[kea-dev] Design for Kea Host Reservation

Tomek Mrugalski tomasz at isc.org
Wed Oct 1 18:52:14 UTC 2014


On 30.09.2014 13:20, Marcin Siodelski wrote:
> I have created a new design document on Kea wiki:
> http://kea.isc.org/wiki/HostReservationDesign for Host Reservation.
Here are my comments. I like the design a lot. I think it will address
our needs. We will probably tweak it a little bit over time, but it a
very solid proposal. Here are my comments:

Information stored for HR section
IP address(es) - should mention that list is for IPv6 only, v4 need a
single address only. There's no need to reserve more than one IPv4
address. Also, please explicitly say that IP means IPv4 and IPv6.

Temporary addresses. We do not support temporary addresses at all and
in a sane environment, there is no requirement to reserve them in any
way. On the contrary, it should be an explicit non-requirement. We
should probably state that somewhere in the design (and hope that odd
users wanting their temporaries to be static will never appear :)

We should add that currently there is no way to define what options
can be sent for a given class. This capability will be added in the
future.

I'd love to see DUID reservation example for DHCPv4.

I like the paragraph explaining what the prefixes (dhcp_, dhcp4_ etc.)
mean. This is sort of intuitive, but it's nice to have it explicitly
stated.

Shouldn't host_id and reservation_id be of type primary key? The diagram
in MySQL database schema suggests it is a simple int. I'm not a
MySQL expert, so this is an honest question. I think that setting a
field to primary key type introduces three things: first, it is being
assigned automatically during inserts. Second, the next value is being
stored and applied automatically (autoincrement). Third, a unique
restriction is added.

I like the BEFORE DELETE trigger, but I think we need something
similar for the other way. Something like BEFORE INSERT that checks
that the host_id exists when adding entries into ipv6_reserverations.
Is there a BEFORE UPDATE trigger? It could be useful, too.

Comment for dhcp_identifier in the table does not mention that it must
not be null. I think we won't support a case when no identifier is
given at all.

Description for ipv4_address is strange: "This value must not be set
to NOT-NULL if dhcp4_subnet_id is not specified." It would be better
to simply say "must be NULL if ..."

I'm thinking that we may specify hostname a bit more precisely. How
about this: it may contain a single label, e.g. "host-foo" or a FQDN,
e.g. "host-foo.example.org.". If only hostname is specified, it will
be appended with domain name specified for a given subnet. I can
imagine that sysadmins would hate to write down their own domain
thousands of times.

Why do we need reservation_id in the ipv6_reservations at all? To
refer to specific option when updating it manually? That's fine, but
if that is the rationale for it, we should say in the description that
the field is not going to be actively used by the server and can be
used by third party during database maintenance.

prefix_len description should mention that allowed length is 0..128.

Thanks for explaining the rationale for splitting v4 and v6 options
into separate tables. "Why?" was my immediate question when I saw the
tables layout. Your explanation answers my doubt.

I'm a total MySQL noob, so forgive my ignorance, but why do we need
ALLOW_INVALID_DATES? MySQL does not simply insert invalid dates, and
sort of sanitizes it to zeros. MySQL doc
(http://dev.mysql.com/doc/refman/5.5/en/sql-mode.html, also applies to
older versions) says that invalid date of 2014-04-31 will be sanitized
to 0000-00-00.

Space description: It may be useful to add a reference to Kea ARM,
section 6.2.9 (v4) and 7.2.11 (v6) for people who are unfamiliar with
the concept of option spaces. It is pretty Kea and ISC DHCP specific.

The section about how we interpret dhcp_client_class, host_id and
subnet_id need to be more precise. People may read it, think that
we'll support something and the be disappointed if we don't. It think
the following approach will give us the most flexibility:

SELECT ... FROM dhcp4_options WHERE subnet_id=X AND (host_id=Y OR
dhcp_client_class in (class1,class2,...,classN))

I'm not sure if the way how you organized the relation between
SrvConfig and CfgHost and HostContainer are optimal. I would think
that since the host reservations are defined per subnet, it should be
SrvConfig keeps Subnet4 and Subnet6 collections (as it does now) and
each Subnet{4,6} can contain zero or more CfgHost objects. The
difference is actually important from performance perspective.
Imagine 100 subnets with 1000 hosts in each. In the solution you
propose you'd have to pick the appropriate reservation out of 100.000
for every request. It will be much slower than picking it out of 1000
reservations. You don't waste any additional time for picking subnet
as that is done in the code already and is unavoidable.

That's it. I have no further comments. As I said, I like the design a
lot. Good job!

Tomek





More information about the kea-dev mailing list