Huge throughput drop when Kong is used to proxy requests to an API

[mrak...@gmail.com]

Hello,

I am experiencing a huge throughput drop when I use Kong to proxy requests to a backend API.

What I basically have is a simple echo service. I use Tsung client to produce load. When I point the clients directly to the echo service the QPS is 170,000. The setup looks like this:

Clients -> Echo service

After that I installed Kong and configured it to proxy the requests to the echo service. This setup looks like this:

Clients -> Kong -> Echo service

The QPS with this setup dropped to 24,000 - 25,000.

I did not do any changes to the Kong Nginx default configuration. I also did not enable any Kong plugins for my backend API.

My questions are:

1. Is it expected Kong to affect the QPS this much when used to proxy requests to the backend API?

2. If not then what the problem might be and what the potential solution might be?

Any help will be much appreciated.

Thank you,

Dimitar Dimitrov

[mrak...@gmail.com]

P.S.  After some further investigation it seems to me that Kong does not keep the connections to the API upstream alive. This might be the problem for the performance drop I have described above.

How can I configure Kong to keep these connections alive?

BTW: I am using Kong 0.7.0

Thank you,

Dimitar Dimitrov

[Thibault Charbonnier]

Hi,

> How can I configure Kong to keep these connections alive?

Kong does not yet support upstream connection pools. It's on our
short-term road-map, probably for 0.9.0 (no date yet).

Many other improvements are to be introduced in 0.9, especially
regarding performance.

[mrak...@gmail.com]

Hi again,

Thank you very much for the response.

Regards,

Dimitar Dimitrov