Simplify rollback arguments

[ironcladlou]

Rolling back a deployment currently requires a reference to a specific deployment/replication controller name. The command should support work in terms of deployment configs and versions, because users shouldn't have to care about specific deployments (which aren't formal API resources).

[ironcladlou]

@smarterclayton @liggitt this needs prioritized.

[smarterclayton]

P2

On Jul 1, 2015, at 11:21 AM, Dan Mace notifications@github.com wrote:

@smarterclayton https://github.com/smarterclayton @liggitt
https://github.com/liggitt this needs prioritized.

—
Reply to this email directly or view it on GitHub
#3338 (comment).

[ironcladlou]

One thing I thought of while revisiting this: just a number is insufficient because multiple deployment configs are likely. The command signature would need to include options like:

rollback <deployment> (config/version inferred)

or

rollback <config> <version> (deployment inferred)

Would it be too confusing to support both? The 2nd syntax seems like the best one.

[0xmichalis]

This is similar to start-build in that it has to handle dcs and deployments while start-build handles bcs and builds. Should this and #1995 have the same solution @smarterclayton @bparees ?

[bparees]

@Kargakis seems reasonable to me.

[deads2k]

I think we should support oc rollback <dc> where it chooses last-known-good-version (if possible) or next-to-last-version by default. That means that the number of args will be insufficient to distinguish these two cases. That takes you into resource builder style names (with a default).

[ironcladlou]

Apparently we should move this conversation to trello. I'm going to close this issue to avoid crosstalk.

Further discussion here: https://trello.com/c/ZREvUIf2

[smarterclayton]

Why are multiple deployment configs likely?

On Jul 8, 2015, at 10:11 AM, Dan Mace notifications@github.com wrote:

One thing I thought of while revisiting this: just a number is insufficient
because multiple deployment configs are likely. The command signature would
need to include options like:

rollback (config/version inferred)

or

rollback (deployment inferred)

Would it be too confusing to support both? The 2nd syntax seems like the
best one.

—
Reply to this email directly or view it on GitHub
#3338 (comment).

[smarterclayton]

A couple of things here. Deploy applies to deployment configs, not to deployments (which aren't an object). The frame of reference is a DC and a version is either an argument (--version 5) or a modification to the name (foo@10, but since we haven't decided how we're doing uid reference I'd like to avoid that for now). So a flag is more appropriate for now.

[smarterclayton]

Why should we move discussion to Trello?