Description of problem: Installation failed at task openshift_health_check with instance of 16GiB Memory Version-Release number of selected component (if applicable): openshift-ansible-3.6.85-1.git.0.109a54e.el7.noarch.rpm How reproducible: Always Steps to Reproduce: 1. launch instances on AWS EC2 with type t2.xlarge 16GiB Memory 2. set up cluster by openshift-ansible-3.6.85-1.git.0.109a54e.el7.noarch.rpm Actual results: Installation failed. "msg": "Available memory (15.6 GB) below recommended value (16.0 GB)" Expected results: Installation SUCCESS 16 GB RAM meets system requirements https://docs.openshift.com/container-platform/3.5/install_config/install/prerequisites.html Additional info:
I bet a lot of cloud platforms will be shaving ~256-512MiB off the quoted size of the instance. We should probably account for that.
There's already a little bit of fudge factor since we're measuring GB instead of GiB. But looking at a t2.xlarge instance it does look like AWS has some disappearing RAM: # dmidecode --type memory Handle 0x1000, DMI type 16, 15 bytes Physical Memory Array Maximum Capacity: 16 GB Handle 0x1100, DMI type 17, 21 bytes Memory Device Size: 16384 MB # cat /proc/meminfo MemTotal: 16005316 kB MemFree: 14720336 kB MemAvailable: 15595096 kB MemTotal is what the check is looking at. It's substantially lower than the "hardware" RAM. I don't think support would deny a customer with this memory setup, so we should probably introduce a bit more fudge factor for those with memory that's "close enough".
I think this is expected and not specific to AWS: https://access.redhat.com/solutions/3006511 So yes we probably have to always give a small % of margin
https://github.com/openshift/openshift-ansible/pull/4301
PR just merged to master, so should be available to test whenever there's a new openshift-ansible build.
Verified on openshift-ansible-3.6.94-1.git.0.fff177b.el7.noarch.rpm Installation SUCCESS Fixed.
never released