docs: add architecture decision records for error boundaries and hooks architecture

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
DevContainer User
2026-03-05 13:49:54 +00:00
parent e755f69023
commit a79b7be961
3 changed files with 310 additions and 0 deletions
@@ -0,0 +1,151 @@
# ADR 006: Error Boundary with Dual Variants
**Status**: Accepted
**Date**: 2026-03-05
**Deciders**: Development Team
---
## Context
The Sealed Secrets plugin registers components at two distinct integration points in Headlamp:
1. **Route-level**: Full-page views (`SealedSecretList`, `SealingKeysView`) registered via `registerRoute`
2. **Section-level**: Injected detail sections (`SecretDetailsSection`) registered via `registerDetailsViewSection`
Each integration point has different error recovery requirements:
- **Route-level errors** typically stem from API connectivity issues (controller not found, RBAC misconfiguration). Users need troubleshooting guidance and a retry mechanism.
- **Section-level errors** are isolated failures within a host page. The error should be contained without disrupting the rest of the detail view. A simple reload is sufficient.
A single error boundary class cannot serve both needs because the error messaging, recovery actions, and visual treatment differ significantly.
---
## Decision
Implement a `BaseErrorBoundary` abstract class with a `renderError()` template method, then derive two concrete variants:
- **`ApiErrorBoundary`**: Used at route level. Displays connectivity troubleshooting guidance (check controller namespace, RBAC permissions, pod status) with a Retry button that resets the error state.
- **`GenericErrorBoundary`**: Used at section level. Displays a compact error message with a Reload button. Designed to fail gracefully without affecting the parent detail page.
Both variants use `getDerivedStateFromError` for error capture and expose a reset mechanism via `setState({ hasError: false })`.
```typescript
abstract class BaseErrorBoundary extends React.Component<Props, State> {
static getDerivedStateFromError(error: Error): State {
return { hasError: true, error };
}
abstract renderError(error: Error): React.ReactNode;
render() {
if (this.state.hasError) {
return this.renderError(this.state.error);
}
return this.props.children;
}
}
```
---
## Consequences
### Positive
**Appropriate error recovery**: Each integration point gets tailored error messages and recovery actions
**Fault isolation**: Section-level errors don't crash the entire detail page
**Shared base class**: Common error capture logic is defined once in `BaseErrorBoundary`
**Consistent with React patterns**: Error boundaries are the recommended React mechanism for catching render errors
### Negative
⚠️ **Class components required**: React error boundaries must be class components, breaking the otherwise all-functional-component convention
⚠️ **Two components to maintain**: Changes to error handling patterns must be applied to both variants
### Mitigation
- The class component exception is documented and limited to `ErrorBoundary.tsx`
- Both variants share `BaseErrorBoundary`, so common logic changes propagate automatically
---
## Alternatives Considered
### 1. Single generic error boundary
**Pros**:
- Simpler — one component for all uses
- Less code to maintain
**Cons**:
- Cannot provide context-specific troubleshooting guidance
- Route-level errors need different recovery UX than section-level errors
- Generic messages are unhelpful for API connectivity issues
**Rejected**: The error recovery requirements differ too much between route and section contexts.
---
### 2. try/catch in each component
**Pros**:
- No class components needed
- Per-component error handling
**Cons**:
- Cannot catch render-phase errors (React limitation)
- Duplicated error handling logic across every component
- Inconsistent error UX
**Rejected**: React error boundaries are the only mechanism for catching render errors.
---
### 3. React error boundary library (react-error-boundary)
**Pros**:
- Functional component API via `ErrorBoundary` wrapper
- Built-in reset mechanisms
- Well-maintained
**Cons**:
- External dependency not available in plugin runtime
- Plugin cannot add npm dependencies beyond Headlamp peer dependencies
**Rejected**: Dependency constraint makes this infeasible.
---
## Implementation
- `ApiErrorBoundary` wraps `SealedSecretList` and `SealingKeysView` in `index.tsx`
- `GenericErrorBoundary` wraps `SecretDetailsSection` in `index.tsx`
- Both are defined in `src/components/ErrorBoundary.tsx`
- Uses MUI `Alert`, `Box`, `Button`, `Typography` for styled error display
---
## References
- [React Error Boundaries](https://react.dev/reference/react/Component#catching-rendering-errors-with-an-error-boundary)
- [Headlamp Plugin Registration API](https://headlamp.dev/docs/latest/development/plugins/)
---
## Related ADRs
- [ADR 005: Custom React Hooks](005-react-hooks-extraction.md) — Hooks architecture that error boundaries wrap
---
## Changelog
- **2026-03-05**: Initial decision
@@ -0,0 +1,157 @@
# ADR 007: Custom Hooks Architecture vs Data Context
**Status**: Accepted
**Date**: 2026-03-05
**Deciders**: Development Team
---
## Context
All other Headlamp plugins in this project family (polaris, rook, intel-gpu, kube-vip, tns-csi) use a single React Context provider (`*DataProvider`) to centralize data fetching and share state across components. This is the established pattern.
The Sealed Secrets plugin has different requirements:
1. **Multiple independent data domains**: Controller health, RBAC permissions, SealedSecret CRUD, and encryption are logically separate concerns with different lifecycles.
2. **CRD class extension**: `SealedSecret` extends Headlamp's `KubeObject` class, providing its own `useList()` hook — making a centralized fetch redundant for the primary resource.
3. **Write-heavy workflows**: Unlike read-only plugins, sealed-secrets creates, encrypts, and rotates resources. The encryption workflow involves multi-step state (certificate fetch → encrypt → create resource).
4. **Independent refresh cadences**: Controller health polls every 30 seconds; SealedSecret list is reactive via `useList()`; RBAC checks run once on mount.
A single context provider would either become a monolithic "god context" or force artificial coupling between unrelated concerns.
---
## Decision
Use **independent custom hooks** instead of a shared data context:
- **`useControllerHealth(autoRefresh?, intervalMs?)`**: Polls controller `/healthz` endpoint. Returns `{ healthy, checking, error, refresh }`.
- **`usePermissions()`**: Queries RBAC capabilities on mount. Returns permission flags for create, delete, encrypt operations.
- **`useSealedSecretEncryption()`**: Orchestrates the encryption workflow (fetch cert → encrypt values → build manifest). Returns workflow state and action functions.
- **`SealedSecret.useList()`**: Headlamp's built-in `KubeObject.useList()` — reactive to cluster changes, no custom fetch needed.
Each hook manages its own loading, error, and refresh state. Components compose multiple hooks as needed.
```typescript
function SealedSecretList() {
const [secrets, error] = SealedSecret.useList();
const { healthy } = useControllerHealth(true);
const { canCreate } = usePermissions();
// Each concern is independent
}
```
---
## Consequences
### Positive
**Separation of concerns**: Each hook encapsulates a single domain (health, permissions, encryption, CRUD)
**Independent lifecycles**: Controller health polls at 30s; RBAC checks once; list is reactive — no unnecessary coupling
**Composable**: Components pick only the hooks they need, avoiding unnecessary data in scope
**Testable in isolation**: Each hook can be unit-tested independently without mocking an entire context provider
**Leverages Headlamp's KubeObject**: `SealedSecret.useList()` provides reactive list updates without custom fetch logic
### Negative
⚠️ **Diverges from project convention**: Other plugins use the `*DataProvider` pattern — contributors must learn a different approach for this plugin
⚠️ **No single source of truth**: State is distributed across hooks rather than centralized — harder to debug "what data does the plugin have right now?"
⚠️ **Potential duplicate fetches**: If two components both call `useControllerHealth()`, the health endpoint is polled twice
### Mitigation
- The convention divergence is documented in `CLAUDE.md` and this ADR
- Controller health polling is lightweight (single `/healthz` call)
- `SealedSecret.useList()` is internally deduplicated by Headlamp's hook system
---
## Alternatives Considered
### 1. Single SealedSecretsDataProvider context
**Pros**:
- Consistent with other plugins in the project
- Single source of truth for all sealed-secrets data
- Deduplicates fetches automatically
**Cons**:
- Would become a "god context" with 10+ fields spanning unrelated concerns
- All consumers re-render when any field changes (health poll triggers list re-render)
- Encryption workflow state doesn't belong in shared context (it's dialog-scoped)
- `SealedSecret.useList()` already provides reactive CRUD — wrapping it in context adds indirection
**Rejected**: The data domains are too independent; a single context would create artificial coupling.
---
### 2. Multiple specialized contexts
**Pros**:
- Separation of concerns (like hooks)
- Consistent with React Context pattern
**Cons**:
- Three or four nested providers in `index.tsx` — deep nesting
- More boilerplate than hooks (provider + context + consumer hook per domain)
- No benefit over standalone hooks when providers don't need to share state
**Rejected**: Contexts add boilerplate without benefit when data domains are independent.
---
### 3. State management library (Zustand, Jotai)
**Pros**:
- Lightweight, no provider nesting
- Built-in deduplication and memoization
**Cons**:
- External dependency not available in plugin runtime
- Plugins cannot add npm dependencies beyond Headlamp peer dependencies
**Rejected**: Dependency constraint makes this infeasible.
---
## Implementation
```
src/hooks/
├── useControllerHealth.ts # Health polling with configurable interval
├── usePermissions.ts # RBAC capability check (runs once)
└── useSealedSecretEncryption.ts # Multi-step encryption workflow
```
- Components in `src/components/` import hooks directly
- No provider wrapping needed in `index.tsx` (except error boundaries)
- `SealedSecret` class in `src/lib/SealedSecretCRD.ts` extends `KubeObject` for `useList()`/`useGet()`
---
## References
- [React Custom Hooks](https://react.dev/learn/reusing-logic-with-custom-hooks)
- [Headlamp KubeObject API](https://headlamp.dev/docs/latest/development/api/classes/lib_k8s_cluster.KubeObject/)
---
## Related ADRs
- [ADR 005: Custom React Hooks](005-react-hooks-extraction.md) — Details the hook extraction process
- [ADR 006: Dual Error Boundaries](006-dual-error-boundaries.md) — Error handling that wraps hook-based components
---
## Changelog
- **2026-03-05**: Initial decision
+2
View File
@@ -26,6 +26,8 @@ Each ADR follows this structure:
| [003](003-client-side-crypto.md) | Client-Side Encryption | Accepted | 2026-02-11 | | [003](003-client-side-crypto.md) | Client-Side Encryption | Accepted | 2026-02-11 |
| [004](004-rbac-integration.md) | RBAC-Aware UI | Accepted | 2026-02-11 | | [004](004-rbac-integration.md) | RBAC-Aware UI | Accepted | 2026-02-11 |
| [005](005-react-hooks-extraction.md) | Custom React Hooks | Accepted | 2026-02-12 | | [005](005-react-hooks-extraction.md) | Custom React Hooks | Accepted | 2026-02-12 |
| [006](006-dual-error-boundaries.md) | Error Boundary with Dual Variants | Accepted | 2026-03-05 |
| [007](007-hooks-vs-context.md) | Custom Hooks Architecture vs Data Context | Accepted | 2026-03-05 |
## Creating New ADRs ## Creating New ADRs