guides / openshift-install-cannot-execute troubleshootingupdated 2026-11-15 readelf -h ./app

openshift-install: cannot execute binary file - wrong CPU

2026-11-15 By luckyray mirror.openshift.com 5.0.1 - x86_64 / aarch64 / ppc64le / s390x

You downloaded openshift-install-linux.tar.gz, extracted it, ran ./openshift-install, and got cannot execute binary file. The file is not corrupt and the mirror is not broken: you downloaded the build for a different CPU, and the filename gave you no way to know.

Verified against mirror.openshift.com on 2026-11-15.


The trap: four architectures, one filename

The OpenShift mirror is organised by architecture in the directory, not in the filename:

/pub/openshift-v4/x86_64/clients/ocp/stable/openshift-install-linux-5.0.1.tar.gz   443.8 MB
/pub/openshift-v4/aarch64/clients/ocp/stable/openshift-install-linux-5.0.1.tar.gz  388.8 MB
/pub/openshift-v4/ppc64le/clients/ocp/stable/openshift-install-linux-5.0.1.tar.gz  392.3 MB
/pub/openshift-v4/s390x/clients/ocp/stable/openshift-install-linux-5.0.1.tar.gz    454.4 MB

The filename is byte-for-byte identical in all four directories; the sizes differ, which is the only visible signal that they are four different binaries.

So the question is not "did I get the right file" but "which directory was I in when I clicked". If you were browsing the aarch64 tree on an x86-64 workstation — or a search result dropped you there — you have an ARM binary and a perfectly correct error message.


Three commands to confirm

$ uname -m
x86_64                                  # what this machine is

$ file ./openshift-install
ELF 64-bit LSB executable, ARM aarch64, version 1 (SYSV),
statically linked, Go BuildID=...        # what the binary is

Machine in the header is the answer. For reference:

file says Runs on
x86-64 Intel / AMD 64-bit
ARM aarch64 ARM 64-bit servers, Apple Silicon VMs, Raspberry Pi 4/5
64-bit PowerPC ppc64le
IBM S/390 s390x

Third check, if you want the ELF identity directly:

readelf -h ./openshift-install | grep Machine
# Machine:  Advanced Micro Devices X86-64      <- or AArch64, or PowerPC64

Note: openshift-install is a statically linked Go binary. That rules out the other common cause of this message — a libc mismatch — because there is no interpreter to miss. For a static binary, the architecture is the only thing that can make the kernel refuse it.


The fix

Re-download from the directory that matches uname -m:

# on an x86-64 host
curl -LO https://mirror.openshift.com/pub/openshift-v4/x86_64/clients/ocp/stable/openshift-install-linux.tar.gz
# on ARM
curl -LO https://mirror.openshift.com/pub/openshift-v4/aarch64/clients/ocp/stable/openshift-install-linux.tar.gz

Then verify before you run it — one second, and it prevents a second 400 MB download:

file ./openshift-install | grep -q "$(uname -m | sed 's/x86_64/x86-64/;s/aarch64/aarch64/')" \
  && echo "architecture matches" || echo "wrong build"

The same rule applies to the whole client set: oc, kubectl, and openshift-install are all published per architecture under the same clients/ocp/<version>/<arch>/ layout. Mixed tools on one workstation is a common cause of "it worked yesterday": someone updated oc from a different directory.


Why copying the binary from another machine fails

Copying openshift-install from an ARM jump host to an x86-64 bastion copies the architecture with it. This works only with emulation:

# user-space emulation, slow but functional for a CLI
sudo apt-get install -y qemu-user-static
sudo update-binfmts --enable qemu-aarch64

For an installer that runs for 30–45 minutes and talks to the cluster API, emulation is the wrong answer: download the native build. Emulation is for containers and tests, not for the tool you use to create production clusters.


What does not help

Attempt Why it fails
chmod +x openshift-install Permission was never the problem; the kernel rejected the contents
sudo ./openshift-install The kernel does not check who you are
Copying a loader from another machine Static binary — there is no loader to fix
Re-extracting the tarball The bytes were fine; the build was for the wrong CPU

Checklist

  1. uname -m — what am I on?
  2. Check the mirror directory you downloaded from, not the filename.
  3. file ./openshift-install before the first run.
  4. Keep oc, kubectl and openshift-install from the same architecture tree.
  5. In scripts, assert the architecture before invoking the installer.

Paste the binary into ELF Inspector if you want the header read in the browser — it shows the machine type and whether the binary is static in one screen.


Mirror names and sizes verified with HEAD requests on 2026-11-15 against mirror.openshift.com. The architecture check outputs above are from a real x86-64 host.

Related

guides
  1. glibc and libstdc++ versions by distribution — measured, not remembered
  2. version `GLIBCXX_3.4.29' not found — and why copying libstdc++ does not fix it
  3. ldd: not a dynamic executable — what it means (nothing is broken)
  4. bash: cannot execute binary file - what bash actually checked
  5. node: cannot execute binary file - the four ways Node breaks
  6. exec format error in Docker - measured, with the fix that works