REPORT: R-82 agent arc v0.97.0 -> v0.102.0 (overwrite)
Four defects found by running it rather than reviewing it, the frozen untargeted contract verified live, the fail-safe directions stated once, and what is NOT done — including that the scheduled restore-test never selects the offsite tier and that R-84 is now closer to a prerequisite than a tidy-up.
This commit is contained in:
Claude Code
@@ -1,211 +1,92 @@
|
||||
# REPORT — v0.97.0: R-82 Slice A — per-target backup tiers (2026-07-26)
|
||||
# REPORT — R-82: per-target backup tiers, agent v0.97.0 → v0.102.0 (2026-07-26)
|
||||
|
||||
Slice A of R-82 (the backup target split). **Mechanism only — no box's behaviour changes until a
|
||||
`backup_targets` entry is added to its config, which is Slice D.** An untouched config resolves to
|
||||
exactly one tier and behaves byte-identically to v0.96.0.
|
||||
**Overwritten** per the standing rule. Live on **demo-felhom** and **demo-hp**.
|
||||
Full cross-repo arc + the Phase-0 gates: `felhom.eu/REPORT.md`.
|
||||
|
||||
Phase-0 gates: `felhom.eu/documentation/audits/SPIKE-r82-phase0-2026-07-26.md`.
|
||||
Operator ruling 2026-07-26: datastore growth deferred, proceed. Capacity state recorded in
|
||||
`felhom.eu/documentation/architecture/07-backup-architecture.md` §9.1.
|
||||
## Versions in this arc
|
||||
|
||||
**STOPPED at the Slice A boundary as specified.** Slices B/C/D/E not started.
|
||||
|
||||
---
|
||||
|
||||
## 1. What was wrong
|
||||
|
||||
`BackupTarget()` returned ONE string and `BackupCadence()` ONE 24 h window. "Local daily **and** PBS
|
||||
weekly" was not expressible at all — which is why the DR promise is unbacked today: demo-felhom holds
|
||||
a single PBS snapshot from 2026-07-18 (a healing artifact, not a job) and demo-hp has **zero, ever**,
|
||||
despite `pbs_dr` reporting `applied` since 07-21.
|
||||
|
||||
## 2. Files modified
|
||||
|
||||
| File | Change |
|
||||
| | |
|
||||
|---|---|
|
||||
| `internal/config/config.go` | `BackupTargetConfig`, `BackupConfig.ExtraTargets` (`backup_targets`), `BackupTier`, `BackupTiers()` |
|
||||
| `internal/localapi/server.go` | per-target due/status/POST, `?target=` resolution, `/backup/tiers`, jobs keyed by (vmid,target), per-tier job ids |
|
||||
| `internal/localapi/backup_tiers.go` | **NEW** — `normalizeBackupTiers` (the compatibility core) |
|
||||
| `cmd/felhom-agent/main.go` | one runner per tier; tier-rejection warnings logged at **ERROR** |
|
||||
| `internal/config/backup_tiers_test.go` | **NEW** — 7 tests |
|
||||
| `internal/localapi/backup_tiers_test.go` | **NEW** — 13 tests |
|
||||
| `CHANGELOG.md`, `REUSE.md` | v0.97.0 entry; new canonical-helper row |
|
||||
| **v0.97.0** | Slice A — per-target cadence/retention, `/backup/tiers`, per-tier due + runner + job keys |
|
||||
| **v0.98.0** | the 30-minute false failure — per-tier vzdump wait bound |
|
||||
| **v0.99.0** | operator rulings — 2-week offsite retention (scoped PBS prune opt-in) + one backup at a time per guest |
|
||||
| **v0.100.0** | the restore tier comes from the ARCHIVE, not the configured target |
|
||||
| **v0.101.0** | a leaked restore-test scratch can no longer auto-start (`onboot=0` at restore time) |
|
||||
| **v0.102.0** | an unprovisioned tier DEFERS instead of firing at a missing storage |
|
||||
|
||||
Commit: **`739b3c3`** on `main`, pushed.
|
||||
## Four defects found by RUNNING it, not reviewing it
|
||||
|
||||
## 3. The compatibility rule — how it is enforced
|
||||
1. **v0.98.0** — a 41-minute PBS backup was recorded `success:false` at 30 minutes **while it was
|
||||
still running**, and it later completed `TASK OK`. Worse than "the backup didn't happen": the
|
||||
tier stays permanently due and the retry collides with the guest lock.
|
||||
2. **v0.100.0** — a `felhom-pbs:` archive was classified `local` and got the 10-minute bound against
|
||||
a 14.46 GB WAN restore, failing at 600 s. **A silent regression of the S4.1 fix** — the mechanism
|
||||
was never removed; its *input* changed when `local_backup_target` was retargeted to `local`. A fix
|
||||
keyed on *"the configured target"* stops holding the moment more than one target exists.
|
||||
3. **v0.101.0** — a leaked scratch guest kept `onboot: 1`, so a host reboot would have started a
|
||||
clone of the live guest. `onboot=0` is now set **at restore time**, because "after" is the path
|
||||
that leaks.
|
||||
4. **v0.102.0** — a tier fires at a storage that does not exist yet (a fresh box before DR
|
||||
provisioning). Now deferred, going live with no restart once the storage appears.
|
||||
|
||||
The agent and controller deploy independently, so the untargeted contract is **frozen**:
|
||||
## The frozen contract
|
||||
|
||||
- `BackupDueResponse.Target` / `BackupResponse.Target` / `BackupStatusResponse.Target` are all
|
||||
`omitempty` and left **EMPTY** for an untargeted request → the response marshals to the pre-R-82
|
||||
**bytes**, not merely a compatible shape. Verified live:
|
||||
`{"ok":true,"data":{"vmid":8200,"due":false,"reason":"within cadence window","age_seconds":7200}}`
|
||||
- The untargeted verdict comes from the **primary tier only** — a fresh backup on another tier must
|
||||
not satisfy it (own test).
|
||||
- The **primary tier's job-id format is unchanged**; only additive tiers carry a target segment.
|
||||
- An unknown `?target=` is a **400, never a silent fallback**. A controller asking about a tier this
|
||||
agent does not serve must find out, not act on a different tier's freshness.
|
||||
- `GET /backup/tiers` 404s on a pre-R-82 agent — that 404 is Slice B's designed capability probe.
|
||||
|
||||
## 4. Design decisions, and why
|
||||
|
||||
**One runner per tier** (not a target-parameterised runner). The runner holds target, mode, notes and
|
||||
retention as immutable construction state, and `localPruneSpec` reads that retention — parameterising
|
||||
a single runner by target would make it possible to pair tier A's target with tier B's retention. One
|
||||
runner per tier makes each tier's policy structurally inseparable from its target.
|
||||
|
||||
**A missing cadence is REJECTED, not defaulted.** Defaulting an extra tier to the 24 h local default
|
||||
would quietly turn a weekly DR tier into a daily one and fill the 37.2 GB datastore. A tier whose
|
||||
cadence you did not state is not a tier. `main.go` logs every rejection at **ERROR** — a silently
|
||||
dropped backup tier is the "applied and empty" fault R-82 exists to fix.
|
||||
|
||||
**`keep_last` unset = never prune.** The fail-safe: a DR tier must not start pruning itself because
|
||||
someone forgot a field. Negative values clamp to the same.
|
||||
|
||||
**Jobs keyed by (vmid, target).** This is what lets the weekly night run both backups inside ONE
|
||||
quiesce window (Slice B). It also fixed a real defect — see §6.
|
||||
|
||||
## 5. The Slice A.4 finding (Slice C depends on this)
|
||||
|
||||
**A PBS-targeted vzdump will appear in BOTH arrays**, and this is expected, not a bug:
|
||||
|
||||
- in `backups[]` as `Backup{TargetID:"felhom-pbs", Archive:"felhom-pbs:backup/ct/9201/…"}` — written
|
||||
by the runner, recorded in the agent's in-memory store;
|
||||
- in `pbs_snapshots[]` — enumerated **independently** from the PBS API by the agent's verify loop.
|
||||
|
||||
**Double-counting is harmless for freshness** (the hub takes a max, which is idempotent).
|
||||
**Mis-attribution is not.** If Slice C treats `backups[]` as "the local tier" and `pbs_snapshots[]`
|
||||
as "the PBS tier", a PBS-target `Backup` record would be attributed to the local tier and make a
|
||||
**stale local tier look fresh**.
|
||||
|
||||
**The rule Slice C must implement:** classify a `backups[]` entry by joining its `target_id` to
|
||||
`storage_targets[].name` and reading `.type == "pbs"` — **by target type, never by array membership.**
|
||||
|
||||
Verified live that the discriminator is present in production reports:
|
||||
Untargeted `/backup/due`, `POST /backup` and `/backup/status` keep the **primary tier and the
|
||||
pre-R-82 response bytes** (`Target` is `omitempty` and stays empty); the primary's job-id format is
|
||||
unchanged. Verified live on demo-felhom:
|
||||
|
||||
```
|
||||
demo-felhom name=felhom-pbs type=pbs content=backup reachable=true
|
||||
demo-felhom name=local type=local content=…,backup reachable=true
|
||||
demo-hp name=felhom-pbs type=pbs content=backup reachable=true
|
||||
drill-r50 name=local type=local content=backup,… reachable=true (no PBS storage yet)
|
||||
/backup/due (untargeted) {"vmid":9201,"due":true,"reason":"no successful backup recorded yet"}
|
||||
/backup/due?target=felhom-pbs {...,"target":"felhom-pbs"}
|
||||
/backup/due?target=bogus 400 "unknown backup target: bogus"
|
||||
```
|
||||
|
||||
This is preferable to keying off `pbs_dr.storage_id`, which is `null` on drill-r50.
|
||||
An unknown `?target=` is a **400, never a silent fallback** — a caller asking about a tier this agent
|
||||
does not serve must find out, not act on another tier's freshness.
|
||||
|
||||
## 6. Tests — 748 → 768 (+20)
|
||||
## Fail-safe directions, stated once
|
||||
|
||||
Full gate green, **`go vet` run unpiped** (rc printed separately; R-81 recorded that piping it into
|
||||
`head` reports rc=0 while failing):
|
||||
- Unparseable timestamp → **due** (a spurious backup is cheap; a skipped one is not).
|
||||
- Storage-view error → **present** (`"I could not check"` is not `"not there"` — reading it that way
|
||||
would silently suppress backups, the rule this project has relearned three times).
|
||||
- Missing tier cadence → **rejected, logged at ERROR** (never defaulted: a weekly DR tier silently
|
||||
running daily would fill the datastore).
|
||||
- Unset `keep_last` → **never prune** (a DR tier must not start pruning itself because someone forgot
|
||||
a field). The primary tier keeps the **absolute** PBS-prune refusal, because its target *and*
|
||||
retention both default and could prune the DR by accident.
|
||||
|
||||
```
|
||||
go build ./... rc=0
|
||||
go vet ./... rc=0
|
||||
go test ./... rc=0 (every package ok)
|
||||
```
|
||||
## Live validation (demo-felhom)
|
||||
|
||||
### Red-proof #1 — old controller ↔ new agent (mandatory) — OBSERVED
|
||||
Both tiers armed: `local 24h keep_last=3 wait 30m` + `felhom-pbs 168h keep_last=2 wait 12h
|
||||
prune_pbs_allowed=true`. First real PBS-targeted backup: **`TASK OK`, 41 minutes, 14.46 GB
|
||||
snapshot**, and it **restored cleanly** (`vzrestore: stopped OK`). Measured incremental cost:
|
||||
**+2.7 GB on disk** for a 14.46 GB logical snapshot (~81 % dedup).
|
||||
|
||||
Removed the untargeted compat branch in `tierFromRequest` so it echoes the primary's id like any
|
||||
other tier:
|
||||
demo-hp: v0.102.0 deployed via the documented break-glass path (binary + config backed up first),
|
||||
tier armed, first-ever PBS backup triggered against a verifiably empty namespace.
|
||||
|
||||
```
|
||||
--- FAIL: TestBackupDue_Untargeted_ResponseBytesUnchanged (0.00s)
|
||||
backup_tiers_test.go:105: UNTARGETED response MUST NOT carry a target key — an old controller
|
||||
sees a changed contract; body: {"ok":true,"data":{"vmid":8200,"due":false,
|
||||
"reason":"within cadence window","age_seconds":7200,"target":"local"}}
|
||||
```
|
||||
## Tests
|
||||
|
||||
Restored; suite green.
|
||||
`go build ./... && go vet ./... && go test ./...` — **rc=0, 29 packages**, vet run unpiped.
|
||||
|
||||
### A defect the tests caught (worth recording)
|
||||
Red-proofs observed and restored: old-controller compat (`"target":"local"` leaking into the
|
||||
untargeted body), the per-tier wait bound, and the `onboot` override. A per-tier single-flight defect
|
||||
was caught **by its own test** before it ever shipped — job ids were unique only by clock luck, so
|
||||
the PBS request was handed the local job's id.
|
||||
|
||||
`TestBackupPost_SingleFlightIsPerTierNotPerGuest` **failed on first run**:
|
||||
**Process failure recorded:** I ran the suite and committed in the same command, read
|
||||
`packages ok: 28` and pushed without checking `rc=1`; five of my own tests were failing (a harness
|
||||
artifact, fixed in the follow-up). That is the exact exit-code trap recorded twice earlier in this
|
||||
arc.
|
||||
|
||||
```
|
||||
backup_tiers_test.go:320: PER-TIER single-flight violated: the PBS request was handed the LOCAL
|
||||
job id "backup-8200-1781092800000000000" — the controller would believe the PBS backup ran
|
||||
```
|
||||
## NOT done
|
||||
|
||||
Job ids were `backup-<vmid>-<unixnano>` — unique only by clock luck, and identical under an injected
|
||||
clock or two tiers starting in the same nanosecond. Fixed so ids are unique **per tier by
|
||||
construction**, with the primary's format left unchanged. This is precisely the failure mode Slice B
|
||||
would have hit on the weekly both-due night.
|
||||
1. The **scheduled** restore-test still only ever selects the **primary** tier, so the offsite tier
|
||||
is never automatically restore-tested — arguably the more important half of "is the DR tier real?".
|
||||
2. The agent does not report per-tier **cadences**, so the hub must infer "PBS ⇒ weekly" from storage
|
||||
type.
|
||||
3. The backup `Store` is still in-memory (**R-84**): every agent restart re-triggers a backup on every
|
||||
tier. On the local tier that is wasted minutes; on the offsite tier it is a wasted multi-hour
|
||||
upload after every deploy — which makes R-84 closer to a prerequisite than a tidy-up.
|
||||
4. The boot+verify half of the restore round-trip has not run under the fixed code.
|
||||
|
||||
## 7. Live validation performed
|
||||
|
||||
**No live deploy — see §8.** The strongest zero-risk check available was run instead: an end-to-end
|
||||
wire test composing the REAL path `agent.json bytes → config.BackupConfig → BackupTiers() →
|
||||
localapi.BackupTier → a real TLS listener`, driven by a real HTTP client (`httptest`'s own trusting
|
||||
client — TLS verification was **not** disabled). Unit tests cover each half; this proves the join,
|
||||
which is where a per-tier mis-wire would actually live.
|
||||
|
||||
```
|
||||
tier armed: target=local cadence=24h0m0s keep_last=3 primary=true
|
||||
tier armed: target=felhom-pbs cadence=168h0m0s keep_last=2 primary=false
|
||||
|
||||
GET /backup/tiers -> 200 {"tiers":[{"target":"local","cadence_seconds":86400,"primary":true},
|
||||
{"target":"felhom-pbs","cadence_seconds":604800,"primary":false}]}
|
||||
GET /backup/due (UNTARGETED) -> 200 {"vmid":8200,"due":false,"reason":"within cadence window","age_seconds":7200}
|
||||
GET /backup/due?target=local -> 200 {…,"due":false,"target":"local"}
|
||||
GET /backup/due?target=felhom-pbs -> 200 {…,"due":true,"reason":"older than cadence","age_seconds":691200,"target":"felhom-pbs"}
|
||||
GET /backup/due?target=bogus -> 400 {"ok":false,"error":"unknown backup target: bogus"}
|
||||
```
|
||||
|
||||
The fixture is the exact `agent.json` shape Slice D will write: fresh daily local (2 h) + a PBS
|
||||
snapshot at 8 days. The DR tier correctly reports **due** while the local tier does not — the
|
||||
behaviour whose absence is the entire bug. Harness deleted; tree clean.
|
||||
|
||||
Binary built and verified: `felhom-agent 0.97.0` (13.9 MB, `/tmp/felhom-agent-0.97.0`). **Not
|
||||
installed anywhere.**
|
||||
|
||||
## 8. NOT live-validated — and the deploy target problem
|
||||
|
||||
**Slice A's "deploy to the drill VM only" step was NOT performed: no drill target is reachable.**
|
||||
|
||||
| Candidate | Status |
|
||||
|---|---|
|
||||
| **drill-r50** (`drill-r50-0a4f9a`) — the correct target: a live, hub-enrolled drill host | **UNREACHABLE.** Its operator path is wg-only (`10.77.0.4:8822`, OOB `felhom_sshd_active:true`). DooPlex has **no wg interface** — TCP connect to `10.77.0.4:8822` fails. No SSH alias exists |
|
||||
| **`/mnt/5_hdd/felhom.eu/drill/drill.qcow2`** on DooPlex | Present but **not a Felhom host** — a bare PVE install whose only snapshot is `virgin` (2026-07-03), powered off. Using it means a full greenfield run (host-install → guest provision → hub enrol), which is a different task with real half-state risk, not "deploy the agent" |
|
||||
| demo-felhom / demo-hp | **Production — explicitly excluded by this slice** |
|
||||
|
||||
Consequently the following are unproven and must be proven at Slice D step 1:
|
||||
|
||||
1. **The tier-arming startup log on a real host** (`backup tier armed` × N, and `backup tier REJECTED`
|
||||
at ERROR for a bad config). Proven by unit test + the composed wire test only.
|
||||
2. **A real vzdump to `felhom-pbs` via the new per-tier runner** — no PBS-targeted backup has been
|
||||
run by this code.
|
||||
3. **The A.4 double-appearance in a real host-report.** §5 is derived from source plus the live
|
||||
`storage_targets` shape; it has not been observed with an actual PBS-target `Backup` record
|
||||
present, because none exists yet.
|
||||
4. **Per-tier retention against a real storage** — `localPruneSpec`'s PBS refusal is unchanged, so no
|
||||
PBS prune can occur, but that is asserted from source, not exercised.
|
||||
5. **The restore round-trip.** Per the task's own bar, a tier is only real once a `pct` restore from
|
||||
its snapshot yields a working guest. Nothing here approaches that.
|
||||
|
||||
## 9. Open decisions carried forward
|
||||
|
||||
- **PBS pruning is still OFF.** Per-tier retention is plumbed and each tier carries its own
|
||||
`keep_last`, but `localPruneSpec` still refuses to prune a PBS-type target, and an unset
|
||||
`keep_last` means never-prune. Turning on automatic pruning of the DR datastore is irreversible and
|
||||
needs an operator ruling — P0.3 already flagged retention as one of the levers. **Left OFF
|
||||
deliberately; with capacity already tight, the wrong default here is unrecoverable.**
|
||||
- **The P0.3 capacity constraint stands.** Growth deferred by operator ruling; the 80 % warn is still
|
||||
projected at roughly the second additional customer.
|
||||
- **drill-r50 has no Tier-3 offsite** (`offsite: null`), so the P0.1 "weekly is sufficient" verdict
|
||||
does **not** hold for it. It is first in the Slice D rollout order — its PBS cadence cannot simply
|
||||
inherit the fleet ruling.
|
||||
|
||||
## 10. Observations
|
||||
|
||||
1. `local_backup_target` on both production boxes is `"local"`, so `BackupTarget()`'s
|
||||
`defaultBackupTarget = "felhom-pbs"` fallback is never exercised in production. A box with that
|
||||
key **absent** would send its primary daily backup to PBS — surprising, and worth checking during
|
||||
Slice D before writing any config.
|
||||
2. The agent's backup `Store` remains in-memory (R-84). Slice A does not change that: after a restart
|
||||
both tiers' records are empty until each next run, so `/backup/due` reports **due** for every tier.
|
||||
That is fail-safe (a spurious backup, not a skipped one), but on the weekly tier it means a
|
||||
restart can pull a PBS backup forward. Worth naming in Slice B's dedup reasoning.
|
||||
3. `RestoreTest.SourceTier` is still typed as `"local"` with a comment saying PBS is "Phase B". With a
|
||||
real PBS tier arriving, the restore-test's tier attribution should be revisited — not in scope
|
||||
here, recorded.
|
||||
|
||||
Reference in New Issue
Block a user