[Freeipa-devel] Structured DNS record API proposal - summary

[Martin Kosek]

On Wed, 2011-09-21 at 11:22 +0200, Martin Kosek wrote:
> On Tue, 2011-09-20 at 11:22 -0500, Endi Sukma Dewata wrote:
> > On 9/20/2011 6:15 AM, Martin Kosek wrote:
> > >>> ACK.  Proposal looks like it will work fairly easily with the UI.
> > >>> We'll have to make some chagnes due to the Add doing something
> > >>> different based on the type, but that is the case anyway.
> > >>
> > >> Yes, I was thinking how can we integrate this new API to WebUI. AFAIK
> > >> you use dnsrecord-add $ZONE $REC --a-rec=... --mx-rec=... for adding a
> > >> new DNS record and dnsrecord-mod $ZONE $REC --mx-rec=... when for
> > >> example the mx record is being modified. All MX values (even the
> > >> unmodified ones) are passed to dnsrecord-mod.
> > >>
> > >> 1) I was wondering how the new dnsrecord-<rrtype>-add commands can be
> > >> used. I suppose WebUI will know a list of DNS record types with these
> > >> new structured commands and offer the user new window to add a record
> > >> for these types instead of typing them directly to the text box as it is
> > >> now.
> > 
> > When adding a DNS record the user will specify the name and the type, 
> > then the UI will show a set of fields based on the selected record type.
> > 
> > So instead of a generic 'data' field like below (click Add):
> > 
> > http://edewata.fedorapeople.org/freeipa/install/ui/index.html#dns=dnszone&identity=dns&navigation=identity&dnszone-facet=default&dnszone-pkey=ayoung.boston.devel.redhat.com
> > 
> > it will be similar to Permissions (click Add):
> > 
> > http://edewata.fedorapeople.org/freeipa/install/ui/index.html#rolebased=permission&ipaserver=rolebased&navigation=ipaserver
> > 
> > The UI will use the type to pick the correct dnsrecord-<rrtype>-add 
> > command and each parameter in that command will have a corresponding 
> > field to enter the value.
> 
> Yes, I think this will work fine. Would it make sense to create
> dnsrecord-<rr-type>-add commands also for non-structured DNS records? I
> mean for example for A, AAAA, PTR, CNAME, ... record, which have just
> one simple value or let plain old dnsrecord-add --a-rec=... handle it?
> 
> > 
> > >> 2) But my main concern here is how the modification of current DNS
> > >> records should work. Say, we have 2 MX records for example.com. How can
> > >> we modify one of it in a new structured interface?
> > >>
> > >> We would have to implement dnsrecord-mx-show method so that you can fill
> > >> all the text areas (preference, mailserver). Question is how to refer
> > >> the value we want to show since DNS records are multivalued. We could
> > >> pass --dnsrecord="..." with DNS record value, e.g. "0 mx.example.com."
> > >> and then use the same value for dnsrecord-mx-mod. The whole command
> > >> sequence would look this way:
> > >>
> > >> dnsrecord-find example.com      -- get all DNS records for example.com
> > >> dnsrecord-show example.com @    -- show DNS records directly in the zone
> > >> NS: "ns.example.com"
> > >> MX: "0 mx1.example.com."
> > >> MX: "1 mx2.example.com."<<  user wants to modify this one ->  new window
> > 
> > I think for each record value the primary keys are the zone name, record 
> > name, and the value itself. To simplify operations, we should use the 
> > value as a single string. For CLI, users can copy & paste the value more 
> > easily.
> 
> Agreed. As Adam Tkac suggested, we can simplify this with interactive
> prompt so that user doesn't have to copy&paste, but just choose a record
> to -show/-mod.
> 
> > 
> > For UI it depends whether (1) we're going to keep the current edit page 
> > where all records with the same name are considered a single entry, or 
> > whether (2) we're going to edit each record value in a separate page. 
> > See ticket #1478.
> > 
> > If we stay with (1), the link to the edit page consists of zone name and 
> > record name only. But if we pick (2) the link consists of zone name, 
> > record name, value, and type (which can be obtained from -find output).
> 
> This is more of a UXD decision, server API will remain intact. I just
> see 2 issues here:
> 
> 1) If you let user edit multiple structured DNS records, you would have
> to call dnsrecord-<rr-type>-show multiple times so that you can populate
> all the fields. This can slow down things.
> 
> 2) Some DNS records may be pretty large. MX record data is small, but
> for example CERT records have an entire certificate stored in it.
> Wouldn't there be a problem if we place the large DNS record in URL?
> 
> > 
> > >> dnsrecord-mx-show example.com --dnsrecord="1 mx1.example.com."
> > >> PREFERENCE: 1	<<  user modifies this to 0
> > >> MAILSERVER: mx2.example.com.
> > 
> > For consistency, the record value should be specified as an argument 
> > instead of an option (like in automount). So it will be like this:
> > 
> > dnsrecord-mx-show "example.com" "@" "1 mx1.example.com."
> > PREFERENCE: 1
> > MAILSERVER: mx2.example.com
> 
> This can be done.
> 
> > 
> > If we stay with (1) the UI will have to call the dnsrecord-<rrtype>-show 
> > for each value to get the value of each fields. The UI will need to 
> > implement a new widget (or section) that can handle multiple fields 
> > which will be duplicated for each value.
> 
> Ah, yes - as I wrote above. This would also take more time to process.
> 
> > 
> > The edit page for (2) is much simpler since it only needs to handle a 
> > single type at a time. The output of the -show command will be used to 
> > populate each field.
> > 
> > >> dnsrecord-mx-mod example.com --dnsrecord="1 mx1.example.com." --preference=0
> > 
> > When updating the value, option (1) is a bit more complicated because 
> > the UI will have to find the dirty record and then find the dirty field. 
> > Option (2) is simpler because it will only need to find the dirty field, 
> > but both will execute the following command:
> > 
> > dnsrecord-mx-mod "example.com" "@" "1 mx1.example.com." --preference=0
> > 
> > I think option (2) is more clear to users because we only have to 
> > introduce 2 concepts: zone and record (which is the individual value). 
> > With option (1) we will have to explain the underlying LDAP entry that 
> > will be deleted automatically when the last record value is deleted.
> > 
> 
> When I look at it, option (2) looks better for our case.
> 
> Martin

I think this entire discussion became too academic. I would like to make
it more down to earth :-) Adam proposed new complex attributes, IMHO it
should be discussed separately from this API proposal. I gave it some
thought and I just don't see the benefits that it would bring to DNS.
Personally, I would like to continue with API that was proposed so far.

Summary of the steps for new DNS types:

1) New API shall be implemented only for non-DNSSEC resource record (RR)
types that are supported by bind-dyndb-ldap.
DNSSEC RR types to be left unimplemented: DS, KEY, NSEC, RRSIG, SIG
Unsupported RR types to be removed at all: APL, DHCID, DLV, DNSKEY, HIP,
IPSECKEY, NSEC3, NSEC3PARAM, RP, TA, TKEY, TSIG
In future, DNSSEC attributes should be generated automatically. But let
us leave this discussion to the future.


2) -add commands shall be implemented for structured RR types:

> ipa dnsrecord-afsdb-add --subtype=INT --hostname=STR
> ipa dnsrecord-cert-add --type=ENUM --tag=INT --algorithm=ENUM --certificate=STR
> ipa dnsrecord-kx-add --preference=INT --exchanger=STR
> ipa dnsrecord-loc-add --lat-deg=INT --lat-min=INT --lat-sec=FLOAT --lat-dir=ENUM --lon-deg=INT --lon-min=INT --lon-sec=FLOAT --lon-dir=ENUM --alt=FLOAT --h-precision=FLOAT --v-precision=FLOAT
> ipa dnsrecord-mx-add --priority=INT --mailserver=STR
> ipa dnsrecord-naptr-add --order=INT --preference=INT --flag=ENUM --service=STR --regexp=STR --replacement=STR
> ipa dnsrecord-srv-add --priority=INT --weight=INT --port=INT --target=STR
> ipa dnsrecord-sshfp-add --algorithm=ENUM --type=ENUM --fingerprint=STR

OPEN QUESTION: should we implement these new commands also for discrete
DNS records types to be consistent? I mean for example A, AAAA, CNAME,
PTR, ... They would look like

> ipa dnsrecord-aaaa-add --ip-address=IPAddress

BENEFITS of this approach (command per RR type):
- use can get all help for RR type by simply typing "ipa help
dnsrecord-mx-add"
- we would be able to implement helper methods consistently on one
place, for example:
dnsrecord-aaaa-add --from-mac=00:1D:BA:06:37:64


3) -show commands shall be implemented for structured RR types:
These commands will let UI populate the (text) fields, ENUMs, etc. They
would look this way

> ipa dnsrecord-mx-show ZONE NAME VALUE

For example:
> dnsrecord-mx-show "example.com" "@" "1 mx1.example.com."

OPEN QUESTION: should we implement also -find methods
(dnsrecord-mx-find) so that UI can for example populate text fields for
all (MX) records for one DNS name?


4) -mod commands shall be implemented for structured RR types:
API would be almost the same as with -add commands. User (WebUI) would
just have to identify which record should be modified:
a) by copy&passing the raw DNS value directly to the command:

> dnsrecord-mx-mod example.com @ "1 mx1.example.com." --preference=0

b) (CLI only) by using an interactive wizard that would let user choose
the modified record like this way:

> dnsrecord-mx-mod example.com @ --preference=0
Which record would you like to change?
[1] 1 mx1.example.com.
[2] 10 mx2.example.com.
DNS record: <user enters the number>

I have already implemented an interactive wizard like this for
dnsrecord-del and it works fine.

Martin