NixOS Infrastructure

Sender	Message	Time
1 Feb 2026
BMG	`~106K` affected paths, with `~95K` of those of the form `nixos-test-driver-%`: D select count() FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet'; ┌──────────────┐ │ count_star() │ │ int64 │ ├──────────────┤ │ 105944 │ └──────────────┘ D select count() FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet' where pname like 'nixos-test-driver-%'; ┌──────────────┐ │ count_star() │ │ int64 │ ├──────────────┤ │ 95032 │ └──────────────┘	14:34:47
Mic92	Looks nixos tests to me. Should be safe to drop	14:34:56
BMG	The other 10k look to be nixos releases: `D select count(*) FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet' where pname like 'nixos-%.%'; ┌──────────────┐ │ count_star() │ │ int64 │ ├──────────────┤ │ 10912 │ └──────────────┘`	14:36:17
BMG	Of the form `nixos-19.03.172539.72c2c58356e`	14:36:28
Mic92	I think that might be actually nixos toplevel closures from tests	14:37:47
Mic92	Is it possible to just include the references as well in the deletion tests?	14:38:46
Mic92	* Is it possible to just include the references as well in the deletion set	14:38:58
BMG	The archeologist account I have for AWS, that is definitely read -only ? I want to test the `--dry-run` against the bucket but I want to be really sure it can't affect anything yet	14:39:14
BMG	I should be able to	14:39:35
Mic92	I can check	14:39:36
Mic92	@brian:bmcgee.ie: only list/get	14:40:30
BMG	ok cool	14:40:38
BMG	Mic92: fixed a bug and improved the progress logging: `{ "dry_run": true, "targets": { "nar_infos": 150581 }, "removed": { "nars": 150581, "nar_infos": 150581, "errors": 0 }, "missing_in_s3": { "nars": 0, "nar_infos": 0 } }`	15:46:09
BMG	I'm going to capture those extra references in a separate file that you can just run through the simple gc same as the nixos images one.	15:46:34
Mic92	Alright	16:07:48
BMG	Mic92: I pushed an updated version of the dangling refs parquet file which you should be able to run through simple gc to remove them just like with nixos images. I also locally tweaked the dangling refs script to load those first order dangling refs into the deletion set and the result was no more dangling refs found: `================================================== SUMMARY ================================================== Entries with dangling refs: 0 No dangling references found`	16:55:49
BMG	So removing the nixos images and the extra dangling refs results in store paths with missing references due to the removals	16:56:26
BMG	* So removing the nixos images and the extra dangling refs results in no store paths with missing references due to the removals	16:56:35
BMG	So to be clear, `narinfos-nixos-images-2026-01-06T01-13Z.parquet` is the proposed deletion, `narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet` is anything that referred to the entries in the first file.	16:57:45
BMG	I'm gonna test with a dry run to be sure there's no issues with the dangling refs format	16:59:07
BMG	Had to fix a schema issue, pushed an update to the dataset and just tested it with a dry run: `{ "dry_run": true, "targets": { "nar_infos": 105944 }, "removed": { "nars": 105944, "nar_infos": 105944, "errors": 0 }, "missing_in_s3": { "nars": 0, "nar_infos": 0 } }`	17:07:55
2 Feb 2026
Arian	If you are nervous about correctness we could enable bucket versioning + a Lifecycle rule that moves deleted objects into glacier	08:30:17
Arian	``` { "Rules": [ { "Id": "archive-deleted-objects", "Status": "Enabled", "NoncurrentVersionTransitions": [ { "NoncurrentDays": 0, "StorageClass": "GLACIER_IR" } ] } ] }```	08:30:25
Arian	@joerg:thalheim.io are you still in Brussels today or tomorrow?	08:31:22
Mic92	Arian: I am.	08:56:48
Mic92	I am on the guix days. Maybe we can meet in the Numtide venue in the evening?	08:57:25
Mic92	* I am on the guix days the next two days. Maybe we can meet in the Numtide venue in the evening?	08:57:34
Mic92	I actually wanted to ask regarding the bucket versioning, because the implications are a bit unclear to me yet.	08:58:04
Mic92	okay, we might than actually want to remotve the dangling refs dataset before we run the nixos images one.	09:00:44
Mic92	* Arian: I am. I will leave on Wednesday	09:00:50

~106K affected paths, with ~95K of those of the form nixos-test-driver-%:

D select count(*) FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet';
┌──────────────┐
│ count_star() │                                                                                                                                                                                                                                    
│    int64     │                                                                                                                                                                                                                                    
├──────────────┤                                                                                                                                                                                                                                    
│    105944    │                                                                                                                                                                                                                                    
└──────────────┘                                                                                                                                                                                                                                    
D select count(*) FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet' where pname like 'nixos-test-driver-%';
┌──────────────┐
│ count_star() │                                                                                                                                                                                                                                    
│    int64     │                                                                                                                                                                                                                                    
├──────────────┤                                                                                                                                                                                                                                    
│    95032     │                                                                                                                                                                                                                                    
└──────────────┘

14:34:47

Mic92

Looks nixos tests to me. Should be safe to drop

14:34:56

BMG

The other 10k look to be nixos releases:

D select count(*) FROM 'datasets/narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet' where pname like 'nixos-%.%';
┌──────────────┐
│ count_star() │                                                                                                                                                                                                                                    
│    int64     │                                                                                                                                                                                                                                    
├──────────────┤                                                                                                                                                                                                                                    
│    10912     │                                                                                                                                                                                                                                    
└──────────────┘

14:36:17

BMG

Of the form nixos-19.03.172539.72c2c58356e

14:36:28

Mic92

I think that might be actually nixos toplevel closures from tests

14:37:47

Mic92

Is it possible to just include the references as well in the deletion tests?

14:38:46

Mic92

* Is it possible to just include the references as well in the deletion set

14:38:58

BMG

The archeologist account I have for AWS, that is definitely read -only ? I want to test the --dry-run against the bucket but I want to be really sure it can't affect anything yet

I should be able to

I can check

@brian:bmcgee.ie: only list/get

ok cool

Mic92: fixed a bug and improved the progress logging:

{
  "dry_run": true,
  "targets": {
    "nar_infos": 150581
  },
  "removed": {
    "nars": 150581,
    "nar_infos": 150581,
    "errors": 0
  },
  "missing_in_s3": {
    "nars": 0,
    "nar_infos": 0
  }
}

15:46:09

BMG

I'm going to capture those extra references in a separate file that you can just run through the simple gc same as the nixos images one.

Alright

Mic92: I pushed an updated version of the dangling refs parquet file which you should be able to run through simple gc to remove them just like with nixos images.

I also locally tweaked the dangling refs script to load those first order dangling refs into the deletion set and the result was no more dangling refs found:

==================================================
SUMMARY
==================================================
Entries with dangling refs: 0
No dangling references found

16:55:49

BMG

So removing the nixos images and the extra dangling refs results in store paths with missing references due to the removals

16:56:26

BMG

* So removing the nixos images and the extra dangling refs results in no store paths with missing references due to the removals

16:56:35

BMG

So to be clear, narinfos-nixos-images-2026-01-06T01-13Z.parquet is the proposed deletion, narinfos-nixos-images-dangling-refs-2026-01-06T01-13Z.parquet is anything that referred to the entries in the first file.

16:57:45

BMG

I'm gonna test with a dry run to be sure there's no issues with the dangling refs format

16:59:07

BMG

Had to fix a schema issue, pushed an update to the dataset and just tested it with a dry run:

{
  "dry_run": true,
  "targets": {
    "nar_infos": 105944
  },
  "removed": {
    "nars": 105944,
    "nar_infos": 105944,
    "errors": 0
  },
  "missing_in_s3": {
    "nars": 0,
    "nar_infos": 0
  }
}

17:07:55

Arian

If you are nervous about correctness we could enable bucket versioning + a Lifecycle rule that moves deleted objects into glacier

08:30:17

Arian

``` { "Rules": [ { "Id": "archive-deleted-objects", "Status": "Enabled", "NoncurrentVersionTransitions": [ { "NoncurrentDays": 0, "StorageClass": "GLACIER_IR" } ] } ] }```

08:30:25

Arian

@joerg:thalheim.io are you still in Brussels today or tomorrow?

Arian: I am.

I am on the guix days. Maybe we can meet in the Numtide venue in the evening?

08:57:25

Mic92

* I am on the guix days the next two days. Maybe we can meet in the Numtide venue in the evening?

08:57:34

Mic92

I actually wanted to ask regarding the bucket versioning, because the implications are a bit unclear to me yet.

08:58:04

Mic92

okay, we might than actually want to remotve the dangling refs dataset before we run the nixos images one.

09:00:44

Mic92

* Arian: I am. I will leave on Wednesday

09:00:50

Load older messages

Show newer messages

	NixOS Infrastructure	383 Members
	Next Infra call: 2024-07-11, 18:00 CEST (UTC+2) \| Infra operational issues backlog: https://github.com/orgs/NixOS/projects/52 \| See #infra-alerts:nixos.org for real time alerts from Prometheus.	116 Servers