fix(deployment): flake import

[Malix-Labs]

Problems

Importing flakes is starting to get outscaled

This has started causing CI issues.
For example, https://github.com/NixOS/nixos-search/actions/workflows/import-to-elasticsearch.yml fails repetitively which makes https://search.nixos.org/flakes unable to be updated

Errors

Disk space seems to be the culprit since errors seems to mostly be

System.IO.IOException: No space left on device

(except the first failing of the failing streak)

And some other are

The job has exceeded the maximum execution time of 6h0m0s

Subject

• first failing of the failing streak: Hourly import to Elasticsearch #30802
• last succeeding before the failing streak: Hourly import to Elasticsearch #30801

nixos-search/.github/workflows/import-to-elasticsearch.yml

Lines 116 to 120 in 701750c

	- name: Import ${{ matrix.group }} group
	if: github.repository_owner == 'NixOS'
	shell: sh
	run: |
	nix run --accept-flake-config .#flake-info -- --push --elastic-schema-version=$(cat ./VERSION) group ./flakes/${{ matrix.group }}.toml ${{ matrix.group }}

That step, and especially for the "manual" matrix group (= "Import manual group" step) is the failing one

Proposed Solutions

Ordered by (@Malix-Labs's) preference

Only one implementation would be sufficient, but a combination could help in case it isn't

1. Enhancing CI algorithm by looping for each flake entry and checking for garbage collecting needs before iterating or by using periodic disk monitoring (which maybe would need to be done in flake-info
2. Keep a single file but split its evaluation into chunks (preferably of ~ equal hardware requirements, maybe by pre-evaluating each flake hardware requirements before the "Import ${{ matrix.group }} group" / "Import manual group" step)
3. Use GitHub large runners
4. Splitting flakes entry into arbitrary files (Split flake lists into groups #1037)

References

• enhance(ci): Workflow "Check Flake Groups" - Increasing Time to Complete #1027 (concerns are somewhat related)

[Malix-Labs]

@raboof the arbitrary file split proposed in #1037 I think is a leaky workaround because as flakes entry grows, more arbitrary file split would be necessary

---

See #1033 (comment) for more

[raboof]

@raboof the arbitrary file split proposed in #1037 I think is a leaky workaround because as flakes entry grows, more arbitrary file split would be necessary

yes

I see 3 other options

nixos-search/.github/workflows/import-to-elasticsearch.yml

Lines 116 to 120 in 701750c
- name: Import ${{ matrix.group }} group
if: github.repository_owner == 'NixOS'
shell: sh
run: |
nix run --accept-flake-config .#flake-info -- --push --elastic-schema-version=$(cat ./VERSION) group ./flakes/${{ matrix.group }}.toml ${{ matrix.group }}

That step, and especially for the "manual" matrix group (= "Import manual group" step) is the failing one (e.g. this instance)

It seems like it could be enhanced by looping for each flake entry and checking if there is a need to garbage collect before iterating (or using periodic disk monitoring); maybe that would need to be done in flake-info

Another alternative would be keep a single file but to split it into chunks (preferably of ~ equal hardware requirements, maybe by pre-evaluating each flake hardware requirements before the "Import ${{ matrix.group }} group" / "Import manual group" step

Another option is to use GitHub large runners

As a side effect, the ability to evaluate partial files might also contributes to #1027, as concerns are somewhat related

Do you want to create new issues for these potential improvements or change the title of this issue?

[Malix-Labs]

#1033 (comment) from @raboof :

Do you want to create new issues for these potential improvements or change the title of this issue?

I just turned this issue into a general enhancement about the flake import in CI, including its title and body

---

@raboof what do you think of the "Proposed Solutions" in the body, and particularly the current order of preference?