Rollback etcd-manager in +1.19

[sp-francisco-manas]

Due to Go upgrade to 1.15 on etcd-manager version 3020210122, kOps upgrades to 1.19 or newer are breaking (or are not being straightforward) because of this. The first master during a rolling update never joins the etcd clusters.

Until a proper fix is implemented I would opt for rolling back the version kOps uses to 3020201209 or a new one if etcd-manager is recompile with Go 1.14.

The current workaround is to add the following settings on clusterSpec for each etcd:

    manager:
      env:
      - name: GODEBUG
        value: x509ignoreCN=0

References:

• Auth Errors on Cluster Upgrade kopeio/etcd-manager#399
• https://golang.org/doc/go1.15#commonname
• crypto/x509: Go 1.15 will break default connections to AWS RDS golang/go#39568

[sp-francisco-manas]

I'm closing the issue as it is working for us now. In our case, the problem was that we enabled etcd-manager metrics and by mistake, we were trying to bind both etcd and etcd-events metrics endpoint to the same metrics port. Thus etcd-events was not starting.

Although the master starts and join the K8s cluster there're still some errors in the beginning about the certificate and Go 1.15 deprecation, but it seems to not affect etcd operation.