In November last year, I started developing an infrastructure that would allow us to
collect, store, search and retrieve high volume data. The idea was
to collect all the URLs on which our homegrown CDN
would serve JS content. Based on our current traffic, we were looking to collect some 10k URLs per
second across four major geographic regions where we run our servers.
In the beginning we tried MySQL, Redis, Riak, CouchDB, MongoDB, ElasticSearch but
nothing worked out for us with that kind of high speed writes. We also wanted our
system to respond very quickly, under 40ms between
internal servers on private network. This post talks about how we were able to
make such a system using C++11, RocksDB and Thrift.
First, let me start by sharing the use cases of such a system in VWO; the
following screenshot shows a feature where users can enter a URL to check if VWO
Smart Code was installed on it.
VWO Smart Code checker
The following screenshot shows another feature where users can see a list of URLs
matching a complex wildcard pattern, regex pattern, string rule etc. while
creating a campaign.
VWO URL Matching Helper
several opensource databases but none of them would fit our requirements except
Cassandra. In clustered deployment, reads from Cassandra were too slow and slower
when data size would grew. After understanding how Cassandra worked under the
hood such as its log structured storage like LevelDB I started playing with opensource
embeddable databases that would use similar approach such as LevelDB and Kyoto Cabinet.
At the time, I found an embeddable persistent key-value store
library built on LevelDB called RocksDB.
It was opensourced by Facebook and had a fairly active developer community so I
with it. I read the project wiki,
wrote some working code and joined their Facebook group to ask questions around
prefix lookup. The community was helpful, especially Igor and
Siying who gave me enough hints
around prefix lookup,
using custom extractors
and bloom filters which helped me
write something that actually worked in our production environment for the first time.
Explaining the technology and jargons is out of scope of this post but I would like
to encourage the readers to read
aboutLevelDB and to read the RocksDB wiki.
RocksDB FB Group
For capturing the URLs with peak velocity up to 10k serves/s, I reused our
distributed queue based infrastructure.
For storage, search and retrieval of URLs I wrote a custom datastore service
using C++, RocksDB and Thrift called HarvestDB. Thrift
provided the RPC mechanism
for implementing this system as a distributed service accessible by various
backend sub-systems. The backend sub-systems use client libraries
generated by Thrift compiler for communicating
with the HarvestDB server.
The HarvestDB service implements five remote procedures - ping, get,
put, search and purge. The following Thrift IDL
describes this service.
Clients use ping to check HarvestDB server connectivity before executing
other procedures. RabbitMQ consumers consume collected URLs and put them to
HarvestDB. The PHP based application backend uses custom Thrift based client
library to get (read) and to search URLs.
A Python program runs as a periodic cron job and uses purge procedure to purge old entries
based on timestamp which makes sure we don’t exhaust our storage
resources. The system is in production for more than five months now and is
capable of handling (benchmarked) workload of up to 24k writes/second while consuming
less than 500MB RAM. Our future work will be on replication, sharding and fault
tolerance of this service. The following diagram illustrates this architecture.
e2e or end-to-end or UI testing is a methodology used to test whether the flow of an application is performing as designed from start to finish. In simple words, it is testing of your application from the user endpoint where the whole system is a blackbox with only the UI exposed to the user.
It can become quite an overhead if done manually and if your application has a large number of interactions/pages to test.
In the rest of the article I’ll talk about webdriverJS and Jasmine to automate your e2e testing, a combination which isn’t talked about much on the web.
What is WebDriverJS?
This was something which took me quite sometime to put my head around and I feel this was more or less due to the various available options for everything related to WebDriver.
So let’s take it from the top and see what its all about.
This having support for almost all major browsers, is a very good alternative to automate our tests in the browser.
So whatever you do in the browser while testing your application, like navigating to pages, clicking a button, writing text in input boxes, submitting forms etc, can be automated using Selenium.
WebDriver (or Selenium 2) basically refers to the language bindings and the implementations of the individual browser controlling code.
WebDriver introduces a JSON wire protocol for various language bindings to communicate with the browser controller.
For example, to click an element in the browser, the binding will send POST request on /session/:sessionId/element/:id/click
So, at one end there is the language binding and a server, known as Selenium server, on the other. Both communicate using the JSON wire protocol.
So as you might guess, WebDriverJS is simply a wrapper over the JSON wire protocol exposing high level functions to make our life easy.
Now if you search webdriver JS on the web, you’ll come across 2 different bindings namely selenium-webdriver and webdriverjs (yeah, lots of driver), both available as node modules. You can use anyone you like, though we’ll stick to the official one i.e. selenium-webdriver.
Done! You can now require the package and with a lil’ configuration you can open any webpage in the browser:
To run your test file, all you do is:
Note: In addition to the npm package, you will need to download the WebDriver implementations you wish to utilize. As of 2.34.0, selenium-webdriver natively supports the ChromeDriver. Simply download a copy and make sure it can be found on your PATH. The other drivers (e.g. Firefox, Internet Explorer, and Safari), still require the standalone Selenium server.
Difference from other language bindings
WebDriverJS has an important difference from other bindings in any other language - It is asynchronous.
So if you had done the following in python:
But it doesn’t stop here. Even with promises, the above code would have become:
Do you smell callback hell in there? To make it more neat, WebDriverJS has a wrapper for Promise called as ControlFlow.
In simple words, this is how ControlFlow prevents callback hell:
It maintains a list of schedule actions.
The exposed functions in WebDriverJS do not actually do their stuff, instead they just push the required action into the above mentioned list.
ControlFlow puts every new entry in the then callback of the last entry of the list, thus ensuring the sequence between them.
And so, it enables us to simply do:
Isn’t that awesome!
Controlflow also provides an execute function to push your custom function inside the execution list and the return value of that function is used to resolve/reject that particular execution. So you can use promises and do any asynchronous thing in your custom code:
Combining WebDriverJS with Jasmine
Our browser automation is setup with selenium. Now we need a testing framework to handle our tests. That is where Jasmine comes in.
If we were to convert our earlier testfile.js to check for correct page title, here is what it might look like:
Now the above file needs to be run with jasmine-node, like so:
This will fire the browser and do the mentioned operations, but you’ll notice that Jasmine won’t give any results for the test. Why?
Well…that happens because Jasmine has finished executing and no expect statement ever executed because of the expectation being inside an asynchronous callback of getTitle function.
To solve such asynchronicity in our tests, jasmine-node provides a way to tell that a particular it block is asynchronous. It is done by accepting a done callback in the specification (it function) which makes Jasmine wait for the done() to be executed. So here is how we fix the above code:
Quick tip: You might want to tweak the time allowed for tests to complete in Jasmine like so:
Bonus for Angular apps
Angular framework has been very testing focused since the very beginning. Needless to say, they have devoted a lot of time on e2e testing as well.
Protractor is a library by the Angular team which is a wrapper on WebDriverJS and Jasmine and is specifically tailored to make testing of Angular apps a breeze.
Checkout some of the neat addons it gives you:
Apart from querying element based on id, css selector, xpath etc, it lets you query on basis of binding, model, repeater etc. Sweet!
It has Jasmine’s expect function patched to accept promises. So, for example, in our previous test where we were checking for title:
can be refactored to a much cleaner:
And more such cool stuff to make end-to-end testing for Angular apps super-easy.
In the end
e2e testing is important for the apps being written today and hence it becomes important for it to be automated and at the same time fun and easy to perform. There are numerous tools available for you to choose and this article talks about one such tool combination.
Hope this helps you get started. So what are you waiting for, lets write some end-to-end tests!
Using an e2e testing stack you want to share? Let us know in the comments.
Our home-grown geo-distributed architecture
latencies possible. Using the same architecture we do data acquisition as well.
Over the years we’ve done a lot of changes to our backend, this post talks
about some scaling and reliability aspects and our recent work on making fast and
reliable data acquisition system using message queues which is in production for
about three months now. I’ll start by giving some background on our previous
Web beacons are widely used to do data
acquisition, the idea is to have a webpage send us data using an HTTP request
and the server sends some valid object. There are many ways to do this. To keep
the size of the returned object small, for every HTTP request we
return a tiny 1x1 pixel gif image and our geo-distributed architecture along with
our managed Anycast DNS service helps us to do this with very low latencies,
we aim for less than 40ms. When an HTTP request hits one of our data acquisition servers, OpenResty
handles it and our Lua based code processes the request in the same process thread.
OpenResty is a nginx mod which among many things bundles luajit that allows
us to write URL handlers in Lua and the code runs within the web server. Our Lua code
does some quick checks, transformations and writes the data to a Redis
server which is used as fast in-memory data sink. The data stored in Redis is
later moved, processed and stored in our database servers.
This was the architecture when I had joined
Wingify couple of months ago. Things were going smooth but the problem was we were
not quite sure about data accuracy and scalability. We used Redis as a fast
in-memory data storage sink, which our custom written PHP based queue infrastructure
would read from, our backend would process it and write to our database servers.
The PHP code was not scalable and after about a week of hacking, exploring options
we found few bottlenecks and decided to re-do the backend queue infrastructure.
We explored many options and decided to use RabbitMQ.
We wrote a few proof-of-concept backend programs in Go, Python and PHP and
did a lot of testing, benchmarking and real-world load testing.
Ankit, Sparsh and I discussed how we should move forward and we finally
decided to explore two models in which we would replace the home-grown PHP queue
system with RabbitMQ. In the first model, we wrote directly to RabbitMQ from the
Lua code. In the second model, we wrote a transport agent which moved data from Redis
to RabbitMQ. And we wrote RabbitMQ consumers in both cases.
There was no Lua-resty library for RabbitMQ, so I wrote one using cosocket APIs
which could publish messages to a RabbitMQ broker over STOMP protocol. The library
opensourced for the hacker community.
Later, I rewrote our Lua handler code using this library and ran a loader.io
load test. It failed this model due to very low throughput,
we performed a load test on a small 1G DigitalOcean instance for both models.
For us, the STOMP protocol
and slow RabbitMQ STOMP adapter were performance bottlenecks. RabbitMQ was not
as fast as Redis, so we decided to keep it and work on the second
model. For our requirements, we wrote a proof-of-concept Redis to RabbitMQ transport
agent called agentredrabbit to leverage Redis as a fast in-memory storage sink and
use RabbitMQ as a reliable broker. The POC worked well in terms of performance,
throughput, scalability and failover. In next few weeks we were able to write a
production level queue based pipeline for our data acquisition system.
For about a month, we ran the new pipeline in production against the existing one,
to A/B test our backend :) To do that we modified our Lua code to write to two
different Redis lists, the original list was consumed by the existing pipeline, the other was
consumed by the new RabbitMQ based pipeline. The consumer would process and write
data to a new database. This allowed us to compare realtime data from the two
pipelines. During this period we tweaked our implementation a lot, rewrote the
producers and consumers thrice and had two major phases of refactoring.
A/B testing of existing and new architecture
Based on results against a 1G DigitalOcean instance like
for the first model and against the A/B comparison of existing pipeline in realtime,
we migrated to the new pipeline based on RabbitMQ. Other issues of HA,
redundancy and failover were addressed in this migration as well.
The new architecture ensures no single point of failure and has mechanisms to
recover from failure and fault.
Queue (RabbitMQ) based architecture in production
We’ve opensourced agentredrabbit
which can be used as a general purpose fast and reliable transport agent for
moving data in chunks from Redis lists to RabbitMQ with some assumptions and queue
name conventions. The flow diagram below has hints on how it works, checkout the
README for details.
When I got an opportunity of interning with the engineering team at Wingify it made me ecstatic because of an exciting office with fascinating transparent walls full of geeky stuff, I came across on my first visit for an interview — and of course Wingify is a becoming a buzz word in IT industry.
On my first day I was a bit nervous, dressed and prepared as I believed anyone working from 10:00 am to 7:00 pm would. When I reached the office only the office boy was present — honestly speaking I had a feeling that I am at a wrong place because there was no way a software company should look like at 10:30 in the morning on a working day. After a while I was surrounded by people in shorts, denims, t-shirts with smiling faces having friendly chats.
Working at Wingify provided me with an entirely new set of skills like software development design patterns and maintenance that is going to be invaluable for my future. My work here mainly included front-end optimization and internationalization.
Worked on template based engine for the translation of web pages in different languages.
I worked along with the marketing team, and this added an additional dimension to my work by interdependent relationship. I also spent my time researching and learning different methods and technologies for various things such as process automation. All these roles and responsibilities taught me to manage time, being attentive and organized, and enhanced problem-solving abilities.
At Wingify you have the solidarity and independence of your own space and an atmosphere where interns like myself would not hesitate to ask questions as they are answered and explained by highly skilled and dedicated engineering team sitting next to you, which makes it easy to get work done. Awesome appreciation mails boost you up with confidence. Personally, I couldn’t have imagined a better internship experience.
Interning with Wingify provides you with a wonderful learning experience. In a nutshell, it is a great place to work and party \m/
This post is about making your web page perform better using a
real world example. As you know, we recently launched a very cool animated
comic on A/B Testing.
It is scroll animation describing what is A/B testing. I’ll talk about
it as an example and walk you through its performance issues, how we
debugged them and finally what we did to extract 60 FPS out of it.
The process we see in following text will applies more or less to all web pages in
general. Here’s what you need to get started:
Determination to make it run as smooth as a hot knife through butter :)
Worry not if you are missing any of the above, you can still read on. Let us begin.
WHAT is causing the issue?
All we know now is that our page is janky. When you scroll up/down
you’ll notice that the animation is quite choppy. There are sudden jumps
occasionally while scrolling which is really irritating and obviously a bad user
experience. We don’t know what is causing this. The very first step we take here
is profile the page using Chrome devtool’s
feature. So I went on and fired up my devtools.
Open the devtools
Devtools in chrome can be fired either going to Tools > Developer Tools or
using the shortcut Ctrl + Shift + I on Windows/Linux and Cmd + Opt + I
Select frames tab
Frames tab basically will let us visualize each frame individually showing how much time was taken by that frame and for what tasks.
Filter out events taking more than 15ms
Note that we are targeting 60 FPS here. A little math here gives us the number
16.666 ms (1 / 60 * 1000). This is the time budget available per frame to
do its thing if we want a consistent 60 FPS.
Therefore, we essentially want to investigate those frames which are crossing
this time limit. To do so, select the >= 15ms option from bottom bar as
Press the ‘Record’ button at the bottom to start devtools record what’s happening
on the page. Once you do that, go back to the page and interact with the page
as one would normally do exposing the issues we are trying to debug.
In my case, the page was feeling choppy while scrolling between slides. So I
simply kept scrolling on the page like a normal user. After interacting for a
while with the page, I get back to the devtools window and press the same button
to stop the recording.
Notice the frames
You now see the frame data for your page… something like in the snapshot
above. In the image you’ll notice a vertical limit with the label 60 FPS just
below the label for 30 FPS. These limits are for the frames under which they
need to do their stuff if the respective framerate is to be achieved. Once you
know this, you’ll straight away conclude that almost all of our frames our
crossing that limit like hell! This is the point where we have actually
visualized and confirmed the issue. Lets find out the cause.
Script events taking more than 15ms
Every frame’s bar is made of different colour components. In the above
snapshots we see only yellow and green ones. A quick look at the color legend
in the bottom bar tells us that yellow is script time and green is painting. A
closer analysis tells us that most frames are in majority made up of yellow
component. This means that most of the frame’s time is spent in executing
Moreover if you hover over any small horizontal yellow bars below, as show in
the snapshot above, you’ll also see the exact time that our scripts are taking
per frame along with the corresponding event that triggered it. In my case, it’s
the scroll event (we expected that…no?). Some of those scroll events are
taking upto 27 ms which is much much more than our budget of 16ms per frame.
Issue detected: Scroll event script
After all this analysis using the devtools we hence come to the conclusion that
it’s the script executing for every scroll event that is the cause of issue
here. Next step in our debug process is finding WHY it is causing it.
WHY is it causing an issue?
Let’s investigate the code
Our code for the callback bound to the Scroll event is as follows:
This callback function will be our target from now on.
Scroll event is too frequent to handle scripts taking time
First thing that striked me was that the Scroll event is fired too frequently.
Every time you scroll on a page, that event is fired multiple times within
seconds. Therefore any code that is attached to the Scroll event will be fired
with the same frequency. And if that code is computation heavy, we are done!
To improve the situation here, we have 2 ways:
A. Make the Scroll event fire less frequently
B. Optimize the callback’s code to take less execution time
FIX A. Make the Scroll event fire less frequently
I could make the Scroll code fire less frequently in our case as it did not had
any usability hit. In fact mostly the code thats required to be executed on
Scroll event can be run on little longer intervals without any user experience
This thing was easy to do. Ben Alman has an awesome
written for throttling/debouncing functions. Its very easy to use too. Simply
get the plugin into your page and pass the throttled function to Scroll event
As you see in above code, I have made my callback to fire atmost once within
350 ms. In other words, there will be atleast an interval of 350 ms between
2 calls to that function. This should probably keep those adjacent long yellow
bar at some distant from each. We’ll see.
We made a small change from our side. But remember, there is no point of it
without actually testing the page and getting a performance boost. So lets
repeat the profiling procedure again.
Here is what we got this time:
Seems to have worked quite a bit! We have lesser frames overshooting the 16ms budget.
FIX B. Optimize the callback’s code to take less execution time
Secondly, its also important to optimize the code inside that callback at that is what is causing the frames to go beyond our 16ms budget.
If you look closely inside the callback’s code and have a basic understanding of what not do while jQuery, you’ll see some horrible things happening there. I’ll not go in much details on why those things are bad as our focus is on using devtools in this article. Lets list out what all jQuery menace we see in it:
Cache jQuery objects
At many places, jQuery is being used to reference element by passing their selectors again and again in the callback. That is BAD. Unless these references will change in future, its wise to calculate them once and cache for future use.
Some of the lines where jQuery is being used unnecessarily:
Have a look at the following code snippet:
The first if checks if we are on the last iteration of the loop or not. If not, then the else part executes. Which means if the loop runs 100 times, 99 times the else part executes. Moreover if you see carefully the code in the else block, it will keep fading in/out certain elements on each iteration, even when it has done the same thing in past iteration. Taking account the heavy animation account cost in jQuery, this is absolutely unnecessary work being done here.
We could simply do that stuff once and set a flag which will be checked next time and we only do it again if the flag is unset somehow.
After the above 2 fixes, here is how our Scroll event callback looks like:
Needless to say, our next step is to test the changes made. Here is what the
timeline says now:
We hardly have any frames overshooting the target line of 60 FPS.
We get an average execute time of 11.71 ms per frame with a standard deviation of around 4.97 ms.
We still see paint (green) events which are causing some frames to overshoot
the border. It is basically on slides where large image are being animated on
the screen. Its not that we can scale down the images or stop them from being
painted. The solution still needs to be figured out to optimize the painting
going on here. Suggestions?