Adjust CacheFileTests.testCacheFileCreatedAsSparseFile

[tlrx]

It looks like some aarch64 architectures have a non 4KB block size but the test expects that only encryption at rest tests use a non default block size. This change removes the check on EAR.

Sorry @DaveCTurner for the iterations required to adjust this test :( I hope that the assertions for non 4KB block size will work everywhere now but we'll have to wait for ARM tests to execute to be 100% sure.

Relates #81362 (comment)

[elasticmachine]

Pinging @elastic/es-distributed (Team:Distributed)

[droberts195]

It looks like some aarch64 architectures have a non 4KB block size

Some background:

On aarch64 RHEL and its derivatives (including CentOS, Alma, Rocky) use a 64KB page size and Debian and its derivatives (including Ubuntu) use a 4KB page size. On x86_64 all of these use a 4KB page size.

This difference has also affected Kibana - see for example elastic/kibana#89382 (comment) and the other pages linked from there.

[DaveCTurner]

LGTM

[tlrx]

Thanks @droberts195. When I adapted the test for aarch64 I wasn't able to test on ARM architectures with CentOS so I picked up Amazon Linux 2 instead, which I thought was derived from RHEL.

(but I should have checked the docs...)

[tlrx]

Thanks both!

[droberts195]

When I adapted the test for aarch64 I wasn't able to test on ARM architectures with CentOS so I picked up Amazon Linux 2 instead, which I thought was derived from RHEL.

Yes, I think you're right that Amazon Linux changed the page size on aarch64 compared to the RHEL code - for example vectordotdev/vector#4392 (comment) agrees with you.

There are "bug for bug" rebuilds of RHEL, where they don't change anything apart from branding, such as CentOS, Alma and Rocky. And then there are slightly modified rebuilds like Oracle Linux and Amazon Linux where they tweak things a bit. So I guess the page size on aarch64 is one thing Amazon tweaked.