Regarding best practices and efficiency within Flink, what are the recommendations of when to split analytics into multiple tasks
For example, given a single topic in Kafka as the source of the data. If there were many simple operations that were to be carried out over the stream, such as: if some value is greater than x, or if x & y etc. What would be the point at which you would stop spending more rules into the same task and start to run them in parallel?
Is there any official recommendation for this?
It's hard to give a general recommendation. Performance-wise, it makes sense to put as much as possible into one job.
However, it's much more important to think about maintenance. I'd put everything in one job that is closely related, such that new features or bug fixes will likely only affect one job while at the same time, you also don't want to stop all analytics when upgrading one particular query.
Another dimension to think about is state size. It's related to restarts and update frequency (point above). If the state size becomes too big, restarting this one monster job takes a long time, which would be inefficient if you only touched a small fraction of the code.
Finally, it also depends on the relevance. If some part of your job is super important as it reflects the one KPI that drives your business, then you probably don't want to mix that with some fragile, unimportant part.
Related
In Cassandra, Hinted Handoff(HH) only happens when consistency level can be met. Also, hints are unreadable to clients. With consistency level > ANY, using HH can improve neither write nor read availability. Requests still fail since the online replicas are insufficient to meet consistency requirements.
What's the point of using Hinted Handoff? Trading capacity for performance?
Why not just synchronize the failed-and-back node with other replica nodes (i.e. re-replication)?
Hinted handoff is just additional anti entropy measure. i.e. you don't have to run repairs right away and data gets consistent (if there was minor outage) when the node comes back online.
I guess it would just be too complicated to take take care of this with replication all the time because you would have to mark somehow the data that hasn't been replicated etc. Basically you would again have something similar to hinted handoff.
Some stuff from official docs:
https://docs.datastax.com/en/cassandra/2.1/cassandra/dml/dml_about_hh_c.html#concept_ds_ifg_jqx_zj__extreme-write-availability
Basically it's to maximise write throughput of the cluster while there are minor outages. It's configurable and you can disable it in case that you described when there are high consistency levels involved in both read and write.
Plus you have to run the "re-replication" i.e. repair anyway. Because hinted handoff can't really take care of it all.
Personally I used them in a situation R-CL: ONE, W-CL: ONE, RF: 2, NODES: 3. They were very helpful because we maintained the write throughput while doing maintenance and rolling restarts on the cluster. So I would say it works well in situation where W-CL < RF.
Then again there is opinions like this one:
https://blog.threatstack.com/scaling-cassandra-lessons-learned
In fact, just disable them in the configuration. It’s too easy to lose data during a prolonged outage or load spike, and if a node went down because of the load spike you’re just going to pass the problem around the ring, eventually taking multiple or all nodes down. We never experienced this on Cassandra, but have on other systems that supported hinted handoffs.
I want to scale an e-commerce portal based on LAMP. Recently we've seen huge traffic surge.
What would be steps (please mention in order) in scaling it:
Should I consider moving onto Amazon EC2 or similar? what could be potential problems in switching servers?
Do we need to redesign database? I read, Facebook switched to Cassandra from MySql. What kind of code changes are required if switched to Cassandra? Would Cassandra be better option than MySql?
Possibility of Hadoop, not even sure?
Any other things, which need to be thought of?
Found this post helpful. This blog has nice articles as well. What I want to know is list of steps I should consider in scaling this app.
First, I would suggest making sure every resource served by your server sets appropriate cache control headers. The goal is to make sure truly dynamic content gets served fresh every time and any stable or static content gets served from somebody else's cache as much as possible. Why deliver a product image to every AOL customer when you can deliver it to the first and let AOL deliver it to all the others?
If you currently run your webserver and dbms on the same box, you can look into moving the dbms onto a dedicated database server.
Once you have done the above, you need to start measuring the specifics. What resource will hit its capacity first?
For example, if the webserver is running at or near capacity while the database server sits mostly idle, it makes no sense to switch databases or to implement replication etc.
If the webserver sits mostly idle while the dbms chugs away constantly, it makes no sense to look into switching to a cluster of load-balanced webservers.
Take care of the simple things first.
If the dbms is the likely bottle-neck, make sure your database has the right indexes so that it gets fast access times during lookup and doesn't waste unnecessary time during updates. Make sure the dbms logs to a different physical medium from the tables themselves. Make sure the application isn't issuing any wasteful queries etc. Make sure you do not run any expensive analytical queries against your transactional database.
If the webserver is the likely bottle-neck, profile it to see where it spends most of its time and reduce the work by changing your application or implementing new caching strategies etc. Make sure you are not doing anything that will prevent you from moving from a single server to multiple servers with a load balancer.
If you have taken care of the above, you will be much better prepared for making the move to multiple webservers or database servers. You will be much better informed for deciding whether to scale your database with replication or to switch to a completely different data model etc.
1) First thing - measure how many requests per second can serve you most-visited pages. For well-written PHP sites on average hardware it must be in 200-400 requests per second range. If you are not there - you have to optimize the code by reducing number of database requests, caching rarely changed data in memcached/shared memory, using PHP accelerator. If you are at some 10-20 requests per second, you need to get rid of your bulky framework.
2) Second - if you are still on Apache2, you have to switch to lighthttpd or nginx+apache2. Personally, I like the second option.
3) Then you move all your static data to separate server or CDN. Make sure it is served with "expires" headers, at least 24 hours.
4) Only after all these things you might start thinking about going to EC2/Hadoop, build multiple servers and balancing the load (nginx would also help you there)
After steps 1-3 you should be able to serve some 10'000'000 hits per day easily.
If you need just 1.5-3 times more, I would go for single more powerfull server (8-16 cores, lots of RAM for caching & database).
With step 4 and multiple servers you are on your way to 0.1-1billion hits per day (but for significantly larger hardware & support expenses).
Find out where issues are happening (or are likely to happen if you don't have them now). Knowing what is your biggest resource usage is important when evaluating any solution. Stick to solutions that will give you the biggest improvement.
Consider:
- higher than needed bandwidth use x user is something you want to address regardless of moving to ec2. It will cost you money either way, so its worth a shot at looking at things like this: http://developer.yahoo.com/yslow/
- don't invest into changing databases if that's a non issue. Find out first if that's really the problem, and even if you are having issues with the database it might be a code issue i.e. hitting the database lots of times per request.
- unless we are talking about v. big numbers, you shouldn't have high cpu usage issues, if you do find out where they are happening / optimization is worth it where specific code has a high impact in your overall resource usage.
- after making sure the above is reasonable, you might get big improvements with caching. In bandwith (making sure browsers/proxy can play their part on caching), local resources usage (avoiding re-processing/re-retrieving the same info all the time).
I'm not saying you should go all out with the above, just enough to make sure you won't get the same issues elsewhere in v. few months. Also enough to find out where are your biggest gains, and if you will get enough value from any scaling options. This will also allow you to come back and ask questions about specific problems, and how these scaling options relate to those.
You should prepare by choosing a flexible framework and be sure things are going to change along the way. In some situations it's difficult to predict your user's behavior.
If you have seen an explosion of traffic recently, analyze what are the slowest pages.
You can move to cloud, but EC2 is not the best performing one. Again, be sure there's no other optimization you can do.
Database might be redesigned, but I doubt all of it. Again, see the problem points.
Both Hadoop and Cassandra are pretty nifty, but they might be overkill.
In a financial system I am currently maintaining, we are relying on the rollback mechanism of our database to simulate the results of running some large batch jobs - rolling back the transaction or committing it at the end, depending on whether we were doing a test run.
I really cannot decide what my opinion is. In a way I think this is great, because then there is no difference between the simulation and a live run - on the other hand it just feels kind of icky, like e.g. depending on code throwing exceptions to carry out your business logic.
What is your opinion on relying on database transactions to carry out your business logic?
EXAMPLE
Consider an administration system having 1000 mortgage deeds in a database. Now the user wants to run a batch job that creates the next term invoice on each deed, using a couple of advanced search criteria that decides which deeds are to be invoiced.
Before she actually does this, she does a test run (implemented by doing the actualy run but ending in a transaction rollback), creating a report on which deeds will be invoiced. If it looks satisfactory, she can choose to do the actual run, which will end in a transaction commit.
Each invoice will be stamped with a batch number, allowing us to revert the changes later if it is needed, so it's not "dangereous" to do the batch run. The users just feel that it's better UX to be able to simulate the results first.
CLARIFICATION
It is NOT about testing. We do have test and staging environments for that. It's about a regular user using our system wanting to simulate the results of a large operation, that may seem "uncontrollable" or "irreversible" even though it isn't.
CONCLUSION
Doesn't seem like anyone has any real good arguments against our solution. As always, context means everything, so in the context of complex functional requirements exceeding performance requirements, using db rollback to implement batch job simulations seems a viable solution.
As there is no real answer to this question, I am not choosing an answer - instead I upvoted those who actually put forth an argument.
I think it's an acceptable approach, as long as it doesn't interfere with regular processing.
The alternative would be to build a query that displays the consequences for review, but we all have had the experience of taking such an approach and not quite getting it right; or finding that the context changed between query and execution.
At the scale of 1000 rows, it's unlikely the system load is burdensome.
Before she actually does this, she does a test run (implemented by doing the actualy run but ending in a transaction rollback), creating a report on which deeds will be invoiced. If it looks satisfactory, she can choose to do the actual run, which will end in a transaction commit.
That's wrong, prone to failure, and must be hell on your database logs. Unless you wrap your simulation and the actual run in a single transaction (which, judging by the timeline necessary to inspect 1000 deeds, would lead to a lot of blocked users) then there's no guaranteed consistency between test run and real run. If somebody changed data, added rows, etc. then you could end up with a different result - defeating the entire purpose of the test run.
A better pattern to do this would be for the test run to tag the records, and the real run to pick up the tagged records and process them. Or, if you have a thick client app, you can pull down the records to the client, show the report, and - if approved - push them back up.
We can see what the user needs to do, quite a reasonable thing. I mean how often do we get a regexp right first time? Refining a query till it does exactly what you want is not unusual.
The business consequences of not catching errors may be quite high, so doing a trial run makes sense.
Given a blank sheet of paper I'm sure we can devise an clean implementation expressed in formal behaviours of the system rather than this somewhat back-door appraoch.
How much effort would I put into fixing that now? Depends on whether the current approach is actually hurting. We can imagine that in a heaviliy used system it could lead to contention in the database.
What I wrote about the PRO FORMA environment in that bank I worked in was also entirely a user thing.
I'm not sure exactly what you're asking here. Taking you literally
What is your opinion on relying on
database transactions to carry out
your business logic?
Well, that's why we have transactions. We do rely on them. We hit an error and abort a transaction and rely on work done in that transaction scope to be rolled-back. So exploiting the transactional beahviours of our systems is a jolly good thing, and we'd need to hand-roll the same thing ourselves if we didn't.
But I think your question is about testing in a live system and relying on not commiting in order to do no damage. In an ideal world we have a live system and a test system and we don't mess with live systems. Such ideals are rarely seen. Far more common is "patch the live system. testing? what do you mean testing?" So in fact you're ahead of the game compared with some.
An alternative is to have dummy data in the live system, so that some actions can actually run all the way through. Again, error prone.
A surprisingly high proportion of systems outage are due to finger trouble, it's the humans who foul up.
It works - as you say. I'd worry about the concurrency of the system since the transaction will hold locks, possibly many locks. It means that your tests will hold up any live action on the system (and any live action operations will hold up your tests). It is generally better to use a test system for testing. I don't like it much, but if the risks from running a test but forgetting to roll it back are not a problem, and neither is the interference, then it is an effective way of attaining a 'what if' type calculation. But still not very clean.
When I was working in a company that was part of the "financial system", there was a project team that had decided to use the production environment to test their conversion procedure (and just rollback instead of commit at the end).
They almost got shot for it. With afterthought, it's a pity they weren't.
Your test environments that you claim you have are for the IT people's use. Get a similar "PRO-FORMA" environment of which you can guarantee your users that it is for THEIR use exclusively.
When I worked in that bank, creating such a PRO-FORMA environment was standard procedure at every year closure.
"But it's not about testing, it's about a user wanting to simulate the results of a large job."
Paraphrased : "it's not about testing, it's about simulation".
Sometimes I wish Fabian Pascal was still in business.
(Oh, and in case anyone doesn't understand : both are about "not perpetuating the results".)
Let's say at your job your boss says,
That system over there, which has lost all institutional knowledge but seems to run pretty good right now, could we dump double the data in it and survive?
You're completely unfamiliar with the system.
It's in SQL Server 2000 (primarily a database app).
There's no test environment.
You might be able to hijack it on the weekends if you needed to run a benchmark.
What would be the 3 things you'd do to convince yourself and then your manager that you could take on that extra load. And if you couldn't do it, on the same hardware... the extra hardware (measured in dollars) it would take to satisfy that request.
To address the response from doofledorfer, you assumptions are almost all 180 degrees off. But that's my fault for an ambiguous question.
One of the main servers runs 7x24 at 70% base and spikes from there and no one knows what it is doing.
This isn't an issue of buy-in or whining... Our company may not have much of a choice in the matter.
Because this is being externally mandated, delays in implementation could result in huge fines. So large meeting to assess risk are almost impossible. There is one risk, that dumping double the data would take the system down for the existing customers.
I was hoping someone would say something like, see if you take the system off line Sunday night at midnight and run SQLIO tests to see how close the storage subsystem is to saturation. Things like that.
Set up a test environment, even if I have to do it on my laptop.
Enable some kind of logging on the production system to get an idea of the volume of transactions in addition to the volume of data.
Read the source code as I run stress tests on my laptop with increasing amounts of data.
Having said that, I sympathize with this assignment, because it's unfair. It's like asking someone in a boat if the boat can float with twice the cargo -- but you can't get out of the boat or take it out of its regular service.
You've just described a typical Agile project. Your answer should be:
I don't know, and I won't be able to tell without testing.
In addition to data volume, there might be issues with usage patterns, application interactions, database and server tuning, etc.
So let's work through a basic list of risk factors, and how we might resolve them.
Once we've done that, let's work through them in inverse order of risk; and make a stop/continue decision as we develop the results.
etc.
Without management buy-in and participation at least at that level, any other answer you might give is high-risk wishing, and "3 most important" is a non sequitur.
I'd be optimistic unless your current system is substantially loaded already. Most servers should run at less than 50% capacity on all resources, or else be on life-support.
And I expect you wouldn't be having the conversation if the existing server were already dealing with load issues; although "seems to run pretty good right now" is imprecise enough to be worrisome.
it mostly depends on its current level. If doubling is going from 2GB to 4GB just do it. If it's going from 1TB to 2TB you've got some planning to do.
I'd collect some info using Performance Monitor and provide it to help make an educated decision.
It depends what you mean by "double the data".
If that is going to affect one table only (say product table) then you are probably safe as most queries that are referring to that one table are most likely to double the time of execution (that assumes that you do not reference the same time twice in a query).
The problem will arise if you double the amount of data in all the tables as the execution time may grow in exponential fashion then and it can lead to some serious issues.
But in general I would support the answer by doofledorfer
I have developed a framework that is used by several teams in our organisation. Those "modules", developed on top of this framework, can behave quite differently but they are all pretty resources consuming even though some are more than others. They all receive data in input, analyse and/or transform it, and send it further.
We planned to buy new hardware and my boss asked me to define and implement a benchmark based on the modules in order to compare the different offers we have got.
My idea is to simply start sequentially each module with a well chosen bunch of data as input.
Do you have any advice? Any remarks on this simple procedure?
Your question is pretty broad, so unfortunately my answer will not be very specific either.
First, benchmarking is hard. Do not underestimate the effort necessary to produce meaningful, repeatable, high-confidence results.
Second, what is your performance goal? Is it throughput (transaction or operations per second)? Is it latency (time it takes to execute a transaction)? Do you care about average performance? Do I care about worst case performance? Do you care about the absolute worst case or I care that 90%, 95% or some other percentile get adequate performance?
Depending on which goal you have, then you should design your benchmark to measure against that goal. So, if you are interested in throughput, you probably want to send messages / transactions / input into your system at a prescribed rate and see if the system is keeping up.
If you are interested in latency, you would send messages / transactions / input and measure how long it takes to process each one.
If you are interested in worst case performance you will add load to the system until up to whatever you consider "realistic" (or whatever the system design says it should support.)
Second, you do not say if these modules are going to be CPU bound, I/O bound, if they can take advantage of multiple CPUs/cores, etc. As you are trying to evaluate different hardware solutions you may find that your application benefits more from a great I/O subsystem vs. a huge number of CPUs.
Third, the best benchmark (and the hardest) is to put realistic load into the system. Meaning, you record data from a production environment, and put the new hardware solution through this data. Getting this done is harder than it sounds, often, this means adding all kinds of measure points in the system to see how it behaves (if you do not have them already,) modifying the existing system to add record/playback capabilities, modifying the playback to run at different rates, and getting a realistic (i.e., similar to production) environment for testing.
The most meaningful benchmark is to measure how your code performs under everyday usage. That will obviously provide you with the most realistic numbers.
Choose several real-life data sets and put them through the same processes your org uses every day. For extra credit, talk with the people that use your framework and ask them to provide some "best-case", "normal", and "worst-case" data. Anonymize the data if there are privacy concerns, but try not to change anything that could affect performance.
Remember that you are benchmarking and comparing two sets of hardware, not your framework. Treat all of the software as a black box and simply measure the hardware performance.
Lastly, consider saving the data sets and using them to similarly evaluate any later changes you make to the software.
If you're system is supposed to be able to handle multiple clients all calling at the same time, then your benchmark should reflect this. Note that some calls will not play well together. For example, having 25 threads post the same bit of information at the same time could lead to locks on the server end, thus skewing your results.
From a nuts-and-bolts point of view, I've used Perl and its Benchmark module to gather the information I care about.
If you're comparing differing hardware, then measuring the cost per transaction will give you a good comparison of the trade offs of hardware for performance. One configuration may give you the best performance, but costs too much. A less expensive configuration may give you adequate performance.
It's important to emulate the "worst case" or "peak hour" of load. It's also important to test with "typical" volumes. It's a balancing act to get good server utilization, that doesn't cost too much, that gives the required performance.
Testing across hardware configurations quickly becomes expensive. Another viable option is to first measure on the configuration you have, then simulate that behavior across virtual systems using a model.
If you can, try to record some operations users (or processes) are doing with your framework, ideally using a clone of the real system. That gives you the most realistic data. Things to consider:
Which functions are most often used?
How much data is transferred?
Do not assume anything. If you think "that is going to be fast/slow", don't bet on it. In 9 out of 10 cases, you're wrong.
Create a top ten for 1+2 and work from that.
That said: If you replace old hardware with new hardware, you can expect roughly 10% faster execution for each year that has passed since you bought the first set (if the systems are otherwise pretty equal).
If you have a specialized system, the numbers may be completely different but usually, new hardware doesn't change much. For example, adding an useful index to a database can reduce the runtime of a query from two hours to two seconds. Hardware will never give you that.
As I see it, there are two kinds of benchmarks when it comes to benchmarking software. First, microbenchmarks, when you try to evaluate a piece of code in isolation or how a system deals with narrowly defined workload. Compare two sorting algorithms written in Java. Compare two web browsers how fast can each perform some DOM manipulation operation. Second, there are system benchmarks (I just made the name up), when you try to evaluate a software system under a realistic workload. Compare my Python based backend running on Google Compute Engine and on Amazon AWS.
When dealing with Java and such like, keep in mind that the VM needs to warm up before it can give you realistic performance. If you measure time with the time command, the JVM startup time will be included. You almost always want to either ignore start-up time or keep track of it separately.
Microbenchmarking
During the first run, CPU caches are getting filled with the necessary data. The same goes for disk caches. During few subsequent runs the VM continues to warm up, meaning JIT compiles what it deems helpful to compile. You want to ignore these runs and start measuring afterwards.
Make a lot of measurements and compute some statistics. Mean, median, standard deviation, plot a chart. Look at it and see how much it changes. Things that can influence the result include GC pauses in the VM, frequency scaling on the CPU, some other process may start some background task (like virus scan), OS may decide move the process on a different CPU core, if you have NUMA architecture, the results would be even more marked.
In case of microbenchmarks, all of this is a problem. Kill what processes you can before you begin. Use a benchmarking library that can do some of it for you. Like https://github.com/google/caliper and such like.
System benchmarking
In case of benchmarking a system under a realistic workload, these details do not really interest you and your problem is "only" to know what a realistic workload is, how to generate it and what data to collect. It is always best if you can instrument a production system and collect data there. You can usually do that, because you are measuring end-user characteristics (how long did a web page render) and these are I/O bound so the code gathering data does not slow down the system. (The page needs to be shipped to the user over the network, it does not matter if we also log a few numbers in the process).
Be mindful of the difference between profiling and benchmarking. Benchmarking can give you absolute time spent doing something, profiling gives you relative time spent doing something compared to everything else that needed doing. This is because profilers run heavily instrumented programs (common technique is to stop-the-world every few hundred ms and save a stack trace) and the instrumentation slows everything down significantly.