In general, stuck pending statuses can be avoided by not enabling
the tiller native wait flag when updating releases, since tiller
then marks the release completed directly after applying the
resources to kubernetes.
However, when updating tiller itself, once kubernetes sees the
updated tiller resource, it can bring tiller down
before it has a chance to mark the release which contains tiller
as completed, leaving it in pending status.
This adds a preStop hook to both the standalone and sidecar tiller
containers to simply sleep to give them a chance to finish updating
their release, before terminating.
Ideally tiller would handle this on its own
via signal handling, but it doesn't. We could try to query for
the absence of PENDING_*** releases via `helm ls` before exiting,
however the helm CLI is not available inside the tiller image, and
those releases could be getting updated from another tiller instance,
or had already got stuck in that state previously, in which case we
don't want to hold up tiller termination.
Change-Id: I300c613f2a89eb1406531ce0a9af85c429a886f2
9573afd3c2