DynamoDB AutoScaling still throttles - amazon-dynamodb

I have an application using DynamoDB and I noticed they just implemented autoscaling which is awesome. I love the concept and the timing for my app is pretty perfect. However I am still getting some issues that I wonder if I can't tweak settings to remove.
My application gets definite spikes in usage so I think this is an ideal thing to use, however with autoscaling on I still am getting some throttling. Here is my read graphs for the last 12 hours:
As you can see, when it spikes the usage is set low, so it throttles for a minute or two until the update kicks in, then works. That's ok I guess and better than not scaling, but I would like it not to throttle at all...
Is there any way to tell DynamoDB to never throttle unless it goes over 100 (or 200 or whatever I set as the top limit)? Just if it gets a surge turn up the throughput for 15 minutes or whatever until the surge is over?

Autoscaling uses CloudWatch. You can see these alarms by going to the CloudWatch dashboard and look for the alarms that include your table name and "DO NOT EDIT OR DELETE" in the description.
Why am I telling you this?
Well, the CloudWatch has some minimal period granularity. Currently it's 1 minute. It means that it will wait at least for 1 minute before firing any event to its listeners. Therefore, it will take at least one minute after the load starts until the capacity is increased. Actually it will be even more, since increasing the capacity also takes time. Bottom line: if you have very large spike, some requests may be throttled, since the the auto-scaling will not yet take effect and bursting can be exhausted.
The simple, but costly, solution will be increasing the initial capacity.
If you know about the upcoming spike in advance (e.g. you have some job running periodically, or customer peaks in certain times) you can use API to modify autoscaling programmatically.

Related

Why wouldn't a small Firebase Functions app just use a single Function to handle logic?

...aside from the benefit in separate performance monitoring and logging.
For logging, I am confident I can get granularity through manually adding the name of the "routine" to each call. This is how it is now with several discrete Functions for different parts of the system:
There are multiple automatic logs: start and finish of the routine, for example. It would be more challenging to find out how expensive certain routines are, but it would not be impossible.
The reason I want the entire logic of the application handled by a single handle function is because of reducing cold starts: one function means only one container that can be persistently kept alive when there are very few users of the app.
If a month is ~2.6m seconds and we assume the system uses 1 GB RAM and 1 GHz CPU frequency at all times, that's:
2600000 * 0.0000025 + 2600000 * 0.000001042 = USD$9.21 a month
...for one minimum instance.
I should also state that all of my functions have the bare minimum amount of global scope code; it just sets up Firebase assets (RTDB and Firestore).
From a billing, performance (based on user wait time), and user/developer experience perspective, is there any reason why it would be smart to keep all my functions discrete?
I'd also accept an answer saying "one single function for all logic is reasonable" as long as there's a reason for it.
Thanks!
If you have very small app with ~5 end points and very low traffic. Sure you could do something like this. But why not do it:
billing and performance
The important thing to realize is that with every request a new instance of your function is created. Which means there could be 10s of them running at the same time.
If you would like to have just 1 instance handling all the traffic you should explore GCP Cloud run, where you have 1 container handling multiple requests and scaling only when it's not sufficient.
Imagine you have several end-points and every one of them have different performance requirements.
1 can need only 128MB or RAM
1 can need 1GB RAM
(FYI: You can control the CPU MHz of the function via the RAM settings too - which can speed up execution in some cases)
If you had only 1 function with 1GB of ram. Every request would allocate such function and in some cases most of the memory could go to waste.
But if you split it into multiple, some requests will require much less resources and can save you $ when we talk about bigger amount of executions / month. (tens of thousands+).
Let's imagine function, 3 second execution, 10k executions/month:
128MB would cost you $0.0693
1024MB would cost you $0.495
As you can see, with small app the difference could be nothing. But if you scale it matters. (*The cost can vary based on datacenter)
As for the logging, I don't think it matters. Usually in bigger systems there could be messages traveling trough several functions so you have to deal with that anyway.
As for the cold start. You just need good UI to facilitate that. At first I was worry about it in our apps but later on, you just get used to it that some action can take ~2s to execute (cold start). And you should have the UI "loading" regardless, because you don't know if the function will take ~100ms or 3s due to bad connection.

How to handle dynamodb WCU limitation ? is it queued when you go above the limit?

I need to feed the db with something like 10k items, I don't need to rush and can/want stay below the 25wcu free plan.
It will take something like 6.5minutes (10000requests/25requests/sec).
Here is the question. I will loop on a json to feed the base do I have to handle the number of request by second myself or can I push to the max and it will be queued ?
I read that I may just have an error message (400?) when I exceed the limit can I just brutally retry the failed requests (eventually making more fail) until my 10k items are put in the db ?
tldr; => best way/strategy to feed the base knowing there is a limit of calls/sec
ps: it's run from a lambda idk if it matters.
Your calculation is a little bit off unless you do this constantly, because AWS actually allows you to burst a little bit (docs):
DynamoDB provides some flexibility in your per-partition throughput
provisioning by providing burst capacity. Whenever you're not fully
using a partition's throughput, DynamoDB reserves a portion of that
unused capacity for later bursts of throughput to handle usage spikes.
DynamoDB currently retains up to 5 minutes (300 seconds) of unused
read and write capacity. During an occasional burst of read or write
activity, these extra capacity units can be consumed quickly—even
faster than the per-second provisioned throughput capacity that you've
defined for your table.
Since 300 (seconds) * 25 (WCU) = 7500 this leaves you with about 7.5k items until it will actually throttle.
Afterwards just responding to the ProvisionedThroughputExceeded error by retrying later is fine - but make sure to add a small delay between retries (e.g. 1 second) as you know that it takes time for the new WCU to flow into the tocken bucket. Immediately retrying and hammering the API is not a kind way to use the service and might look like a DoS attack.
You can also write the items in Batches to reduce the amount of network requests as network throughput is also limited in Lambda. This makes handling the ProvisionedThroughputExceeded error slightly more cumbersome, because you need to inspect the response which items failed to write, but will probably be a net positive.

Amazon DynamoDB throttling

I've enabled auto-scaling for our dynamo-db table. It has a target utilization at 30% but it keeps throttling.
See this example screenshot where throttling is happening
As you can see it's exactly scaling up as you want it too. But I don't understand why it's still throttling. Its almost always below the provisioned throughput.
Can anyone explain what's going wrong and why it's still throttling?
Thanks,
Hendrik
Very hard to tell from the graph, and limited information.
Some thoughts:
AutoScaling can take 5 - 10 minutes to kick in. This is not fast enough if there is a sudden increase in usage. Perhaps you are seeing throttling in that 5 - 10 minute window before it scales up.
If you set CloudWatch Metrics to 1 min interval, you might see whats going on in a big more detail.
As mkobit mentioned, you might be hitting throughput on your Partition, depending on how your data is structured.
Your capacity units are evenly distributed between your partitions. So you may hit the capacity limit on your partition you have and which records you are trying to access, but not your table throughput.
This also depends on the amount of data you have stored, number of partitions etc.
HTH

How can I find the average number of concurrent users for IIS to simulate during a load/performance test?

I'm using JMeter for load testing. I'm going through and exercise of finding the max number of concurrent threads (users) that our webserver can handle by simply increasing the # of threads in my distributed JMeter test case, and firing off the test.
Then -- it struck me, that while the MAX number may be useful, the REAL number of users that my website actually handles on average is the number I need to make the test fruitful.
Here are a few pieces of information about our setup:
This is a mixed .NET/Classic ASP site. Upon login, a browser session (with timeout) is created in both for the users.
Each session times out after 60 minutes.
Is there a way using this information, IIS logs, performance counters, and/or some calculation that will help me determine the average # of concurrent users we handle on our production site?
You might use logparser with the QUANTIZE function to determine the peak number of requests over a suitable interval.
For a 10 second window, it would be something like:
logparser "select quantize(to_localtime(to_timestamp(date,time)), 10) as Qnt,
count(*) as Hits from yourLogFile.log group by Qnt order by Hits desc"
The reported counts won't be exactly the same as threads or users, but they should help get you pointed in the right direction.
The best way to do exact counts is probably with performance counters, but I'm not sure any of the standard ones works like you would want -- you'd probably need to create a custom counter.
I can see a couple options here.
Use Performance Monitor to get the current numbers or have it log all day and get an average. ASP.NET has a Requests Current counter. According to this page Classic ASP also has a Requests current, but I've never used it myself.
Run the IIS logs through Log Parser to get the total number of requests and how long each took. I'm thinking that if you know how many requests come in each hour and how long each took, you can get an average of how many were running concurrently.
Also, keep in mind that concurrent users isn't quite the same as concurrent threads on the server. For one, multiple threads will be active per user while content like images is being downloaded. And after that the user will be on the page for a few minutes while the server is idle.
My suggestion is that you define the stop conditions first, such as
Maximum CPU utilization
Maximum memory usage
Maximum response time for requests
Other key parameters you like
It is really subjective to choose the parameters and I personally cannot provide much experience on that.
Secondly you can see whether performance counters or IIS logs can map to the parameters. Then you set up proper mappings.
Thirdly you can start testing by simulating N users (threads) and see whether the stop conditions hit. If not hit, you can go to a higher number. If hit, you can use a smaller number. Recursively you will find a rough number.
However, that never means your web site in real world can take so many users. No simulation so far can cover all the edge cases.

Asp.net guaranteed response time

Does anybody have any hints as to how to approach writing an ASP.net app that needs to have a guaranteed response time?
When under high load that would normally cause us to exceed our desired response time, we want to throw out an appropriate number of requests, so that the rest of the requests can return before the max response time. Throwing out requests based on exceeding a fixed req/s is not viable, as there are other external factors that will control response time that cause the max rps we can safely support to fiarly drastically drift and fluctuate over time.
Its ok if a few requests take a little too long, but we'd like the great majority of them to meet the required response time window. We want to "throw out" the minimal or near minimal number of requests so that we can process the rest of the requests in the allotted response time.
It should account for ASP.Net queuing time, ideally the network request time but that is less important.
We'd also love to do adaptive work, like make a db call if we have plenty of time, but do some computations if we're shorter on time.
Thanks!
SLAs with a guaranteed response time require a bit of work.
First off you need to spend a lot of time profiling your application. You want to understand exactly how it behaves under various load scenarios: light, medium, heavy, crushing.. When doing this profiling step it is going to be critical that it's done on the exact same hardware / software configuration that production uses. Results from one set of hardware have no bearing on results from an even slightly different set of hardware. This isn't just about the servers either; I'm talking routers, switches, cable lengths, hard drives (make/model), everything. Even BIOS revisions on the machines, RAID controllers and any other device in the loop.
While profiling make sure the types of work loads represent an actual slice of what you are going to see. Obviously there are certain load mixes which will execute faster than others.
I'm not entirely sure what you mean by "throw out an appropriate number of requests". That sounds like you want to drop those requests... which sounds wrong on a number of levels. Doing this usually kills an SLA as being an "outage".
Next, you are going to have to actively monitor your servers for load. If load levels get within a certain percentage of your max then you need to add more hardware to increase capacity.
Another thing, monitoring result times internally is only part of it. You'll need to monitor them from various external locations as well depending on where your clients are.
And that's just about your application. There are other forces at work such as your connection to the Internet. You will need multiple providers with active failover in case one goes down... Or, if possible, go with a solid cloud provider.
Yes, in the last mvcConf one of the speakers compares the performance of various view engines for ASP.NET MVC. I think it was Steven Smith's presentation that did the comparison, but I'm not 100% sure.
You have to keep in mind, however, that ASP.NET will really only play a very minor role in the performance of your app; DB is likely to be your biggest bottle neck.
Hope the video helps.

Resources