An escape char is needed to prevent "$ nix" from being evaluated and
expanded to an empty string. With this change the resulting text is
$ nix-env -qaP | grep wget
instead of
-env -qaP | grep wget
systemd escaping rules translate this into a string containing '\'
which is treated by some code paths as quoted, and by others as unquoted
causing the affected units to fail.
Currently, the restartTriggers are abusing the systemd unit file in that
the cfg.carbon.config/storageAggregation/... option text is pasted into
the unit file. Even though this sort-of works (the service is restarted
if the config changes) this causes systemd to print error messages about
invalid sections (rightfully so!).
The correct use of restartTriggers is to list storage paths, which is
what this change does. If any of the
cfg.carbon/config/storageAggregation/... options change, configDir will
get a new hash. It is not as "fine grained" as the current version, but
it is not abusing the interface.
Also, remove unneeded 'waitress' in one of the restartTriggers, because
it is already listed as part of the service config.
When formatting a nixos usb using my windows machine I noticed that the
disk labeling using periods was not compatible with my linux kernel /
udev recognition. When labeling a volume NIXOS_14.10 under Windows, it would
show up as NIXOS_14_10 on NixOS. This meant that /dev/root would never come
up at boot time, preventing the livecd from starting.
This patch works around this issue by eliminating any special characters
other than underscores. The previous versioning didn't seem all that
useful, especially when consdering there are many different version of
the year.month named iso.
graphitePort must point to the port that carbon-cache listens on, not
the graphite webUI port.
With this change I finally got data from statsd to graphite.
It's "aggregation" with two 'g's.
Fixes this:
carbon-cache[9363]: [console] /nix/store/drxq4jj92sjk3cjik2l4hnsndbray3i4-graphite-config/storage-aggregation.conf not found, ignoring.
Attempt to fix race condition in installer tests, especially the
grub1 test.
The latter was failing when running "parted /dev/sda ..." because
/dev/sda didn't exist yet.
