These IP addresses are supposed to be unreachable in most networks. For
some reason they are routed differently on Canonical's infrastructure
and sent to the proxy, which makes some tests to fail because they
expect a timeout instead when trying to reach the IPs.
Blocking these addresses might have unintended side effects since
a quick search in Debian Code Search shows that many packages uses
10.255.255.1 in their test suite. However, it seems to me that most
packages indeed expect it to be unreachable, so I guess making that a
reality should in fact help more than one package, but it's difficult to
be certain that it won't break one particular package somewhere.
fix: web/download-all-results: iterate through container in batches
I had been previously trying to run download-all-results for all
supported and esm releases except noble and oracular, and
download-all-results was consistently running out of memory with message
"Killed".
I assumed this was because we were fetching the container with:
full_listing=True
And storing the names and run ids of every object in a given container.
Our "release" containers e.g. autopkgtest-jammy are massive, with at
least 100,000 swift objects, which was causing the script to OOM.
This commit amends the issue by iterating through each individual
container in batches of size 1000. I've tested it and the script no
longer OOM's.
