osbuild-exec: add buildlog and monitor support

[croissanne]

This commit adds a new BuildLog option to the OSBuildOptions that
can be used to generate a streamed buildlog (e.g. to a file or a
websocket). This supports the --with-buildlog option in
image-builder-cli.

With the buildlog option set, stderr is forwarded to stdout to make sure
statements appear in order. If BuildLogMu is set, this mutex will be
used for the synced writer which writes to both stdout and the
buildlog. Callers can use the same mutex to write statements to the same
buildlog.

The errorWriter option got removed, and callers can just use Stdout
and Stderr to set the stdout and stderr of the command.

[schutzbot]

This PR changes the images API or behaviour causing integration failures with osbuild-composer. The next update of the images dependency in osbuild-composer will need work to adapt to these changes.

This is simply a notice. It will not block this PR from being merged.

[lzap]

Some concerns around FD and shared mutex. It does not appear that multiple goroutines would need to create multiple instances of that writer.

[croissanne]

woops, accidentally deleted the branch 🤦

[croissanne]

Ok this still fails go test -race with image-builder-cli so problem was in dereffing the mutex

[mvo5]

This looks good, thanks for working on this! Its very cool that this moves from ibcli and that we can reuse it for composer now. Some tiny comments inline and I think the linter had some (valid) comments too.

Unfortunately doing the build log its not super easy, I'm not sure we can do much about it but we could rethink the approach. In ibcli we really wanted to have a buildlog file that is streamed to in realtime with the details of the build but the user sees a normal progress. We could move this task into osbuild, i.e. something like osbuild --monitor JSONSeqMonitor:3 --monitor LogMontior:1 where osbuild would montior to fd=3 via jsonseqmoniotor and to stdout via the log monitor. But it would probably not help that much (I think) because the problem is that stderr/crashes from osbuild itself would still have to somehow synced and end up in the buildlog.txt so we would still need something that ensures that and the problem just moves sideways (but maybe I'm missing something here, I have not tried it).

[mvo5]

Thank you!

comments applied