The COMMIT state requires good network access to perform all the checks before committing the deployment/update. However, doing a rollback (or generally aborting the deployment) because of a temporary network outage could be an expensive thing to do potentially even blocking devices from being updated if they have (windows of) bad network. We either need to:

• add a new PRE_COMMIT state in which the client would sit waiting for good network, or

• be able to stay in the COMMIT state waiting for good network before performing checks.

In both cases, the Mender work has to be rescheduled to let other things run in the meantime. The number of retries and intervals between them should be configurable, either with specific options or using the backoff config options that we already have.