Read only disks, no size limit on local disks

.claude/skills/debug-e2e/SKILL.md

@@ -0,0 +1,106 @@
+---
+name: debug-e2e
+description: Debug flaky Playwright E2E test failures from CI
+---
+
+# Debugging Flaky E2E Test Failures
+
+Use this skill when investigating flaky Playwright E2E test failures from CI.
+
+## Workflow
+
+### 1. Get CI failure details
+
+```bash
+# View PR checks status
+gh pr checks <PR_NUMBER>
+
+# Get failed test logs
+gh run view <RUN_ID> --log-failed 2>&1 | head -500
+
+# Search for specific failure patterns
+gh run view <RUN_ID> --log-failed 2>&1 | rg -C 30 "FAILED|Error:|Expected|Timed out"
+```
+
+### 2. Reproduce locally with repeat-each
+
+Run the failing test multiple times to reproduce flaky behavior:
+
+```bash
+# Run specific test 20 times
+npm run e2e -- test/e2e/<file>.e2e.ts --repeat-each=20 -g "<test name pattern>"
+
+# Target specific browser if failure is browser-specific
+npm run e2e -- test/e2e/<file>.e2e.ts --repeat-each=20 -g "<test name>" --project=<browser>
+```
+
+### 2.1 Calibrate test duration
+
+Start small to get signal quickly, then scale up only if needed.
+
+- 5-10 repeats on a single browser is usually under a few minutes.
+- 20+ repeats across all browsers can take a long time, especially for full files.
+
+Always run repeats on a single test or at most one file. Never repeat the whole suite.
+Prefer narrowing with `-g` and `--project` first, then increase `--repeat-each` once the fix looks stable.
+
+### 3. Analyze failure artifacts
+
+When tests fail, Playwright saves traces and error context:
+
+```bash
+# View error context (page snapshot at failure time)
+cat test-results/<test-name>-<browser>/error-context.md
+
+# Open trace viewer (interactive)
+npx playwright show-trace test-results/<test-name>-<browser>/trace.zip
+```
+
+### 4. Common flakiness patterns
+
+**React Aria NumberInput flakiness**: When `fill()` on a NumberInput doesn't stick, it's often because:
+
+- The component re-renders after a prop change (e.g., `maxValue` changing)
+- Hydration race conditions
+
+**Fix pattern - NumberInput helper**:
+
+```typescript
+import { fillNumberInput } from './utils'
+
+await fillNumberInput(input, 'value')
+```
+
+**Form hydration issues**: Wait for a field that only renders after mount:
+
+```typescript
+await expect(page.getByRole('radiogroup', { name: 'Block size' })).toBeVisible()
+```
+
+**State change timing**: When clicking changes form state, wait for visual confirmation:
+
+```typescript
+await page.getByRole('radio', { name: 'Local' }).click()
+// Wait for dependent UI to update
+await expect(page.getByRole('radiogroup', { name: 'Block size' })).toBeHidden()
+```
+
+### 5. Sleep as last resort
+
+If deterministic waits don't work, use `sleep()` from `test/e2e/utils.ts`:
+
+```typescript
+import { sleep } from './utils'
+
+await sleep(200) // Use sparingly, prefer deterministic waits
+```
+
+### 6. Verify fix is stable
+
+Run at least 30-50 iterations to confirm flakiness is resolved:
+
+```bash
+npm run e2e -- test/e2e/<file>.e2e.ts --repeat-each=50 -g "<test name>"
+```
+
+A good target is 0 failures out of 100+ runs. Test on all browsers that failed in CI.

AGENTS.md

@@ -22,6 +22,7 @@
 # Testing code
 
 - Run local checks before sending PRs: `npm run lint`, `npm run tsc`, `npm test run`, and `npm run e2ec`; pass `-- --ui` for Playwright UI mode or project/name filters like `npm run e2ec -- instance -g 'boot disk'`.
+- You don't usually need to run all the e2e tests, so try to filter by filename. CI will run the full set.
 - Keep Playwright specs focused on user-visible behavior—use accessible locators (`getByRole`, `getByLabel`), the helpers in `test/e2e/utils.ts` (`expectToast`, `expectRowVisible`, `selectOption`, `clickRowAction`), and close toasts so follow-on assertions aren’t blocked.
 - Cover role-gated flows by logging in with `getPageAsUser`; exercise negative paths (e.g., forbidden actions) alongside happy paths as shown in `test/e2e/system-update.e2e.ts`.
 - Consider `expectVisible` and `expectNotVisible` deprecated: prefer `expect().toBeVisible()` and `toBeHidden()` in new code.

OMICRON_VERSION

@@ -1 +1 @@
-c765b3539203e34f65cd402f139cf604035d5993
+44e65c3b30720e20b3d3be9bab4efbf6cac0ee2c

app/components/StateBadge.tsx

@@ -90,3 +90,9 @@ export const DiskTypeBadge = (props: { diskType: DiskType; className?: string })
     {props.diskType}
   </Badge>
 )
+
+export const ReadOnlyBadge = () => (
+  <Badge color="neutral" className="relative">
+    Read only
+  </Badge>
+)

app/components/form/fields/DiskSizeField.tsx

@@ -12,8 +12,6 @@ import type {
   ValidateResult,
 } from 'react-hook-form'
 
-import { MAX_DISK_SIZE_GiB } from '@oxide/api'
-
 import { NumberField } from './NumberField'
 import type { TextFieldProps } from './TextField'
 
@@ -22,6 +20,8 @@ interface DiskSizeProps<
   TName extends FieldPath<TFieldValues>,
 > extends TextFieldProps<TFieldValues, TName> {
   minSize?: number
+  /** Undefined means no client-side limit (e.g., for local disks) */
+  maxSize: number | undefined
   validate?(diskSizeGiB: number): ValidateResult
 }
 
@@ -32,6 +32,7 @@ export function DiskSizeField<
   required = true,
   name,
   minSize = 1,
+  maxSize,
   validate,
   ...props
 }: DiskSizeProps<TFieldValues, TName>) {
@@ -41,7 +42,7 @@ export function DiskSizeField<
       required={required}
       name={name}
       min={minSize}
-      max={MAX_DISK_SIZE_GiB}
+      max={maxSize}
       validate={(diskSizeGiB) => {
         // Run a number of default validators
         if (Number.isNaN(diskSizeGiB)) {
@@ -50,8 +51,8 @@ export function DiskSizeField<
         if (diskSizeGiB < minSize) {
           return `Must be at least ${minSize} GiB`
         }
-        if (diskSizeGiB > MAX_DISK_SIZE_GiB) {
-          return `Can be at most ${MAX_DISK_SIZE_GiB} GiB`
+        if (maxSize !== undefined && diskSizeGiB > maxSize) {
+          return `Can be at most ${maxSize} GiB`
         }
         // Run any additional validators passed in from the callsite
         return validate?.(diskSizeGiB)