Site can't be reached undeterministically

[eclGen]

I am using mahimahi to record and replay various webpages repeatedly. However, there seems to be an undeterministical glitch somewhere in my set up that only manifests when using mahimahi.

Set up:
Selenium grid (3.11) with Java bindings and browsers (firefox, opera, chrome) in the record or replay shell respectively on an Ubuntu 16.04 server machine.

What I do:
I record a webpage, e.g. google.com, once and replay it multiple times from this recording.

What I expect:
Every replay should successfully produce the webpage in the browser.

What happens:
Seemingly by random, the page load fails in an error page (site can't be reached). I attached two sample videos two show what I mean. One video shows a successful replay, whereas the other one ends in a site can't be reached error. Additionally, I captured the network traffic generated for the two cases.
issue.zip

What surprises me is that at some point in the failing capture an Ip address is requested which fails and seems to lead to the issue and I could not determine from where that Ip address comes from as the host file used by mahimahi does not list it.

I experienced the same issue with a different set up based on python from one of my colleagues.

Have you encountered this before or have any idea what might cause this? Thanks for looking into the issue