Company Overview
Tennishub.in is the top-rated tennis ecommerce website in India, since 2013. It is a go-to website for professionals and amateurs alike, carrying brands like Wilson and Yonex. When you buy a racquet here, you get it strung with the tension you want, the grip and a logo.
The site was hosted on AWS and was self / agency managed.
Key Challenges:
The agency, our partners Techsevin approached us with a problem – the site used to frequently go down with a cloudflare 524. A reboot of the application server was required to get the site back up. They had tried blocking other countries – as their primary business was in India but had run into issues with google crawlers also getting blocked.
Our Approach
As part of our proposal, we analyzed their nginx logs and found surge in traffic for short times when the site would go down. We were confident of resolving the problem due to our better analysis and tracking of BOTs as well as superior BOT detection capability, we would help reduce such DOS attacks, as we have done for many websites.
Win 1: Blocking user agents (BOTs)
After onboarding tennishub.in on luroConnect stack installed on their AWS account, we started observing and blocking BOTs, after discussing with the customer.
That brought down hits for these BOTs, but we yet faced the issue of downtimes – though less frequent than before.
Win 2: Out of Memory
We observed some outages coincided with Out of Memory alerts on the server. However, some others did not. Upon restricting memory with php-fpm, we were able to get hits to appear in the php-fpm access and slow logs. However, now the servers ran out of CPU.
CPU and Memory usage of the App Server
June 19 site went down. June 20 : fix to kill long running php. June 21 code fix
Having a horizontally scalable architecture, we were able to immediately add servers to ensure the website did not go down. This while we and the developers debugged the problem.
Win 3: Traced php-fpm slow logs
With using php-fpm slow log trace and our ability to uniquely trace a request from nginx to php access log, we were able to trace a few problem-hits – pointing to an infinite loop. We also observed that these hits came together, causing the OOM we saw earlier. Along with the developers we were able to determine that some users had content in the cart that caused this issue. We were able to determine that once a user gets this problem, they are unable to use the website.
luroConnect was able to enable enough logs to isolate the carts that caused this problem and provide information to the development team.
Hosting with luroConnect feels different. They can give us information to help debug such issues that no hosting provider does. For debugging this problem, we may have to change code to write logs to understand the problem. Even then, we may not have come to these observations as quickly as luroConnect.
Mehul Goswami, Co-Founder, Techsevin
