commit e950cf7aa2d9309c23380bf76606871ca5c3b56d
author Jan Kundrát <jan.kundrat@cesnet.cz> Fri Sep 06 17:46:11 2024 +0200
committer Jan Kundrát <jan.kundrat@cesnet.cz> Tue Sep 17 14:44:08 2024 +0200
tree 147cb636789aec3ec150719582ebe725b27bc7a0
parent febc9991a13351300356103ab70cb2d7a1f5e6ca

libyang v3

This is a huge commit which changes a lot of things in the system,
including the boot process and configuration handling. There are also
massive updates all over the stack which have accumulated in the
upstream projects. In no particular order:

- The configuration schema has changed in an incompatible way. That's
  due to an update of some drafts related to the ietf-netconf-server
  YANG model and its dependencies; as a result, the way how SSH keys and
  SSH listening configuration is managed in YANG data is incompatible
  with the previous approach, and that means that we *cannot* import the
  previous config into sysrepo which already has new YANG modules.
  "Solve" that by simply dropping the entire YANG config (!) because we
  decided that it's sometimes better to start from scratch. Yay.

- Config migrations now work at the JSON level, not through sysrepo.
  There is currently no real "migration" apart from the initial one (the
  one that brings the config schema from the implicit/empty "v0" to
  "v9"). The TL;DR version is that future migrations will work on an
  JSON input, producing some new JSON output, and only after the
  migration is done, we push stuff into sysrepo. Also, the migrations
  will (likely) not be performed in a sequence, on a cumulative basis,
  but it is now a one-shot process with no intermediate steps. This
  should hopefully simplify the launching/wrapping code. From the
  developer's standpoint, any migration should focus on bringing the
  system to the "final state" (v_new), not to v_old + 1.

  These migrations will use `jq` heavily. So far, we only have some
  "simple" code which merges objects (and arrays!) recursively, courtesy
  of Peter Koppstein (https://stackoverflow.com/a/53666584/2245623).

- Some Buildroot options which should have been hardcoded (e.g.,
  persisting of OpenSSH server keys) are now unconditional. Once the
  czechlight-cfg-fs package is selected, these cannot be disabled.

- We no longer nuke the entire sysrepo SW stack upon a failure. If,
  e.g., cla-sysrepo fails, this does not bring the NETCONF server down
  anymore, for example. (We might restore the PartOf=... if this proves
  to be a problem.)

- There's a bunch of YANG rules which try to make sure that the NETCONF
  server is reasonably configured. Previously, we would simply
  reprovision the server config at the next boot, now we have nice error
  messages at the YANG level already.

- The RESTCONF server has been vastly improved, and some changes all
  over that SW stack were made. Also, cleaned up the packaging a bit
  (especially those parts which would be a PITA to split into extra
  commits due to conflicts everywhere -- sorry).

- Use `jq` patterns for JSON filtering (to make sure that we can provide
  an exclusion filter for the crypto material). Also, since we drop
  anything but v9 configs now, let's drop the existing migration tests
  (there's nothing to test, really).

Change-Id: Id1bb5b9ee66c6deb7a886289e4d768ce3ff7c9b2
Depends-on: https://gerrit.cesnet.cz/plugins/gitiles/github/buildroot/buildroot/+/refs/heads/cesnet/2024-09-06
Depends-on: https://gerrit.cesnet.cz/c/CzechLight/gammarus/+/7570