I've been discussing this with Nazrul, but I'd like to get input from
more people...
If you look at what create-instance does in v2, there are two parts:
1. Create the entries in domain.xml to represent the new instance.
2. Create the entries in the filesystem on the instance node
for the new instance.
In v2, the asadmin command "create-instance" asks the DAS
to do #1. The DAS then asks the node agent to do #2. In
the v3 design of asadmin, this command would be a completely
remote command, with no special local knowledge.
What we need in v3, until we have a node agent, is a local
asadmin command that runs on the node that will host the
new instance. This command needs to ask the DAS to do
#1, then do #2 locally.
This is subtly different from what the existing "create-instance"
command does. I'm torn as to whether this should be an
option to the create-instance command, or a new command.
If we support ssh as a replacement for the node agent, it
would be good to have an asadmin command that just
does #2. The DAS implementation of create-instance could
do #1, then use ssh to run a command that does #2 on the
remote node. create-instance-filesystem seems like a good
name for that command.
One way to think of this is that we need a couple of primitive
commands to do #1 and #2, and then we need some "macro commands"
or "scripts" that combine them for the two different use cases.
The primitives could be "_" commands - _create-instance
and _create-instance-filesystem, for instance. This will
cause them to not show up in asadmin list-commands.
Here's one way to do it...
"create-instance" is an asadmin remote command.
On the server (DAS) the create-instance command
calls the _create-instance command, then (if possible)
executes the _create-instance-filesystem command on
the remote node.
"_create-instance-filesystem" is an asadmin local command
that does #2 above, and nothing else.
"create-local-instance" is an asadmin local command that
executes the remote command _create-instance, then
executes the other local command _create-instance-filesystem.
Here's another way to do it...
"create-instance" is an asadmin local command. Its behavior
depends on some option, which I'll call --createfilesystem for
now, but which needs a better name. If --createfilesystem is
not specified, the local create-instance command simply
calls the remote create-instance command.
On the server (DAS) the create-instance command
calls the _create-instance command, then (if possible)
executes the _create-instance-filesystem command on
the remote node. (Just like above.)
If --createfilesystem is specified, the local create-instance
command executes the remote _create-instance command, then
executes the other local command _create-instance-filesystem.
"_create-instance-filesystem" is an asadmin local command
that does #2 above, and nothing else.
Either approach works. And the same primitive commands
are needed on both cases.
The question is, which will users find more natural and understandable?
A single command with an option to control the different behavior?
Or two different commands for the different use cases?
Comments? Which would *you* prefer?