message to a new queue.
Once the queue is empty, you would need to read from the new queue and send them back to the original queue (which should preserve ordering).
If you frequently require more capabilities that Amazon SQS provides, you could consider using Amazon MQ – Managed message broker service for ActiveMQ. It has many more capabilities (but is accordingly less 'simple').
If you set another MessageGroupId, you can get another 10 messages, even you don't release or delete the previous ones.
https://docs.aws.amazon.com/AWSSimpleQueueService/latest/SQSDeveloperGuide/using-messagegroupid-property.html
If you're using PUSH sockets, you'll find that the first PULL socket to connect will grab an unfair share of messages. The accurate rotation of messages only happens when all PULL sockets are successfully connected, which can take some milliseconds. As an alternative to PUSH/PULL, for lower data rates, consider using ROUTER/DEALER and the load balancing pattern.
So one way to do sync in PUSH/PULL is using the load balancing pattern.
For this specific case below, I wonder whether there is another way to do sync:
I could set the PULL endpoint in worker to block until the connection successfully setup, and then send a special message via worker's PULL endpoint to 'sink'. After 'sink' receives #worker's special messages, 'sink' sends a message with REQ-REP to 'ventilator' to notify that all workers ready. 'ventilator' starts to distribute jobs to workers.
Is it reliable?
The picture is from here
Yes, so long as the Sink knows how many Workers to wait for before telling the Ventilator that it's OK to start sending messages. There's the question of whether the special messages from the Workers get through if they start up before the Sink connects - but you could solve that by having them keep sending their special message until they start getting data from the Ventilator. If you do this, the Sink would of course simply ignore any duplicates it receives.
Of course, that's not quite the same as the Workers having a live, working connection to the Ventilator, but that could itself be sending out special do-nothing messages that the Workers receive. When they receive one of those that's when they can start sending a special message to the Sink.
I have a Kinesis producer which writes a single type of message to a stream. I want to process this stream in multiple, completely different consumer applications. So, a pub/sub with a single publisher for a given topic/stream. I also want to make use of checkpointing to ensure that each consumer processes every message written to the stream.
Initially, I was using the same App Name for all consumers and producers. However, I started getting the following error once I started more than one consumer:
com.amazonaws.services.kinesis.model.InvalidArgumentException: StartingSequenceNumber 49564236296344566565977952725717230439257668853369405442 used in GetShardIterator on shard shardId-000000000000 in stream PackageCreated under account ************ is invalid because it did not come from this stream. (Service: AmazonKinesis; Status Code: 400; Error Code: InvalidArgumentException; Request ID: ..)
This seems to be because consumers are clashing with their checkpointing as they are using the same App Name.
From reading the documentation, it seems the only way to do pub/sub with checkpointing is by having a stream per consumer application, which requires each producer to know about all possible consumers. This is more tightly coupled than I want; it's really just a queue.
It seems like Kafka supports what I want: arbitrary consumption of a given topic/partition, since consumers are completely in control of their own checkpointing. Is my only option to move to Kafka, or some other alternative, if I want pub/sub with checkpointing?
My RecordProcessor code, which is identical in each consumer:
override def processRecords(processRecordsInput: ProcessRecordsInput): Unit = {
log.trace("Received record(s) from kinesis")
for {
record <- processRecordsInput.getRecords
json <- jawn.parseByteBuffer(record.getData).toOption
msg <- decode[T](json.toString).toOption
} yield subscriber ! msg
processRecordsInput.getCheckpointer.checkpoint()
}
The code parses the message and sends it off to the subscriber. For now, I'm simply marking all messages as successfully received. I can see messages being sent on the AWS Kinesis dashboard, but no reads happen, presumably because each application has its own AppName and doesn't see any other messages.
The pattern you want, that of one publisher to & multiple consumers from one Kinesis stream, is supported. You don't need a separate stream per consumer.
How do you do that? You need to give a different application-name to every consumer. That way, checkpointing info of one consumer won't collide with that of another.
Check the first response to this: https://forums.aws.amazon.com/message.jspa?messageID=554375
I am new to Amazon Web Services and am currently trying to get my head around how Simple Queue Service (SQS) works.
In the link ReceiveMessage the following is mentioned:
Short poll is the default behavior where a weighted random set of
machines is sampled on a ReceiveMessage call. This means only the
messages on the sampled machines are returned. If the number of
messages in the queue is small (less than 1000), it is likely you will
get fewer messages than you requested per ReceiveMessage call. If the
number of messages in the queue is extremely small, you might not
receive any messages in a particular ReceiveMessage response; in which
case you should repeat the request.
What I understand there is one queue and many machines/instances can read the messages. What is not clear to me is what does "weighted random set of machines" means? Is there more than one queue on a number of machines? Clearly I am lacking some knowledge on on SQS works.
I believe what this means is that because SQS is geographically distributed, not all of the machines (amazon's servers that have your queue) will have the exact same queue content at all times because they won't always be in sync with each other at every instant.
You don't know or control from which of amazons servers it will serve messages from, it uses an algorithm to figure out which messages are sent to you when you request some. That is why you don't always get messages when you ask for them, and occasionally the same message will get served up more than once; you need to make sure whatever your processing entails it can deal with the possibility that it is processing something that has already been processed by another of your worker machines.