Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

fix: reuse retained Ollama staging only from the same source, reject duplicate imports up front (#13124)

Adam Eivy3d0f0ee6f7d9
diff --git a/server/services/ollamaManager.js b/server/services/ollamaManager.js
index 7542913b93..53738e8d7a 100644
--- a/server/services/ollamaManager.js
+++ b/server/services/ollamaManager.js
@@ -1558,6 +1558,21 @@ async function downloadHfImportFile({ repo, remotePath, destination, headers, pr
* @param {(p: { status: string, percent: number|null, completed?: number, total?: number, importing?: boolean }) => void} [onProgress]
*/
async function importModelFromHfSafetensors(spec, onProgress) {
+ const { modelId } = spec
+ // One import per model at a time: a second would mistake the first's live
+ // staging for one left behind. Checked before any network round-trip.
+ if (activeImports.has(modelId)) {
+ return { success: false, error: `${modelId} is already being imported.`, modelId, code: 'OLLAMA_IMPORT_IN_PROGRESS' }
+ }
+ activeImports.add(modelId)
+ try {
+ return await prepareSafetensorsImport(spec, onProgress)
+ } finally {
+ activeImports.delete(modelId)
+ }
+}
+
+async function prepareSafetensorsImport(spec, onProgress) {
const { modelId, repo, subdir, minVersion = '0.19.0' } = spec
if (!(await checkOllamaAvailable())) {
return { success: false, error: 'Ollama not available', modelId }
@@ -1607,20 +1622,10 @@ async function importModelFromHfSafetensors(spec, onProgress) {
return { success: false, error: `The ${subdir} checkpoint is incomplete or unavailable on Hugging Face.`, modelId }
}
- // One import per model at a time: a second would mistake the first's live
- // staging for one left behind.
- if (activeImports.has(modelId)) {
- return { success: false, error: `${modelId} is already being imported.`, modelId, code: 'OLLAMA_IMPORT_IN_PROGRESS' }
- }
- activeImports.add(modelId)
- try {
- // Resolve-not-throw, like every other failure here. A staging write that
- // throws leaves its marker behind; the next attempt reuses or retires it.
- return await runSafetensorsImport({ modelId, repo, subdir, files, headers, onProgress })
- .catch((err) => ({ success: false, error: err.message, modelId }))
- } finally {
- activeImports.delete(modelId)
- }
+ // Resolve-not-throw, like every other failure here. A staging write that
+ // throws leaves its marker behind; the next attempt reuses or retires it.
+ return runSafetensorsImport({ modelId, repo, subdir, files, headers, onProgress })
+ .catch((err) => ({ success: false, error: err.message, modelId }))
}
const heldImportResult = (modelId, error, reason) => ({
@@ -1641,7 +1646,10 @@ async function runSafetensorsImport({ modelId, repo, subdir, files, headers, onP
.catch((err) => console.error(`❌ Safetensors staging sweep failed: ${err.message}`))
const retained = await findImportStaging(modelId)
if (retained) {
- const complete = await stagedFilesComplete(join(retained.tempRoot, 'model'), files)
+ // Equal file sizes alone could pass a checkpoint from another repo or
+ // subdir off as this one; a marker that cannot name its source is retired.
+ const sameSource = retained.marker.repo === repo && retained.marker.subdir === subdir
+ const complete = sameSource && await stagedFilesComplete(join(retained.tempRoot, 'model'), files)
if (complete) {
// Reusing is non-destructive, but stacking a second create on one the
// daemon is still running is not; wait only for a holder we can see.
@@ -1658,7 +1666,7 @@ async function runSafetensorsImport({ modelId, repo, subdir, files, headers, onP
} else {
const release = await awaitStagingRelease(retained.tempRoot, { attempts: 1 })
if (!release.free) {
- return heldImportResult(modelId, `A previous import of ${modelId} left an incomplete checkpoint.`, release.reason)
+ return heldImportResult(modelId, `A previous import of ${modelId} left a checkpoint it cannot reuse.`, release.reason)
}
await removeImportStaging(retained.tempRoot)
}
diff --git a/server/services/ollamaManager.test.js b/server/services/ollamaManager.test.js
index 19ba0d9764..9716cd64c2 100644
--- a/server/services/ollamaManager.test.js
+++ b/server/services/ollamaManager.test.js
@@ -829,6 +829,7 @@ describe('ollamaManager curated Hugging Face Safetensors import', () => {
let modelsDir
let lsof
let downloads
+ let metadataFetches
let stagingRoots
// Each test imports its own model id so retained staging from a parallel
@@ -842,11 +843,13 @@ describe('ollamaManager curated Hugging Face Safetensors import', () => {
vi.useFakeTimers({ toFake: ['setTimeout', 'clearTimeout', 'setInterval', 'clearInterval', 'Date'] })
hfTokenMock.token = 'hf_example_token'
downloads = 0
+ metadataFetches = 0
stagingRoots = new Set()
vi.stubGlobal('fetch', vi.fn(async (url) => {
const href = String(url)
if (href.endsWith('/api/version')) return versionResponse()
if (href.includes('/api/models/example/Qwen-MLX')) {
+ metadataFetches++
return Response.json({ siblings: [
{ rfilename: '4-bit/config.json', size: files.config.length },
{ rfilename: '4-bit/model.safetensors', lfs: { size: files.weights.length } }
@@ -971,6 +974,53 @@ describe('ollamaManager curated Hugging Face Safetensors import', () => {
await expect(stat(root)).rejects.toMatchObject({ code: 'ENOENT' })
})
+ it('re-downloads instead of reusing retained staging staged from a different source', async () => {
+ const modelSpec = specFor('moved-source')
+ const stale = await mkdtemp(join(tmpdir(), 'portos-ollama-hf-import-'))
+ stagingRoots.add(stale)
+ await mkdir(join(stale, 'model'), { recursive: true })
+ // Same file names and byte sizes as the current recipe, different repo.
+ await writeFile(join(stale, 'model', 'config.json'), files.config)
+ await writeFile(join(stale, 'model', 'model.safetensors'), files.weights)
+ const at = new Date().toISOString()
+ await writeFile(join(stale, 'portos-import.json'), JSON.stringify({
+ modelId: modelSpec.modelId, repo: 'example/Other-MLX', subdir: '4-bit', phase: 'retained', createdAt: at, retainedAt: at
+ }))
+ const { importModelFromHfSafetensors } = await loadManager()
+ let createdFrom
+ await scriptCreate((c, _cmd, args) => { createdFrom = stagingOf(args); queueMicrotask(() => c.exit(0)) })
+
+ let result
+ importModelFromHfSafetensors(modelSpec).then((r) => { result = r })
+ await advanceUntil(() => result, { stepMs: 0 })
+
+ expect(result).toEqual({ success: true, modelId: modelSpec.modelId })
+ expect(downloads).toBe(2)
+ expect(createdFrom).not.toBe(stale)
+ await expect(stat(stale)).rejects.toMatchObject({ code: 'ENOENT' })
+ })
+
+ it('rejects a second import of the same model before any Hugging Face request', async () => {
+ const modelSpec = specFor('duplicate')
+ const { importModelFromHfSafetensors } = await loadManager()
+ let child
+ await scriptCreate((c, _cmd, args) => { child = c; stagingOf(args) })
+
+ let first
+ importModelFromHfSafetensors(modelSpec).then((r) => { first = r })
+ await advanceUntil(() => child, { stepMs: 0 })
+ const fetchesBefore = globalThis.fetch.mock.calls.length
+
+ const second = await importModelFromHfSafetensors(modelSpec)
+
+ expect(second).toMatchObject({ success: false, code: 'OLLAMA_IMPORT_IN_PROGRESS' })
+ expect(globalThis.fetch.mock.calls.length).toBe(fetchesBefore)
+ expect(metadataFetches).toBe(1)
+ child.exit(0)
+ await advanceUntil(() => first, { stepMs: 0 })
+ expect(first).toEqual({ success: true, modelId: modelSpec.modelId })
+ })
+
it('reclaims week-old retained staging for other models before importing, once nothing holds it', async () => {
const day = 24 * 60 * 60_000
const stage = async (modelId, retainedAgoMs) => {