The current kubernetes module only allows you to set a single DNS
resolver for the kubelet. Historically, this has not mattered as the
value was passed to a cli argument as a string and as per the kubelet's
configuration parsing mechanism, multiple values could be passed as a
comma-delimited string. However, recently, the module was refactored to
make configure kubernetes components via configuration files rather than
the deprecated command-line arguments. These files more strongly-typed
than CLI arguments and to pass multiple values, one must define a list
in the file. When this change was made, an incorrect assumption was made
that only a single DNS server could be specified and forced a
single-item list into this configuration file. We need to introduce a
breaking change to the module in order to allow the user to supply their
own list with however many dns resolvers they wish to use.
Every now and then, kubernetes adds new configuration parameters to the
kubelet configuration. Since this is defined using a nix attrset which
is then converted to json/yaml, it would be nice to have an escape hatch
similar to the extraOpts one that exists for additional CLI arguments.
The typical use case would be to configure new settings before they are
officially supported in the nixos module.
Strongly inspired by the forgejo counterpart[1], for the following
reasons:
* The feature is broken with the current module and crashes on
authentication with the following stacktrace (with a PAM service
`gitea` added):
server # Stack trace of thread 1008:
server # #0 0x00007f3116917dfb __nptl_setxid (libc.so.6 + 0x8ddfb)
server # #1 0x00007f3116980ae6 setuid (libc.so.6 + 0xf6ae6)
server # #2 0x00007f30cc80f420 _unix_run_helper_binary (pam_unix.so + 0x5420)
server # #3 0x00007f30cc8108c9 _unix_verify_password (pam_unix.so + 0x68c9)
server # #4 0x00007f30cc80e1b5 pam_sm_authenticate (pam_unix.so + 0x41b5)
server # #5 0x00007f3116a84e5b _pam_dispatch (libpam.so.0 + 0x3e5b)
server # #6 0x00007f3116a846a3 pam_authenticate (libpam.so.0 + 0x36a3)
server # #7 0x00000000029b1e7a n/a (.gitea-wrapped + 0x25b1e7a)
server # #8 0x000000000047c7e4 n/a (.gitea-wrapped + 0x7c7e4)
server # ELF object binary architecture: AMD x86-64
server #
server # [ 42.420827] gitea[897]: pam_unix(gitea:auth): unix_chkpwd abnormal exit: 159
server # [ 42.423142] gitea[897]: pam_unix(gitea:auth): authentication failure; logname= uid=998 euid=998 tty= ruser= rhost= user=snenskek
It only worked after turning off multiple sandbox settings and adding
`shadow` as supplementary group to `gitea.service`.
I'm not willing to maintain additional multiple sandbox settings for
different features, especially given that it was probably not used for
quite a long time:
* There was no PR or bugreport about sandboxing issues related to
PAM.
* Ever since the module exists, it used the user `gitea`, i.e. it had
never read-access to `/etc/shadow`.
* Upstream has it disabled by default[2].
If somebody really needs it, it can still be brought back by an overlay
updating `tags` accordingly and modifying the systemd service config.
[1] 07641a91c9
[2] https://docs.gitea.com/usage/authentication#pam-pluggable-authentication-module
This splits a dev output to make the default output not depend on any
build dependencies anymore. This also avoids removing references from
pgxs' Makefile this way, which should, at least theoretically, be good
to build extensions via pgxs, making sure they use the same tooling.
ecpg is the "embedded SQL C preprocessor", which is certainly a dev
tool.
Most important, for closure size anyway, is to move pg_config to the dev
output, since it retains paths to all the other outputs.
The only thing with references to the dev output remaining is then the
postgres binary itself. It contains all the output paths, because it
shows those in the pg_config system view. There is no other way than
to nuke those references to avoid circular dependencies between outputs
- and blowing up closure size again.
