Wrong binary is uploaded when deploying to multiple architectures

[mjaschen]

The wrong nginx-prometheus-exporter binary is installed with the nginx exporter role when the target servers run under multiple processor architectures.

sample from playbook:

roles:
  - prometheus.prometheus.nginx_exporter

What happens if different target processor archictectures are involved?

The role downloads the archives for each architecture (nginx-prometheus-exporter_1.1.0_linux_amd64.tar.gz, nginx-prometheus-exporter_1.1.0_linux_arm64.tar.gz etc.) and extracts them – but to the same target file (/tmp/nginx-prometheus-exporter).

The upload task picks the target file and uploads it to all servers. This means that all server get the same binary.

A possible fix could look like this:

• ensure that the extracted binaries get unique names for each processor architecture, e.g. /tmp/nginx-prometheus-exporter-amd64, /tmp/nginx-prometheus-exporter-arm64 etc.
• adjust the src setting for the copy task, so that it picks up the correct file for the target machine

[gitperr]

It also might make sense to download the file on the target machine then, because we should be able to find the arch with ansible facts.

[gardar]

My suggestion for now would be to group the architectures and then have different tasks in the playbook targeting the different groups.

[gitperr]

Yep, this is a problem that can be solved by maintaining your inventory in a peculiar way. ☝️