cli: don't allow balancer to be specified via command line flags

[spencerkimball]

Choosing between usage and range-count for rebalancing decisions is not something we should expect operators to specify on the command line. Let's choose the appropriate default for now and then move towards an automated approach.

@jseldess don't document the --balance_mode flag.

[mrtracy]

If we're going to remove this flag, that is going to have consequences for testing cockroach with multiple nodes on a single machine (which is the reason this flag was added in the first place). There are a few ways to do this:

1. Declare that "rangecount" is the new default. I'm fine with this.
2. Allow stores to be specified with a maximum size that is independent of the disk size. You would also need to update the way that a store's Capacity is computed to account for this. In this case, "usage" can remain the default and the single-node scenario still works.
3. Declare that running multiple nodes on a single machine is not a valid testing scenario for the purposes of rebalancing - I don't like this scenario, but it's possible.

Additionally, I don't know if we'll ever have an "automated approach" for this use case; this flag was intended for developers to try out alternative balance strategies, and while it initially is only used to solve the "multi-node, same hard drive" conundrum, it could eventually allow developers to compare different algorithms without needing special builds. Instead, I think it should moved to gossiped cluster configuration; the main problem with this flag is that it's a cluster-wide setting, not a per-node setting, and thus makes no sense as a command line argument.

[bdarnell]

@BramGruneir is working on the second option, to be able to cap the size of each store.

[spencerkimball]

@mrtracy again, I'm trying to avoid having @jseldess explain what these flags mean. And once we release a flag in the wild, we're going to have a much more difficult time removing it.

[mrtracy]

@spencerkimball I'm fine with removing the flag, we just need to pick an alternative solution for the problem it was solving. It looks like we're going to go with solution #2.

[BramGruneir]

Just a head's up that the maximum size work will be done by end of week.

[spencerkimball]

@BramGruneir should this be closed?

[BramGruneir]

As soon as I get the size changed merged.
The code should be ready to go this evening.

On Sun, Mar 6, 2016 at 1:26 PM, Spencer Kimball notifications@github.com
wrote:

@BramGruneir https://github.com/BramGruneir should this be closed?

—
Reply to this email directly or view it on GitHub
#4732 (comment)
.

[jseldess]

--store=size documented here: https://www.cockroachlabs.com/docs/start-a-node.html#store