This works better in the current case when master branch with the most
recent 5.7.0 tag got merged into the 6.0 branch.  In that case we do not
want the most recent *generally* reachable tag, and git-describe has
probably good logic for this (using first-parent jumps I guess).

This leverages Environments on GitLab to expose different versions of
Knot Resolver docs. The `docs:build` job builds the documentation and
exposes it via job artifacts. Then `docs:develop` (for branches) and
`docs:release` (for tags) take these artifacts and expose them via an
Environment link (an example of this in action may be seen at
[https://gitlab.nic.cz/ostava/knot-resolver/-/environments]).

There is also an optional, manually runnable `docs:public` job, which,
when run, propagates the documentation to the main GitLab Pages of the
project (e.g. [https://knot.pages.nic.cz/knot-resolver]) - this will
probably be mostly used for the latest release, although this setup
pretty much allows us to swap it for whatever version we like at any
time.

Aleš Mrázek authored 1 year ago and Vladimír Čunát committed 1 year ago

- pyproject.tom: poethepoet for python3.7

Oto Šťáva authored 1 year ago and Vladimír Čunát committed 1 year ago

This updates `poethepoet` to version `^0.22.0`
(https://github.com/nat-n/poethepoet/releases/tag/v.0.22.0), which
allows tasks to have a working directory different from the project
path.

This breaks the `poe` script on Python `<3.8`, but discussions took
place on Slack where we came to the conclusion that this is fine. The
script is meant for developers only and does not affect end users on
systems that do not provide newer Python versions.

Oto Šťáva authored 1 year ago and Vladimír Čunát committed 1 year ago

Previously this penalization was only triggered if the remote server
closed TCP.  Now it's extended to us closing it when the server
(only) sends back some nonsense.  At least for the cases which I could
see immediately.

That's just three trivial one-line additions; the rest is refactoring.

Adapted to 6.0 from commit 6468ab22 by Oto Šťáva <oto.stava@nic.cz>

Co-Authored-By: Vladimír Čunat <vladimir.cunat@nic.cz>

release 5.7.0

See merge request !1448

This was changed long ago in bb6e2de2
(and 6.0 docs was correct already)

manager: use self._type in KresID.__eq__()

See merge request !1449

Fixes a case where a GC and KRESD KresID with the same `self._id` would
be considered equal. Said behaviour breaks listing of all running
subprocesses, where `kresd0` would be missing, because it has the same
`self._id` as `cache-gc`.

Vladimír Čunát authored 1 year ago and Aleš Mrázek committed 1 year ago

Previously this penalization was only triggered if the remote server
closed TCP.  Now it's extended to us closing it when the server
(only) sends back some nonsense.  At least for the cases which I could
see immediately.

That's just three trivial one-line additions; the rest is refactoring.

Mentioning just answers is misleading.  Padding is very important
for queries as well; for us that applies during forwarding over TLS.

Also describe /tls/auto_discovery as experimental in the
configuration  schema.  It's a rather dead experiment from long ago.
I hope it can't be confused with more recent things like
https://datatracker.ietf.org/doc/draft-ietf-dprive-unilateral-probing/

Closes #808

We had this behavior in 5.x.
Lua level: affects hints.set() and hints['name'] and hints.add_hosts()
YAML level: /local-data/addresses and /local-data/addresses-files

I considered various approaches when writing this.  This one won because
in /etc/hosts like files a name can be repeated with arbitrary lines
in between, and users can reasonably expect it to collect all addresses.

manager/poetry.lock: remove

See merge request !1445

Oto Šťáva authored 1 year ago and Aleš Mrázek committed 1 year ago

Lockfiles are generally encouraged to be committed into VCS to ensure
reproducible builds on all machines. Our situation is slightly
different - we are more interested in supporting a wide range of setups
on a wide range of systems.

Our `poetry.lock` also contained some outdated library versions that had
CVEs reported on them, and GitHub did not like that, so this
incidentally also solves that problem for us. Since the lock file was
only used during development, this will not affect end users in any way
(runtime dependencies are managed by each distro's package manager).

Record TTL is always meant in seconds by default.
More precise resolution does not exist.