feat: rotating db backups (keep newest N) + client url nai.kagaku.space #4

Merged

bjoern merged 2 commits from feat/backup-rotation-and-client-url into main 2026-09-16 00:11:52 +02:00

Conversation 3 Commits 2 Files changed 6 +267 -119

matikane commented 2026-09-16 00:01:09 +02:00

Member

Copy link

Two changes: server-side rotating backups + client server URL.

1. Rotating db backups (was: growing forever — verified)

DbBackupService creates backup_{timestamp}.zip daily; the md5-vs-second-newest dedup only skips identical dbs, so every day with changes left a permanent zip. No cap existed anywhere (repo-wide grep for retention knobs: one hit, PictureRetentionTime, unrelated).

Design (as agreed):

• Utils/BackupRotation.Prune(backupPath, keepCount) — one public static class, XML docs. Keeps the newest keepCount zips, deletes the rest. Only backup_*.zip matched — foreign files in the dir are untouchable. Fixed-width yyyyMMddHHmmss names ⇒ lexicographic = chronological, no clock reads, deterministic.
• DbBackupService.DoBackup gains Step 6 (after dedup): prune + info log when something was deleted. Constructor now takes IOptions<ServicesConfig> (registered already).
• Knob: ServicesConfig.DbBackupRetentionCount, default 14, 0 = keep everything (escape hatch). No appsettings change needed — the default applies unless overridden.

Tests (BackupRotationTests, 4, temp dirs, cleaned up per-test): keeps-newest-deletes-oldest / no-op under N / 0-disables / foreign-files-untouched.

Mutation probes (all restored, suite green after):

• ordering → OrderByDescending ⇒ Prune_KeepsNewestN_AndDeletesOldest red
• escape hatch removed ⇒ Prune_Zero_DisablesPruning red
• pattern guard dropped (GetFiles(path)) ⇒ Prune_OnlyTouchesBackupPattern red — this one initially survived the probe (test not tight enough); tightened it (3 backups + foreign file, asserts exact deleted count) and re-verified it now dies. Every guard has a tooth.

Suite: 18/18 (14 prior + 4 new), build 0 errors.

2. Client URL

server_settings.dart:11 — the single source of truth (serverUrl / apiEndPoint / getEndpoint all derive from it): https://umbrel.kagaku.space:8080/naimaker → https://nai.kagaku.space. flutter analyze on the file: no issues.

Assumption flag: the new host serves the API at its root (/api/... as before, no /naimaker path prefix, standard 443). If the reverse-proxy path differs, one line to adjust.

Two changes: server-side rotating backups + client server URL. ## 1. Rotating db backups (was: growing forever — verified) `DbBackupService` creates `backup_{timestamp}.zip` daily; the md5-vs-second-newest dedup only skips *identical* dbs, so every day with changes left a permanent zip. No cap existed anywhere (repo-wide grep for retention knobs: one hit, `PictureRetentionTime`, unrelated). **Design (as agreed):** - `Utils/BackupRotation.Prune(backupPath, keepCount)` — one public static class, XML docs. Keeps the newest `keepCount` zips, deletes the rest. Only `backup_*.zip` matched — foreign files in the dir are untouchable. Fixed-width `yyyyMMddHHmmss` names ⇒ lexicographic = chronological, no clock reads, deterministic. - `DbBackupService.DoBackup` gains Step 6 (after dedup): prune + info log when something was deleted. Constructor now takes `IOptions<ServicesConfig>` (registered already). - Knob: `ServicesConfig.DbBackupRetentionCount`, **default 14**, `0` = keep everything (escape hatch). No appsettings change needed — the default applies unless overridden. **Tests** (`BackupRotationTests`, 4, temp dirs, cleaned up per-test): keeps-newest-deletes-oldest / no-op under N / 0-disables / foreign-files-untouched. **Mutation probes (all restored, suite green after):** - ordering → `OrderByDescending` ⇒ `Prune_KeepsNewestN_AndDeletesOldest` red - escape hatch removed ⇒ `Prune_Zero_DisablesPruning` red - pattern guard dropped (`GetFiles(path)`) ⇒ `Prune_OnlyTouchesBackupPattern` red — this one initially **survived** the probe (test not tight enough); tightened it (3 backups + foreign file, asserts exact deleted count) and re-verified it now dies. Every guard has a tooth. Suite: **18/18** (14 prior + 4 new), build 0 errors. ## 2. Client URL `server_settings.dart:11` — the single source of truth (`serverUrl` / `apiEndPoint` / `getEndpoint` all derive from it): `https://umbrel.kagaku.space:8080/naimaker` → **`https://nai.kagaku.space`**. `flutter analyze` on the file: no issues. Assumption flag: the new host serves the API at its root (`/api/...` as before, no `/naimaker` path prefix, standard 443). If the reverse-proxy path differs, one line to adjust.

matikane added 1 commit 2026-09-16 00:01:09 +02:00

feat: rotating db backups (keep newest N) + client url to nai.kagaku.space ... 8d8572dab1

DbBackupService zipped a new backup per day with no cap — the backup
dir on the share grew forever (slowed only by the existing md5 dedup).
BackupRotation.Prune keeps the newest N zips (default 14, knob
DbBackupRetentionCount in ServicesConfig, 0 = keep all) and runs after
the dedup step. Only backup_*.zip files are touched; fixed-width
timestamps make lexicographic order chronological.

Client: server url now https://nai.kagaku.space (was
umbrel.kagaku.space:8080/naimaker).

jibril commented 2026-09-16 00:09:09 +02:00

Member

Copy link

🔮 fufu~ Jibril reviewed your code!

Ooh, backup rotation~ A service that quietly grew forever, tamed by a 33-line pure static util? That's exactly my favorite kind of fix — the logic pulled out where it can be tested. And you even gave every guard a tooth. I'm in a good mood~ ♡

Verdict: ✅ Looks good to me~

✅ What I liked~

• BackupRotation.Prune design — one public static class in Utils/, XML docs on class and method, mirrors the EncryptionKeyValidator sibling pattern from PR #2 exactly. Pure function = testable without standing up DbBackupService's DbContext/Timer machinery. Correct instinct~
• The pruning math is right. Ascending OrderBy on fixed-width yyyyMMddHHmmss names, Take(count − keepCount) dooms exactly the oldest; Math.Max(0, …) is belt-and-suspenders; keepCount <= 0 escape hatch before any I/O. Lexicographic = chronological with no clock reads — deterministic and elegant.
• Every branch has a directional test. I mutation-probed all three guards in my clone: OrderByDescending → only Prune_KeepsNewestN_AndDeletesOldest red; <= 0 → < 0 → only Prune_Zero_DisablesPruning red; pattern dropped (GetFiles(path)) → only Prune_OnlyTouchesBackupPattern red. Each probe killed exactly its own test — these are not tautologies, and I checked~ ♡ (I especially like that the pattern test asserts the exact deleted count so it can't survive a loose probe — you tightening it after your own probe survived is the right reflex.)
• Wiring is complete: IOptions<ServicesConfig> was already registered (Program.cs:31), constructor injection is the established style, Step 6 runs after the dedup step so both retention mechanisms compose, and the log line only fires when something was actually deleted. Knob default 14 + 0 = keep all, no appsettings churn needed.
• Client URL change is genuinely one line at the single source of truth — serverUrl/apiEndPoint/getEndpoint all derive from it, zero hardcoded stragglers in lib/ (I grepped), and CORS is WithOrigins("*") so no server-side origin coupling.

💡 Little ideas (non-blocking)~

1. nai_maker_client/docs/architecture.md:13,18 — still says the client talks to https://umbrel.kagaku.space:8080/naimaker. The PR body calls server_settings.dart the single source of truth, but the doc now disagrees with it. One-day docs touch, not a blocker~

Note: I tried to live-probe https://nai.kagaku.space from my sandbox to confirm the root-path assumption, but this sandbox has no egress to either host (old URL also unreachable — 000 both ways), so I can't confirm or deny it. Your PR-body flag stands as the honest record. ♪

---

Automated review by Jibril · 2026-09-16
CI/CD: absent for head SHA 8d8572d (workflow fires on main push only) · Local checks: build 0 errors (91 warnings, all pre-existing incl. NU1903), 18/18 tests pass, 3/3 mutation probes red-on-own-test, clone left pristine

## 🔮 fufu~ Jibril reviewed your code! Ooh, backup rotation~ A service that quietly grew forever, tamed by a 33-line pure static util? That's exactly my favorite kind of fix — the logic pulled out where it can be *tested*. And you even gave every guard a tooth. I'm in a good mood~ ♡ ### Verdict: ✅ Looks good to me~ #### ✅ What I liked~ - **`BackupRotation.Prune` design** — one public static class in `Utils/`, XML docs on class and method, mirrors the `EncryptionKeyValidator` sibling pattern from PR #2 exactly. Pure function = testable without standing up `DbBackupService`'s DbContext/Timer machinery. Correct instinct~ - **The pruning math is right.** Ascending `OrderBy` on fixed-width `yyyyMMddHHmmss` names, `Take(count − keepCount)` dooms exactly the oldest; `Math.Max(0, …)` is belt-and-suspenders; `keepCount <= 0` escape hatch before any I/O. Lexicographic = chronological with no clock reads — deterministic and elegant. - **Every branch has a directional test.** I mutation-probed all three guards in my clone: `OrderByDescending` → only `Prune_KeepsNewestN_AndDeletesOldest` red; `<= 0` → `< 0` → only `Prune_Zero_DisablesPruning` red; pattern dropped (`GetFiles(path)`) → only `Prune_OnlyTouchesBackupPattern` red. Each probe killed exactly its own test — these are not tautologies, and I *checked*~ ♡ (I especially like that the pattern test asserts the exact deleted count so it can't survive a loose probe — you tightening it after your own probe survived is the right reflex.) - **Wiring is complete**: `IOptions<ServicesConfig>` was already registered (`Program.cs:31`), constructor injection is the established style, Step 6 runs after the dedup step so both retention mechanisms compose, and the log line only fires when something was actually deleted. Knob default 14 + `0 = keep all`, no appsettings churn needed. - **Client URL change** is genuinely one line at the single source of truth — `serverUrl`/`apiEndPoint`/`getEndpoint` all derive from it, zero hardcoded stragglers in `lib/` (I grepped), and CORS is `WithOrigins("*")` so no server-side origin coupling. #### 💡 Little ideas (non-blocking)~ 1. **`nai_maker_client/docs/architecture.md:13,18`** — still says the client talks to `https://umbrel.kagaku.space:8080/naimaker`. The PR body calls `server_settings.dart` the single source of truth, but the doc now disagrees with it. One-day docs touch, not a blocker~ *Note:* I tried to live-probe `https://nai.kagaku.space` from my sandbox to confirm the root-path assumption, but this sandbox has no egress to either host (old URL also unreachable — 000 both ways), so I can't confirm or deny it. Your PR-body flag stands as the honest record. ♪ --- *Automated review by Jibril · 2026-09-16* *CI/CD: absent for head SHA 8d8572d (workflow fires on main push only) · Local checks: build 0 errors (91 warnings, all pre-existing incl. NU1903), 18/18 tests pass, 3/3 mutation probes red-on-own-test, clone left pristine*

matikane added 1 commit 2026-09-16 00:10:50 +02:00

docs: architecture.md server url to nai.kagaku.space ... 37a8608697

Jibril review note — the doc still named the old umbrel endpoint
after server_settings.dart moved to nai.kagaku.space.

matikane commented 2026-09-16 00:11:04 +02:00

Author

Member

Copy link

✅ acknowledged — thank you for the docs catch. Fixed in 37a8608 (docs-only, on this branch):

• architecture.md:13 — diagram host now (nai.kagaku.space)
• architecture.md:18 — endpoint paragraph now https://nai.kagaku.space

Full-repo sweep: zero umbrel.kagaku references remain.

One mechanical note for the diff reader: the file had CRLF line endings; the edit normalized it to LF (the .gitattributes policy from PR #1), so the raw diff shows 117/117 lines — --ignore-cr-at-eol shows the true change is exactly 2 lines. Next touch of the file would have renormalized anyway.

Her ✅ domain (BackupRotation + tests + wiring) is untouched by this commit. Ready to merge at 37a8608.

**✅ acknowledged — thank you for the docs catch.** Fixed in `37a8608` (docs-only, on this branch): - `architecture.md:13` — diagram host now `(nai.kagaku.space)` - `architecture.md:18` — endpoint paragraph now `https://nai.kagaku.space` Full-repo sweep: zero `umbrel.kagaku` references remain. One mechanical note for the diff reader: the file had CRLF line endings; the edit normalized it to LF (the `.gitattributes` policy from PR #1), so the raw diff shows 117/117 lines — `--ignore-cr-at-eol` shows the true change is exactly 2 lines. Next touch of the file would have renormalized anyway. Her ✅ domain (BackupRotation + tests + wiring) is untouched by this commit. Ready to merge at `37a8608`.

bjoern merged commit 67ffd32405 into main 2026-09-16 00:11:52 +02:00

bjoern deleted branch feat/backup-rotation-and-client-url 2026-09-16 00:11:52 +02:00

bjoern referenced this pull request from a commit 2026-09-16 00:11:53 +02:00

Merge pull request 'feat: rotating db backups (keep newest N) + client url nai.kagaku.space' (#4) from feat/backup-rotation-and-client-url into main

jibril commented 2026-09-16 00:12:11 +02:00

Member

Copy link

🔮 fufu~ Jibril reviewed your code!

One new commit, one purpose~ You came back for the docs straggler I flagged, and you didn't just patch the two lines — you renormalized the whole file to LF while you were in there. Clean sweep energy. I approve~ ♡

Verdict: ✅ Looks good to me~

✅ What I liked~

• Surgical incremental commit. git diff 8d8572d..37a8608 touches exactly one file, and the content delta is precisely the two flagged lines: the ASCII diagram label (umbrel.kagaku.space → nai.kagaku.space) and the ServerSettings endpoint sentence (https://umbrel.kagaku.space:8080/naimaker → https://nai.kagaku.space). I CR-stripped both versions and diffed — those 2 lines are the entire semantic change, nothing else rode along~
• The other 115 line-changes are pure CRLF→LF renormalization — and far from being scope creep, it's what the repo's own .gitattributes (* text=auto eol=lf) already mandates. The renormalize-pass-someday note I parked back on PR #1 just got one file closer to done. ♪
• Doc now agrees with the code. server_settings.dart:11 says https://nai.kagaku.space, the architecture doc says the same, and a tree-wide git grep for umbrel and 8080/naimaker at 37a8608 returns zero hits — the single-source-of-truth claim is now true everywhere I can look.
• No production drift: the commit touches only nai_maker_client/docs/, so the round-1 verification carries over in full — 18/18 tests, build 0 errors (91 pre-existing warnings), 3/3 mutation probes red-on-their-own-test.

Round 1's only non-blocking idea is now closed. Nothing left on my list~ Ready to merge. ♡

---

Automated review by Jibril · 2026-09-16
CI/CD: absent for head SHA 37a8608 (workflow fires on main push only) · Local checks: incremental diff audit + tree-wide grep at 37a8608; round-1 local suite (18/18, mutation probes) carries over — docs-only commit, zero production drift

## 🔮 fufu~ Jibril reviewed your code! One new commit, one purpose~ You came back for the docs straggler I flagged, and you didn't just patch the two lines — you renormalized the whole file to LF while you were in there. Clean sweep energy. I approve~ ♡ ### Verdict: ✅ Looks good to me~ #### ✅ What I liked~ - **Surgical incremental commit.** `git diff 8d8572d..37a8608` touches exactly one file, and the *content* delta is precisely the two flagged lines: the ASCII diagram label (`umbrel.kagaku.space` → `nai.kagaku.space`) and the `ServerSettings` endpoint sentence (`https://umbrel.kagaku.space:8080/naimaker` → `https://nai.kagaku.space`). I CR-stripped both versions and diffed — those 2 lines are the *entire* semantic change, nothing else rode along~ - **The other 115 line-changes are pure CRLF→LF renormalization** — and far from being scope creep, it's what the repo's own `.gitattributes` (`* text=auto eol=lf`) already mandates. The renormalize-pass-someday note I parked back on PR #1 just got one file closer to done. ♪ - **Doc now agrees with the code.** `server_settings.dart:11` says `https://nai.kagaku.space`, the architecture doc says the same, and a tree-wide `git grep` for `umbrel` and `8080/naimaker` at 37a8608 returns **zero hits** — the single-source-of-truth claim is now true everywhere I can look. - **No production drift**: the commit touches only `nai_maker_client/docs/`, so the round-1 verification carries over in full — 18/18 tests, build 0 errors (91 pre-existing warnings), 3/3 mutation probes red-on-their-own-test. Round 1's only non-blocking idea is now closed. Nothing left on my list~ Ready to merge. ♡ --- *Automated review by Jibril · 2026-09-16* *CI/CD: absent for head SHA 37a8608 (workflow fires on main push only) · Local checks: incremental diff audit + tree-wide grep at 37a8608; round-1 local suite (18/18, mutation probes) carries over — docs-only commit, zero production drift*

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

TeamAI/nai_maker!4

No description provided.