nix-output-monitor
Pipe your nix-build output through the nix-output-monitor (aka nom) to get additional information while building.
While your build runs nom will draw something like this at the bottom of your build log:
(note that to reduce clutter nom only shows timers over 1s build or download time.)
Status
This was an experimental fun project, which proved to be useful to quite a lot of people. The purpose of it is to write something fun and useful in Haskell. You are free and very welcome to contribute feedback, issues or PRs. I do not commit to maintain this project over a long time period, but it doesn’t look, like I am stopping anytime soon.
By now, nom is quite fully featured with support for nix v1 commands (e.g. nix-build
) and nix v2 command (e.g. nix build
).
Issues and pull requests welcome under https://github.com/maralorn/nix-output-monitor.
Call for screenshots: I would be very glad to receive screenshot of your nom use. I never know how it looks on your machine and seeing how you use nom might help me in making the UX better.
Support
If your question is not answered in this README you can ask it in #nix-output-monitor:maralorn.de on matrix or open an issue here on github.
Installing
- nixpkgs: nom is in nixpkgs. Just install
pkgs.nix-output-monitor
in the usual way. You might want to install it from nixos-unstable to get the newest version. - cabal: Install
cabal-install
and runcabal install
in the checked out repo. - nix: or run
nix-build
ornix-env
or include thedefault.nix
of this repo in your nixos config.
Running
The Easy Way
Warning: The displayed build tree might be incomplete with new-style commands like nix build
for nix versions <2.10.
The nom
binary (starting from version 2.0) behaves as a nix
drop in, with much more colorful output, for the following commands:
nom build <args>
: Behaves like nix build <args>
.
nom shell <args>
: Behaves like nix shell <args>
.
nom develop <args>
: Behaves like nix develop <args>
.
The latter two commands work by calling nix shell
or nix develop
twice, the first time with overridden --run exit
and monitoring the output, the second time passing output through to the user. This will incur a performance cost by doubling eval time.
Furthermore when called via the corresponding provided symlinks, nom is also a drop-in for the following commands:
nom-build <args>
: Behaves like nix-build <args>
.
nom-shell <args>
: Behaves like nix-shell <args>
.
All aliases internally use the json-based approach (see next section) and propagate error codes. If you want nom support for other nix commands please open an issue.
The Flexible Way
JSON based
nix-build --log-format internal-json -v |& nom --json
Warning: Don‘t forget to redirect stderr. That's what the &
, does.
Human readable log parsing
It his highly recommended to always append --log-format internal-json -v
(or use the above mentioned aliases.) and call nom
with --json
. That will give you much more informative output.
If you are in a situation, where you can‘t use the json based nix output you can still use
nix-build |& nom
Warning: Don‘t forget to redirect stderr. That's what the &
, does.
This has the advantage to also work with other commands like nixos-rebuild
or home-manager
, where it is not trivial to pass in the --log-format internal-json -v
flag. nom will pass everything it reads through, if it does not understand it. This makes it ideal to attach it to scripts which output more then just nix
output.
Preserving Colored Text
Colored text will work as expected in json-mode.
In human-readable log mode you can preserve the color of the redirected text by using the unbuffer
command from the expect
package.
unbuffer nix-build |& nom
Explanation
Legend
Nom tries to convey information via symbols and colors
⏵
, yellow: running builds✔
, green: completed builds⏳︎︎︎
, blue: planned builds⚠
, red: failed builds↓ ⏵
, yellow: running downloads↑ ⏵
, yellow: running uploads↓ ✔
, green: completed downloads↑ ✔
, green: completed uploads↓ ⏳︎︎︎
, blue: waiting downloads∅
: a moving average over past builds of this derivation⏱︎
: running time∑
: a summary over all packages and hosts
If you can‘t see all icons you maybe need another terminal font.
I recommend any font from pkgs.nerdfonts
e.g. "JetBrainsMono Nerd Font"
.
Also different terminals might work differently well. I recommend: pkgs.foot
.
How to Read the Dependency Graph
- Every entry in the nom tree stands for one derivation.
- Children of a node are direct dependencies.
- nom will try to show you the most relevant part of the dependency tree, roughly aiming to fill a third of your terminal
- No build will be printed twice in the tree, it will only be shown for the lower most dependency.
- nom will do it’s best to print all running or failed builds, downloads and uploads, but it does not print every direct child of a node.
- Use the colors from above to read the summary
Example Runs
An example with a lot of downloads:
Implementation
Right now nom uses four sources of information:
- The parsed nix-build output (json or human-readable)
- it checks if build results exist in the nix-store (only in human-readable mode)
- it queries
.drv
files for information about theout
output path. - It caches build times in
$XDG_CACHE_HOME/nix-output-monitor/build-reports.csv
.
Limitations
- This will fail in unexpected and expected ways.
- Luckily I don‘t think this program screws up anything more than your terminal.
- remote builds will sometimes be shown as running even when they are actually still waiting for uploads or downloads. This is how nix reports it.
- Terminal clearing and reprinting is brittle. It might fail with your terminal or terminal width. But at this point I‘ve invested some effort to make it usable.
- This program also makes assumptions like your nix-store is at "/nix/store".
For human-readable log parsing mode:
- nix-output-monitor receives most it's information from parsing nix-build output. The parser might be to strict or to loose for use cases I didn‘t think of. Then the numbers displayed will be off!
- nix-build does not show info when a download or upload is finished, so we currently cannot differentiate between started and completed downloads.
- For completed build detection we assume that every derivation has an output called "out".