What is lod in `carts_lod`?

[phamquiluan]

Thank you for your work, @azamikram! Please let me know what is lod in the sock-shop data. I can't find anything about this in the paper and in your code

[phamquiluan]

Is it container_network_receive_packets_total as workload ?
or container_processes as CPU load?

[azamikram]

The lod is the number of requests received between two time intervals. We used the following Prometheus query to extract it sum(rate(request_duration_seconds_count [{DURATION}])) by (name) where DURATION is the length of the interval.

[phamquiluan]

@azamikram Thank you so much for your reply! I am wondering if you can give me the full Prometheus queries that you use to extract sock shop data, including the duration and other configurations?

[azamikram]

You can find that script in sock-shop-data now.

[phamquiluan]

Thank you so much for your support @azamikram! I see that you don't put the err metrics into the sock-shop data, may I know why?

[azamikram]

I cannot recall why I decided not to collect data for err. One thing that comes to mind is that err was only available for two services (front-end and catalogue) but I'm not sure if that was the reason.

[phamquiluan]

Thank you so much for your answer @azamikram , it really helps me!

I plotted your data and have another question: In this figure, the fault is "payment-mem", the memory usage of the payment container is increasing, I can understand that. But
Q. Why did the memory usage of other containers decrease or increase so suddenly?

I can't figure it out why, please help me 😄

[azamikram]

Failure propagation chain! Change in one service affects how other parts of the system behave.

[phamquiluan]

Thank you so much for your answer @azamikram! Could you please do another favour by public the stress-ng command that you used to inject the fault into the sock-shop? I also use stress-ng but I'm incapable of reproducing the failure propagation chain like you said 😄