Build: use `nix path-info` to test for build output availability #1352

Ma27 · 2024-01-26T17:00:08Z

Hacky workaround for NixOS/nix#9859, but after I managed to fix the issue on my end, I decided I could still submit it against upstream.

When using the Perl bindings, this usually takes the wrong store, i.e. auto even if store_uri is a flat-file binary cache.

See also NixOS/nix#9859

Using fork/exec here on purpose to make sure that the output of nix path-info isn't spammed into stderr. This could probably done with system("foo &>/dev/null"), but I didn't want to worry about shell injection issues then. And finally, I can't seem to silence STDERR for system with a list as argument. Not a Perl dev though, so not sure if there's a better way.

When using the Perl bindings, this usually takes the wrong store, i.e. `auto` even if `store_uri` is a flat-file binary cache. See also NixOS/nix#9859 Using fork/exec here on purpose to make sure that the output of `nix path-info` isn't spammed into stderr. This could probably done with `system("foo &>/dev/null")`, but I didn't want to worry about shell injection issues then. And finally, I can't seem to silence STDERR for `system` with a list as argument. Not a Perl dev though, so not sure if there's a better way.

Ericson2314 · 2024-01-26T17:31:02Z

I don't quite understand the things you are saying in that issue, but I soon as well just add the missing functionality. We have a nix-next branch so we can try it out right away.

lheckemann · 2024-01-26T17:34:20Z

Does this even work when using the nixos module? hydra-server.service doesn't seem to have nix on its PATH. I also wonder if this is an appropriate approach given the other ways in which the nix command could fail (experimental-features = nix-command missing?) and with no distinction drawn between that and the path not being valid.

Ma27 · 2024-01-26T17:43:00Z

It definitely works since I just realized I forgot to silence stdout and now I get paths printed to hydra-server's journal whenever I open a build view. ~~Not sure why yet 🙃~~

Nix is part of hydra-server's env:

hydra root ~ # cat /nix/store/d4w2gj8y1anymj9lsm3bih4v95f03bw3-hydra-0.1.20240126.1d1ec0b/bin/hydra-server|grep nix-
PATH=${PATH/':''/nix/store/7cibfivz81rpv8jhr53fj5hjlidpnykb-nix-2.19.3/bin'':'/':'}
PATH='/nix/store/7cibfivz81rpv8jhr53fj5hjlidpnykb-nix-2.19.3/bin'$PATH
export NIX_RELEASE='nix-2.19.3'

could fail (experimental-features = nix-command missing?)

Fair. Should probably use a nix-* equivalent then 🤷

Ericson2314 · 2024-01-26T17:51:41Z

I am looking at the perl binding stuff.

…well `nix-store -q` (and `--outputs`) doesn't seem to complain if a path doesn't exist in a store anymore, so let's do it that way. This approach works fine on Nix 2.3 as well, it only warns about `experimental-features` being an unknown option, but that's silenced along with the rest of stderr.

Ericson2314 · 2024-01-26T22:25:23Z

I made NixOS/nix#9863 to generalize the Perl bindings so we can fix this the right way. It would be nice to replace this with PR against nix-next using that branch.

I think we can basically do a big sed to update Hydra: it's fine to call openStore() again and again (without an argument, that is) because there is caching.

Implements support for Nix's new Perl bindings[1]. The current state basically does `openStore()`, but always uses `auto` and doesn't support stores at other URIs. Even though the stores are cached inside the Perl implementation, I decided to instantiate those once in the Nix helper module. That way store openings aren't cluttered across the entire codebase. Also, there are two stores used later on - MACHINE_LOCAL_STORE for `auto`, BINARY_CACHE_STORE for the one from `store_uri` in `hydra.conf` - and using consistent names should make the intent clearer then. This doesn't contain any behavioral changes, i.e. the build product availability issue from NixOS#1352 isn't fixed. This patch only contains the migration to the new API. [1] NixOS/nix#9863

Replaces / Closes NixOS#1352 Consider the following setup: * `store_uri` points to a "special" store, e.g. `s3://` or `file://`. * Hydra runs inside e.g. an nspawn container and exclusively builds on remote systems. Then, build products are never in the nspawn's `/nix/store`. This in turn means that `$MACHINE_LOCAL_STORE->isValidPath` is always `false` (or `0` 🤷) for output paths and thus Hydra wrongly claims that the output got garbage collected. Instead, use the BINARY_CACHE_STORE to look for the availability in the correct store. Everything touching the `drv` rather than the output paths still uses MACHINE_LOCAL_STORE: this is because `hydra-eval-jobs` does `openStore()` on `auto`, so the derivations end up there rather than on the BINARY_CACHE_STORE.

Implements support for Nix's new Perl bindings[1]. The current state basically does `openStore()`, but always uses `auto` and doesn't support stores at other URIs. Even though the stores are cached inside the Perl implementation, I decided to instantiate those once in the Nix helper module. That way store openings aren't cluttered across the entire codebase. Also, there are two stores used later on - MACHINE_LOCAL_STORE for `auto`, BINARY_CACHE_STORE for the one from `store_uri` in `hydra.conf` - and using consistent names should make the intent clearer then. This doesn't contain any behavioral changes, i.e. the build product availability issue from NixOS#1352 isn't fixed. This patch only contains the migration to the new API. [1] NixOS/nix#9863

Replaces / Closes NixOS#1352 Consider the following setup: * `store_uri` points to a "special" store, e.g. `s3://` or `file://`. * Hydra runs inside e.g. an nspawn container and exclusively builds on remote systems. Then, build products are never in the nspawn's `/nix/store`. This in turn means that `$MACHINE_LOCAL_STORE->isValidPath` is always `false` (or `0` 🤷) for output paths and thus Hydra wrongly claims that the output got garbage collected. Instead, use the BINARY_CACHE_STORE to look for the availability in the correct store. Everything touching the `drv` rather than the output paths still uses MACHINE_LOCAL_STORE: this is because `hydra-eval-jobs` does `openStore()` on `auto`, so the derivations end up there rather than on the BINARY_CACHE_STORE.

Implements support for Nix's new Perl bindings[1]. The current state basically does `openStore()`, but always uses `auto` and doesn't support stores at other URIs. Even though the stores are cached inside the Perl implementation, I decided to instantiate those once in the Nix helper module. That way store openings aren't cluttered across the entire codebase. Also, there are two stores used later on - MACHINE_LOCAL_STORE for `auto`, BINARY_CACHE_STORE for the one from `store_uri` in `hydra.conf` - and using consistent names should make the intent clearer then. This doesn't contain any behavioral changes, i.e. the build product availability issue from NixOS#1352 isn't fixed. This patch only contains the migration to the new API. [1] NixOS/nix#9863

Replaces / Closes NixOS#1352 Consider the following setup: * `store_uri` points to a "special" store, e.g. `s3://` or `file://`. * Hydra runs inside e.g. an nspawn container and exclusively builds on remote systems. Then, build products are never in the nspawn's `/nix/store`. This in turn means that `$MACHINE_LOCAL_STORE->isValidPath` is always `false` (or `0` 🤷) for output paths and thus Hydra wrongly claims that the output got garbage collected. Instead, use the BINARY_CACHE_STORE to look for the availability in the correct store. Everything touching the `drv` rather than the output paths still uses MACHINE_LOCAL_STORE: this is because `hydra-eval-jobs` does `openStore()` on `auto`, so the derivations end up there rather than on the BINARY_CACHE_STORE.

Implements support for Nix's new Perl bindings[1]. The current state basically does `openStore()`, but always uses `auto` and doesn't support stores at other URIs. Even though the stores are cached inside the Perl implementation, I decided to instantiate those once in the Nix helper module. That way store openings aren't cluttered across the entire codebase. Also, there are two stores used later on - MACHINE_LOCAL_STORE for `auto`, BINARY_CACHE_STORE for the one from `store_uri` in `hydra.conf` - and using consistent names should make the intent clearer then. This doesn't contain any behavioral changes, i.e. the build product availability issue from #1352 isn't fixed. This patch only contains the migration to the new API. [1] NixOS/nix#9863

Ma27 mentioned this pull request Feb 7, 2024

Update perl bindings for nix#9863 #1359

Merged

Ma27 mentioned this pull request Feb 12, 2024

Use binary cache at store_uri for certain operations #1360

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Build: use `nix path-info` to test for build output availability #1352

Build: use `nix path-info` to test for build output availability #1352

Ma27 commented Jan 26, 2024

Ericson2314 commented Jan 26, 2024

lheckemann commented Jan 26, 2024

Ma27 commented Jan 26, 2024 •

edited

Loading

Ericson2314 commented Jan 26, 2024

Ericson2314 commented Jan 26, 2024

Build: use nix path-info to test for build output availability #1352

Are you sure you want to change the base?

Build: use nix path-info to test for build output availability #1352

Conversation

Ma27 commented Jan 26, 2024

Ericson2314 commented Jan 26, 2024

lheckemann commented Jan 26, 2024

Ma27 commented Jan 26, 2024 • edited Loading

Ericson2314 commented Jan 26, 2024

Ericson2314 commented Jan 26, 2024

Build: use `nix path-info` to test for build output availability #1352

Build: use `nix path-info` to test for build output availability #1352

Ma27 commented Jan 26, 2024 •

edited

Loading