Peter,
Thanks for the review. Comments below :
On 06/ 8/10 10:03 PM, Peter Tribble wrote:
On Tue, Jun 8, 2010 at 1:10 PM, Matt Keenan<[email protected]> wrote:
Apologies,
Forgot to add external link :
http://hub.opensolaris.org/bin/download/Project+caiman/auto_install/ai-multi-design-0.1.pdf
For data pools, you say you don't support cache/log/spare devices. I see
this as a significant omission and an inconsistency. I can see a reasonable
limitation being that you only support 1- to N-way mirrored root pools; if
you're going to support data pools at all then allowing cache/log/spare devices
seems essential. (And not terribly difficult.) Besides, I expect such pool
configurations to be very common.
For this iteration of development cache/log/spare devices will not be
supported as they can easily be added post install. However the schema
is designed to facilitate expansion and therefore to possible include
this functionality in a later development cycle.
However we really want to keep the schema as simple as possible, and for
try and keep complex pool layouts to a serialization schema (yet to be
defined).
There's an example in 3.2.4.1:
zpool create mypool mirror d1 d2 raidz d3 d4 d5 d6
seriously, I hope this would be rejected!
d1 d2 etc, are just for example purposes and do not relate to actual
physical devices. Other than that why would this be rejected ?
I'll change the example to include cNtNdN type device specifications.
There's also
mirror c0t0d0 c0t1d0 c0t2d0
it may be a minor thing, but I immediately think of a root pool when
given those devices, and that example then suggests that I can use EFI
labelled disks rather than SMI partitions.
In the above scenario this would actually be a data pool, I can change
the text in the document to clearly indicate it's a data pool example.
Appendix 1:
default name is rpool - does this mean the name of the root pool
can be changed? Is the default name of a data pool still rpool?
The default name for root pool if not specified in the manifest is
"rpool", as is currently the case. The root pool can be any name as
specified in the manifest.
For data pools, the default name if not specified will be "dpool".
Comment above "zpool_file_system_properties" - is that zpool
create or zfs create? And -O or -o? (It's not immediately obvious
why there are two sets of property settings here, when dataset
properties can also be specified later on.)
In the comment above the section I do state "zpool create -O" for
zpool_file_system_properties. Agreed dataset properties can be set in
the zpool_dataset element, however if the user decides to just have the
default dataset layout, then this provides a means of applying
properties to these.
"Pool type, concatenated or mirror. Default if
this element is specified is concatenation."
Missing not? And the list is longer than just concatenated or mirror.
Yep well spotted missing "not" from comment.
Target devices - does this mean that if you just tell it to create a
raidz, without listing devices, then it will just pick some at random?
No, as part of the zpool validation if raidz is specified, I do validate
that the user has specified at least the correct number of devices for
that pool type. In section 3.2.2.4 I specify for validation:
"Correct number of targets specified for each pool type, e.g. of raidz2
must have at least 3 targets specified."
Is this sufficient ?
again thanks for the review.
regards
Matt
That seems horribly broken - I would expect that to just fail with
an error. If you want auto-selection, then I would prefer to say
so explicitly.
Thanks,
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
