Repository navigation
Question about service-to-service communication patterns in Online Boutique #3515
Replies: 1 comment 7 replies
|
Yes, an early plateau is consistent with the stock workload when you're counting service pairs, rather than pod/IP pairs or individual RPC methods. The load generator repeatedly chooses from a fixed set of actions: home page, currency selection, product browsing, adding to cart, viewing the cart, and checkout. Products, quantities and customer details vary, but they don't continually introduce new service types or a growing set of destinations. Checkout is sampled less often than product browsing, so its downstream paths can take longer to appear. For your experiment I'd distinguish two claims:
A useful cross-check is to compare the observed edges against the application's architecture diagram and service descriptions, then exercise checkout explicitly and see whether it adds any missing edges. For reproducibility, pin the application/load-generator revision and record the user count, run duration and any deployment options. Also keep DNS, health checks and telemetry traffic separate from business-service edges, and aggregate replicas consistently using workload/service identity. Otherwise pod churn or infrastructure traffic can look like new application relationships. Running the same fixed workload longer primarily increases confidence in the observed graph; it doesn't guarantee additional pairs will appear. |
Uh oh!
There was an error while loading. Please reload this page.
Hello,
I am a master's student conducting research on Kubernetes network security using the Online Boutique application and Cilium Hubble.
I am collecting network flow logs generated by the built-in load generator and analyzing the service-to-service communication patterns.
In my experiment, I collected the network flows for one hour and divided the observation period into one-minute segments. When I cumulatively analyzed the segments, I found that the number of distinct service-to-service communication pairs stopped increasing relatively early and remained stable for the rest of the observation period.
I would like to ask whether this is expected behavior for the Online Boutique application and its load generator.
Are the main service-to-service communication paths designed to be exercised repeatedly within a relatively short period of time? Or would you normally expect new communication pairs to appear over a longer observation period?
Thank you for your help.
All reactions