[stork-users] Stork installation feedback

[Ben Monroe]

Hi all. I am new to the list and Stork, though I have been using Kea for a
number of years. I recently setup Stork but ran into several issues that I
hope can be reported as constructive feedback.



For the last few years I have been running BIND and Kea in containers. This
has worked well, though it was a little fragile and required some
administration after an unplanned power outage. This gave me an opportunity
to revisit this setup. I decided to re-implement this as a Kubernetes
cluster, which gave me an opportunity to implement Stork.



Note: I am not interested in debating whether this should be run as a
container. It suits my environment and use-case. For those wondering, I am
using MetaLB for load balancing and Longhorn for distributed storage. As
for Kea itself, it is located on a separate VLAN with DHCP relay on
switches forwarding broadcasts for VLANs that it services.



[1] The documentation has an "Installing on Alpine" section. I spent a
while trying to get this to work  but ran into multiple problems.

[1.1] No package requirements are listed, though apparently  curl, bash,
and sed are needed. Should specify.



[1.2] Section 2.5.1.3 indicates to curl
https://dl.cloudsmith.io/public/isc/stork/cfg/setup/setup.alpine.sh'. This
returns  "Not found". A bit of investigation indicated to use '
bash.alpine.sh' and pipe it to bash instead.



[1.3] Finally, trying to add the Stork server package results in the
following error:

  / # apk add --allow-untrusted isc-stork-server

  ERROR: unable to select packages:

    isc-stork-server (no such package):

      required by: world[isc-stork-server]

  / #

Apparently CloudSmith does not even host Alpine binaries for Stork. Is the
documentation in error or will Alpine binaries be added in the near future?

Links:

https://stork.readthedocs.io/en/latest/install.html#installing-on-alpine

https://cloudsmith.io/~isc/repos/stork/packages/



[2] I am using Alpine as it is typically the most light-weight for building
containers. So I compiled my own build. This required a few dependencies
not listed in the documentation. For reference, I will include them here
(along with a Docker step).

  docker run -it --rm alpine:latest /bin/sh

  apk update

  apk add git ruby go npm protoc python3 openjdk17 openssh

  gem install rake

  git clone https://gitlab.isc.org/isc-projects/stork

  cd stork

  rake build

  rake install:agent

  rake install:server

Following this, copy the output in stork/dist to an external location,
create a dockerfile copying in Stork Server and upload to an image
repository. Repeat for Kea and add in the Stork Agent.



[3] Stork Agent and all Kea components (kea-ctrl-agent, kea-dhcp4,
kea-dhcp6, and kea-dhcp-ddns) all need to be installed together. Previously
I ran each Kea component in separate containers. However, 1) Stork Agent
must reside with Kea CA, and 2) Kea CA can only connect to DHCP4, DHCP6,
and DDNS via Unix sockets. Hence, a monolithic container containing all was
required. In my opinion, this is undesirable, at least for my type of
setup. Other connectivity options are desirable.



[4] Initialization using current PostgreSQL required appending the
following:

  GRANT ALL ON SCHEMA public TO stork;



[5] Configuration of STORK_AGENT_HOST required special handling. Within the
Kea container, I do not know what the IP address will be. Setting it to
0.0.0.0 allows Stork Agent to bind, but Stork Server will naturally never
be able to connect to this. Stork Server communicates with Kea via a
ClusterIP service pointing to Kea pod(s). However, within the Kea pod,
Stork Agent cannot bind to this ClusterIP service. In the end, I had to
resort to DNS trickery: I let the Stork Server pod resolve the ClusterIP
service name as normal, though within the Kea pod this ClusterIP service
resolves to 0.0.0.0, which is sufficient for Stork Agent to bind. (For
those interested, I used hostAliases in the deployment YAML to override the
pod DNS.) Could another environment variable be introduced to distinguish
between the Stork Agent hostname as called by Stork Server and what Stork
Agent binds to?


After resolving these problems Stork is up and running. It looks like it
will be very useful. Kea functionality is very good. I hope to see more
BIND functionality added in the future.


Best regards,

Ben Monroe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/stork-users/attachments/20230830/6600cedb/attachment.htm>