HAProxy resolved `redis-node-{0,1}.redis-headless.redis.svc.cluster.local`
once at pod startup and cached the IPs forever. When redis-node pods
cycled (new pod IPs), HAProxy kept connecting to the dead IPs — backends
flapped between "Connection refused" and "Layer4 timeout", and Immich's
ioredis client hit EPIPE until max-retries exhausted and the pod entered
CrashLoopBackOff. This caused an Immich outage on 2026-04-19.
Fix:
- Add `resolvers kubernetes` stanza pointing at kube-dns (10s hold on
every category so we pick up pod IP changes within a DNS TTL window).
- Add `resolvers kubernetes init-addr last,libc,none` to every backend
server line so HAProxy resolves at startup AND uses the dynamic
resolver for runtime refresh.
- Add `checksum/config` pod annotation to the HAProxy Deployment so a
haproxy.cfg change actually rolls the pods (including this one).
Closes: code-fd6
## Context
`null_resource.patch_redis_service` uses `triggers = { always = timestamp() }`,
so every `scripts/tg plan` on `stacks/redis` reports `1 to destroy, 1 to add`
even when nothing has changed. That noise hides real drift in the signal and
trains us to ignore redis-stack plans — which is exactly what you don't want
on a load-bearing patch.
The patch itself is still load-bearing (three consumers hard-code bare
`redis.redis.svc.cluster.local` — `stacks/immich/chart_values.tpl:12`,
`stacks/ytdlp/yt-highlights/app/main.py:136`, `config.tfvars:214` — plus
Bitnami's own sentinel scripts set `REDIS_SERVICE=redis.redis.svc.cluster.local`
and call it during pod startup). Removing the null_resource is a follow-up
(beads T0) once those consumers migrate to `redis-master.redis.svc`. For now
the goal is just: stop being noisy.
## This change
1. Replace the `always = timestamp()` trigger with two inputs that only change
when re-patching is genuinely required:
- `chart_version = helm_release.redis.version` — changes only on a Bitnami
chart version bump, which is the one code path that rewrites the `redis`
Service selector back to `component=node`.
- `haproxy_config = sha256(kubernetes_config_map.haproxy.data["haproxy.cfg"])`
— changes only when HAProxy config is edited; aligned with the existing
`checksum/config` annotation that rolls the Deployment on config change.
Both attributes are known at plan time (verified against `hashicorp/helm`
v3.1.1 provider binary). Rejected alternatives — `metadata[0].revision`
(not exposed in the plugin-framework v3 rewrite), `sha256(jsonencode(values))`
(readability unverified on v3), and `kubernetes_deployment.haproxy.id`
(static `namespace/name`, never changes) — don't meet the bar.
2. Add a **Redis Service Naming** section to `AGENTS.md` that explicitly
states the write/sentinel/avoid endpoints, so new consumers start from
`redis-master.redis.svc` (the documented `var.redis_host`) and long-lived
connections (PUBSUB, BLPOP, Sidekiq) route around HAProxy's `timeout
client 30s` via the sentinel headless path. Uptime Kuma's Redis monitor
already learned that lesson the hard way (memory id=748).
## What is NOT in this change
- Deleting `null_resource.patch_redis_service` — still load-bearing (T0).
- Deleting `kubernetes_service.redis_master` — stays as the declared write API.
- Migrating any consumer off bare `redis.redis.svc` — T0 epic.
- Per-client sentinel migration — T1 epic.
- Retiring HAProxy — T2 epic (blocked on T1 + T3).
## Before / after
Before (steady state):
```
scripts/tg plan
Plan: 1 to add, 2 to change, 1 to destroy.
# null_resource.patch_redis_service must be replaced
# triggers = { "always" = "<timestamp>" } -> (known after apply)
```
After (steady state, post-apply):
```
scripts/tg plan
No changes. Your infrastructure matches the configuration.
```
After (chart version bump):
```
scripts/tg plan
# null_resource.patch_redis_service must be replaced
# triggers = { "chart_version" = "25.3.2" -> "25.4.0" }
```
— the trigger fires only when it actually needs to.
## Test Plan
### Automated
`scripts/tg plan` pre-change (confirms baseline noise):
```
# module.redis.null_resource.patch_redis_service must be replaced
-/+ resource "null_resource" "patch_redis_service" {
~ triggers = { # forces replacement
~ "always" = "2026-04-19T10:39:40Z" -> (known after apply)
}
}
Plan: 1 to add, 2 to change, 1 to destroy.
```
`scripts/tg plan` post-edit (confirms the one-time structural replacement):
```
# module.redis.null_resource.patch_redis_service must be replaced
-/+ resource "null_resource" "patch_redis_service" {
~ triggers = { # forces replacement
- "always" = "2026-04-19T10:39:40Z" -> null
+ "chart_version" = "25.3.2"
+ "haproxy_config" = "989bca9483cb9f9942017320765ec0751ac8357ff447acc5ed11f0a14b609775"
}
}
```
Apply is deferred to the operator — the working tree on the same file also
contains an unrelated HAProxy DNS-resolvers fix (for today's immich outage)
that needs its own review before rolling out together. No `scripts/tg apply`
run from this session.
### Manual Verification
Reproduce locally:
1. `cd infra/stacks/redis && ../../scripts/tg plan`
2. Before apply: expect `null_resource.patch_redis_service` to be replaced
exactly once, with the trigger map transitioning from `{always = <ts>}`
to `{chart_version, haproxy_config}`.
3. After apply: `../../scripts/tg plan` twice in a row must both report
`No changes.` (excluding unrelated drift from other work-in-progress).
4. Cluster-side invariant (must hold pre- and post-apply):
`kubectl -n redis get svc redis -o jsonpath='{.spec.selector}'`
→ `{"app":"redis-haproxy"}`
`kubectl -n redis get svc redis-master -o jsonpath='{.spec.selector}'`
→ `{"app":"redis-haproxy"}`
5. Regression test for the trigger doing its job: bump `helm_release.redis.version`
in a branch, `tg plan`, expect the null_resource to replace. Revert.
## Context
redis-node-1 was stuck in CrashLoopBackOff for 5d10h with 120 restarts.
Cluster-health check flagged it as WARN; Prometheus was firing
`StatefulSetReplicasMismatch` (redis/redis-node: 1/2 ready) and
`PodCrashLooping` alerts continuously.
## Root cause
Memory limit 64Mi is too tight. Master steady-state is only 21Mi, but
the replica needs transient headroom during PSYNC full resync:
- RDB snapshot transfer buffer
- Copy-on-write during AOF rewrite (`fork()` + writes during snapshot)
- Replication backlog tracking
The replica RSS crossed 64Mi during sync and was OOM-killed (exit 137),
looping forever. This also broke Sentinel quorum when master would
fail — no healthy replica to promote.
## Fix
Master + replica: 64Mi → 256Mi (both requests and limits, per
`CLAUDE.md` resource management rule: `requests=limits` based on
VPA upperBound).
Sentinels stay at 64Mi — they don't store data.
## Deployment note
Helm upgrade initially deadlocked because StatefulSet uses
`OrderedReady` podManagementPolicy: the update rollout refuses to start
until all pods Ready, but redis-node-1 could not be Ready without the
update. Recovered via:
helm rollback redis 43 -n redis
kubectl -n redis patch sts redis-node --type=strategic \
-p '{...memory: 256Mi...}'
kubectl -n redis delete pod redis-node-1 --force
Then `scripts/tg apply` cleanly reconciled state. Deadlock-recovery
runbook to be written under `code-cnf`.
## Verification
kubectl -n redis get pods
redis-node-0 2/2 Running 0 <bounce>
redis-node-1 2/2 Running 0 <bounce>
kubectl -n redis get sts redis-node -o jsonpath='{.spec.template.spec.containers[?(@.name=="redis")].resources.limits.memory}'
256Mi
## Follow-ups filed
- code-a3j: lvm-pvc-snapshot Pushgateway push fails sporadically
(separate root cause; surfaced via same cluster-health run)
- code-cnf: runbook / TF tweak for the OrderedReady + atomic-wait
deadlock recovery
Closes: code-pqt
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
## Context
Wave 3A (commit c9d221d5) added the `# KYVERNO_LIFECYCLE_V1` marker to the
27 pre-existing `ignore_changes = [...dns_config]` sites so they could be
grepped and audited. It did NOT address pod-owning resources that were
simply missing the suppression entirely. Post-Wave-3A sampling (2026-04-18)
found that navidrome, f1-stream, frigate, servarr, monitoring, crowdsec,
and many other stacks showed perpetual `dns_config` drift every plan
because their `kubernetes_deployment` / `kubernetes_stateful_set` /
`kubernetes_cron_job_v1` resources had no `lifecycle {}` block at all.
Root cause (same as Wave 3A): Kyverno's admission webhook stamps
`dns_config { option { name = "ndots"; value = "2" } }` on every pod's
`spec.template.spec.dns_config` to prevent NxDomain search-domain flooding
(see `k8s-ndots-search-domain-nxdomain-flood` skill). Without `ignore_changes`
on every Terraform-managed pod-owner, Terraform repeatedly tries to strip
the injected field.
## This change
Extends the Wave 3A convention by sweeping EVERY `kubernetes_deployment`,
`kubernetes_stateful_set`, `kubernetes_daemon_set`, `kubernetes_cron_job_v1`,
`kubernetes_job_v1` (+ their `_v1` variants) in the repo and ensuring each
carries the right `ignore_changes` path:
- **kubernetes_deployment / stateful_set / daemon_set / job_v1**:
`spec[0].template[0].spec[0].dns_config`
- **kubernetes_cron_job_v1**:
`spec[0].job_template[0].spec[0].template[0].spec[0].dns_config`
(extra `job_template[0]` nesting — the CronJob's PodTemplateSpec is
one level deeper)
Each injection / extension is tagged `# KYVERNO_LIFECYCLE_V1: Kyverno
admission webhook mutates dns_config with ndots=2` inline so the
suppression is discoverable via `rg 'KYVERNO_LIFECYCLE_V1' stacks/`.
Two insertion paths are handled by a Python pass (`/tmp/add_dns_config_ignore.py`):
1. **No existing `lifecycle {}`**: inject a brand-new block just before the
resource's closing `}`. 108 new blocks on 93 files.
2. **Existing `lifecycle {}` (usually for `DRIFT_WORKAROUND: CI owns image tag`
from Wave 4, commit a62b43d1)**: extend its `ignore_changes` list with the
dns_config path. Handles both inline (`= [x]`) and multiline
(`= [\n x,\n]`) forms; ensures the last pre-existing list item carries
a trailing comma so the extended list is valid HCL. 34 extensions.
The script skips anything already mentioning `dns_config` inside an
`ignore_changes`, so re-running is a no-op.
## Scale
- 142 total lifecycle injections/extensions
- 93 `.tf` files touched
- 108 brand-new `lifecycle {}` blocks + 34 extensions of existing ones
- Every Tier 0 and Tier 1 stack with a pod-owning resource is covered
- Together with Wave 3A's 27 pre-existing markers → **169 greppable
`KYVERNO_LIFECYCLE_V1` dns_config sites across the repo**
## What is NOT in this change
- `stacks/trading-bot/main.tf` — entirely commented-out block (`/* … */`).
Python script touched the file, reverted manually.
- `_template/main.tf.example` skeleton — kept minimal on purpose; any
future stack created from it should either inherit the Wave 3A one-line
form or add its own on first `kubernetes_deployment`.
- `terraform fmt` fixes to pre-existing alignment issues in meshcentral,
nvidia/modules/nvidia, vault — unrelated to this commit. Left for a
separate fmt-only pass.
- Non-pod resources (`kubernetes_service`, `kubernetes_secret`,
`kubernetes_manifest`, etc.) — they don't own pods so they don't get
Kyverno dns_config mutation.
## Verification
Random sample post-commit:
```
$ cd stacks/navidrome && ../../scripts/tg plan → No changes.
$ cd stacks/f1-stream && ../../scripts/tg plan → No changes.
$ cd stacks/frigate && ../../scripts/tg plan → No changes.
$ rg -c 'KYVERNO_LIFECYCLE_V1' stacks/ --include='*.tf' --include='*.tf.example' \
| awk -F: '{s+=$2} END {print s}'
169
```
## Reproduce locally
1. `git pull`
2. `rg 'KYVERNO_LIFECYCLE_V1' stacks/ | wc -l` → 169+
3. `cd stacks/navidrome && ../../scripts/tg plan` → expect 0 drift on
the deployment's dns_config field.
Refs: code-seq (Wave 3B dns_config class closed; kubernetes_manifest
annotation class handled separately in 8d94688d for tls_secret)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
## Context
Wave 3B-continued: the Goldilocks VPA dashboard (stacks/vpa) runs a Kyverno
ClusterPolicy `goldilocks-vpa-auto-mode` that mutates every namespace with
`metadata.labels["goldilocks.fairwinds.com/vpa-update-mode"] = "off"`. This
is intentional — Terraform owns container resource limits, and Goldilocks
should only provide recommendations, never auto-update. The label is how
Goldilocks decides per-namespace whether to run its VPA in `off` mode.
Effect on Terraform: every `kubernetes_namespace` resource shows the label
as pending-removal (`-> null`) on every `scripts/tg plan`. Dawarich survey
2026-04-18 confirmed the drift. Cluster-side count: 88 namespaces carry the
label (`kubectl get ns -o json | jq ... | wc -l`). Every TF-managed namespace
is affected.
This commit brings the intentional admission drift under the same
`# KYVERNO_LIFECYCLE_V1` discoverability marker introduced in c9d221d5 for
the ndots dns_config pattern. The marker now stands generically for any
Kyverno admission-webhook drift suppression; the inline comment records
which specific policy stamps which specific field so future grep audits
show why each suppression exists.
## This change
107 `.tf` files touched — every stack's `resource "kubernetes_namespace"`
resource gets:
```hcl
lifecycle {
# KYVERNO_LIFECYCLE_V1: goldilocks-vpa-auto-mode ClusterPolicy stamps this label on every namespace
ignore_changes = [metadata[0].labels["goldilocks.fairwinds.com/vpa-update-mode"]]
}
```
Injection was done with a brace-depth-tracking Python pass (`/tmp/add_goldilocks_ignore.py`):
match `^resource "kubernetes_namespace" ` → track `{` / `}` until the
outermost closing brace → insert the lifecycle block before the closing
brace. The script is idempotent (skips any file that already mentions
`goldilocks.fairwinds.com/vpa-update-mode`) so re-running is safe.
Vault stack picked up 2 namespaces in the same file (k8s-users produces
one, plus a second explicit ns) — confirmed via file diff (+8 lines).
## What is NOT in this change
- `stacks/trading-bot/main.tf` — entire file is `/* … */` commented out
(paused 2026-04-06 per user decision). Reverted after the script ran.
- `stacks/_template/main.tf.example` — per-stack skeleton, intentionally
minimal. User keeps it that way. Not touched by the script (file
has no real `resource "kubernetes_namespace"` — only a placeholder
comment).
- `.terraform/` copies (e.g. `stacks/metallb/.terraform/modules/...`) —
gitignored, won't commit; the live path was edited.
- `terraform fmt` cleanup of adjacent pre-existing alignment issues in
authentik, freedify, hermes-agent, nvidia, vault, meshcentral. Reverted
to keep the commit scoped to the Goldilocks sweep. Those files will
need a separate fmt-only commit or will be cleaned up on next real
apply to that stack.
## Verification
Dawarich (one of the hundred-plus touched stacks) showed the pattern
before and after:
```
$ cd stacks/dawarich && ../../scripts/tg plan
Before:
Plan: 0 to add, 2 to change, 0 to destroy.
# kubernetes_namespace.dawarich will be updated in-place
(goldilocks.fairwinds.com/vpa-update-mode -> null)
# module.tls_secret.kubernetes_secret.tls_secret will be updated in-place
(Kyverno generate.* labels — fixed in 8d94688d)
After:
No changes. Your infrastructure matches the configuration.
```
Injection count check:
```
$ rg -c 'KYVERNO_LIFECYCLE_V1: goldilocks-vpa-auto-mode' stacks/ | awk -F: '{s+=$2} END {print s}'
108
```
## Reproduce locally
1. `git pull`
2. Pick any stack: `cd stacks/<name> && ../../scripts/tg plan`
3. Expect: no drift on the namespace's goldilocks.fairwinds.com/vpa-update-mode label.
Closes: code-dwx
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- NFS CSI: fix liveness-probe port conflict (29652 → 29653)
- Immich ML: add gpu-workload priority class to enable preemption on node1
- dbaas: right-size MySQL memory limits (sidecar 6Gi→350Mi, main 4Gi→3Gi)
- Redis: add redis-master service via HAProxy for master-only routing,
update config.tfvars redis_host to use it
- CoreDNS: forward .viktorbarzin.lan to Technitium ClusterIP (10.96.0.53)
instead of stale LoadBalancer IP (10.0.20.200)
- Trading bot: comment out all resources (no longer needed)
- Vault: remove trading-bot PostgreSQL database role
Deploy topolvm/pvc-autoresizer controller that monitors kubelet_volume_stats
via Prometheus and auto-expands annotated PVCs. Annotated all 9 block-storage
PVCs (proxmox-lvm) with per-PVC thresholds and max limits. Updated PVFillingUp
alert to critical/10m (means auto-expansion failed) and added PVAutoExpanding
info alert at 80%.
/proc/self/io inside $(awk ...) resolves to the awk subprocess PID,
not the parent bash shell. Use $$ (bash PID) to read the correct
process IO counters.
- Add /proc/self/io read/write tracking to vault raft-backup and etcd backup
- Push backup_duration_seconds, backup_read_bytes, backup_written_bytes,
backup_last_success_timestamp to Pushgateway from all 6 backup CronJobs
(etcd skipped — distroless image has no wget/curl)
- Add cloudsync_duration_seconds metric to cloudsync-monitor
- New "Backup Health" Grafana dashboard with 8 panels: time since last backup,
overview table, duration/IO trends, cloud sync status, alerts, CronJob schedule
Phase 3: all 27 platform modules now run as independent stacks.
Platform reduced to empty shell (outputs only) for backward compat
with 72 app stacks that declare dependency "platform".
Fixed technitium cross-module dashboard reference by copying file.
Woodpecker pipeline applies all 27+1 stacks in parallel via loop.
All applied with zero destroys.
