Files
superset2/docs/developer_docs/extensions/storage.md
Michael S. Molina 34ebe3d22a feat(extensions): add a Storage API for extensions (#39171)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Evan Rusackas <evan@preset.io>
Co-authored-by: Amin Ghadersohi <amin.ghadersohi@gmail.com>
Co-authored-by: Benyovszky Máté <benyovszky.mate.work@outlook.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-21 16:23:56 -03:00

466 lines
19 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
title: Storage
sidebar_position: 8
---
<!--
Licensed to the Apache Software Foundation (ASF) under one
or more contributor license agreements. See the NOTICE file
distributed with this work for additional information
regarding copyright ownership. The ASF licenses this file
to you under the Apache License, Version 2.0 (the
"License"); you may not use this file except in compliance
with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
KIND, either express or implied. See the License for the
specific language governing permissions and limitations
under the License.
-->
# Storage
Superset Extensions have access to a managed storage API for persisting data. The storage system provides multiple tiers with different persistence characteristics, allowing extensions to choose the right storage for their needs.
Each extension receives its own isolated storage namespace. When Superset loads your extension, it binds storage to your extension's unique identifier, ensuring data privacy—two extensions using the same key will never collide, and extensions cannot access each other's data.
Storage is always accessed through the extension context via `getContext()`. This binding to the context is what ties every operation to the current extension's namespace.
## Storage Tiers
| Tier | Storage Type | Context Property | Use Case |
| ---- | ----------------- | ------------------------------------------ | -------------------------------------- |
| 1 | Browser storage | `ctx.storage.local`, `ctx.storage.session` | UI state, wizard progress, draft forms |
| 2 | Server-side cache | `ctx.storage.ephemeral` | Job progress, temporary results |
| 3 | Database | `ctx.storage.persistent` | Saved artifacts, durable config |
## Tier 1: Local State
Browser-based storage that persists on the user's device. Use this for UI state and settings that don't need to sync across devices.
### Why Use the API Instead of localStorage Directly?
You might wonder why extensions should use `ctx.storage.local` instead of directly accessing `window.localStorage`. The managed API provides several benefits:
- **Automatic namespacing**: Each extension's data is isolated. Two extensions using the same key name won't collide.
- **User isolation**: By default, data is scoped to the current user, preventing data leakage between users on shared devices.
- **Clean uninstall**: When an extension is uninstalled, all its data can be cleanly removed using prefix-based deletion.
- **Future sandboxing**: The async API is designed for a future sandboxed execution model where extensions run in isolated contexts without direct DOM access.
- **Consistent patterns**: The same API shape works across all storage tiers, making it easy to switch between them.
### localState
Data persists across browser sessions until explicitly deleted or the user clears browser storage.
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
// Save sidebar state
await ctx.storage.local.set('sidebar_collapsed', true);
// Retrieve it later
const isCollapsed = await ctx.storage.local.get('sidebar_collapsed');
// Remove it
await ctx.storage.local.remove('sidebar_collapsed');
```
### sessionState
Data is cleared when the browser tab is closed. Use for transient state within a single session.
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
// Save wizard progress (lost when tab closes)
await ctx.storage.session.set('wizard_step', 3);
await ctx.storage.session.set('unsaved_form', { name: 'Draft' });
// Retrieve on page reload within same tab
const step = await ctx.storage.session.get('wizard_step');
```
Tier 1 has no `shared` accessor. Browser storage is per-device, so a "shared" value would only be visible to other users of that same browser, not to other users of the extension generally — see `.shared` on [Tier 2](#tier-2-ephemeral-state) or [Tier 3](#tier-3-persistent-state) for storage that's actually shared across users.
### When to Use Tier 1
- UI state (sidebar collapsed, panel sizes)
- Recently used items
- Draft form values
- Any data acceptable to lose if user clears browser
### Limitations
- Per-browser, per-device (not shared across devices)
- Subject to browser storage quotas (~5-10 MB)
- Not accessible from backend code
## Tier 2: Ephemeral State
Server-side cache storage with automatic TTL expiration. Use for temporary data that needs to be shared between frontend and backend, or persist across page reloads.
TTL is required for every `set` call. The maximum allowed TTL is controlled by `MAX_TTL` in the server configuration.
### Frontend Usage
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
// Store with a 5-minute TTL
await ctx.storage.ephemeral.set('job_progress', { pct: 42, status: 'running' }, { ttl: 300 });
// Retrieve
const progress = await ctx.storage.ephemeral.get('job_progress');
// Remove
await ctx.storage.ephemeral.remove('job_progress');
```
### Backend Usage
```python
from superset_core.extensions.context import get_context
from superset_core.extensions.storage.ephemeral import EphemeralSetOptions
ctx = get_context()
# Store job progress with a 5-minute TTL
ctx.storage.ephemeral.set(
'job_progress', {'pct': 42, 'status': 'running'}, EphemeralSetOptions(ttl=300)
)
# Retrieve
progress = ctx.storage.ephemeral.get('job_progress')
# Remove
ctx.storage.ephemeral.remove('job_progress')
```
### Shared State
For data that needs to be visible to all users:
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
await ctx.storage.ephemeral.shared.set('shared_result', { data: [1, 2, 3] }, { ttl: 3600 });
const result = await ctx.storage.ephemeral.shared.get('shared_result');
```
```python
from superset_core.extensions.context import get_context
from superset_core.extensions.storage.ephemeral import EphemeralSetOptions
ctx = get_context()
ctx.storage.ephemeral.shared.set(
'shared_result', {'data': [1, 2, 3]}, EphemeralSetOptions(ttl=3600)
)
result = ctx.storage.ephemeral.shared.get('shared_result')
```
### When to Use Tier 2
- Background job progress indicators
- Cross-request intermediate state
- Query result previews
- Temporary computation results
- Any data that can be recomputed if lost
### Limitations
- Not guaranteed to survive server restarts
- Subject to cache eviction under memory pressure
- TTL-based expiration (data disappears after timeout)
- TTL is required on every `set` call
## Tier 3: Persistent State
Database-backed storage that survives server restarts, cache evictions, and browser clears. Use for any data that must not be lost.
### Frontend Usage
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
// Store a saved SQL snippet
await ctx.storage.persistent.set('snippet:top_customers', { sql: 'SELECT ...' });
// Retrieve
const snippet = await ctx.storage.persistent.get('snippet:top_customers');
// Remove
await ctx.storage.persistent.remove('snippet:top_customers');
```
### Backend Usage
```python
from superset_core.extensions.context import get_context
ctx = get_context()
# Store a saved SQL snippet
ctx.storage.persistent.set('snippet:top_customers', {'sql': 'SELECT ...'})
# Retrieve
snippet = ctx.storage.persistent.get('snippet:top_customers')
# Remove
ctx.storage.persistent.remove('snippet:top_customers')
```
### Shared State
For data that should be visible to all users of the extension:
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
await ctx.storage.persistent.shared.set('global_config', { version: 2 });
const config = await ctx.storage.persistent.shared.get('global_config');
```
```python
from superset_core.extensions.context import get_context
ctx = get_context()
ctx.storage.persistent.shared.set('global_config', {'version': 2})
config = ctx.storage.persistent.shared.get('global_config')
```
### Listing Entries
Both the frontend and backend accessors support listing entries in the caller's scope, without needing to know every key in advance. `page` and `pageSize`/`page_size` are required on every call — `list()` always returns one page of results, never the whole result set, so there's no default that could make that fact easy to miss. Check the returned `count` (total entries matching the scope/filters, across all pages) against `pageSize`/`page_size` to know whether more pages exist.
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
// Filter by resource
const result = await ctx.storage.persistent.list({
resourceType: 'dashboard',
resourceUuid: '1234-uuid',
page: 0,
pageSize: 25,
});
result.entries.forEach(entry => {
console.log(entry.key, entry.value, entry.codec);
});
console.log(result.count); // total matching entries across all pages
// Shared (global) scope
const shared = await ctx.storage.persistent.shared.list({ page: 0, pageSize: 10 });
```
```python
from superset_core.extensions.context import get_context
from superset_core.extensions.storage.persistent import PersistentListOptions
ctx = get_context()
# Filter by resource
result = ctx.storage.persistent.list(
PersistentListOptions(
resource_type="dashboard",
resource_uuid="1234-uuid",
page=0,
page_size=25,
)
)
for entry in result.entries:
print(entry.key, entry.value, entry.codec)
print(result.count) # total matching entries across all pages
# Shared (global) scope
shared = ctx.storage.persistent.shared.list(PersistentListOptions(page=0, page_size=10))
```
`resource_type`/`resource_uuid` filter to entries linked to one resource — since a single resource can have several keyed entries (e.g. `"layout"`, `"notes"`), this is how an extension enumerates all of them without tracking key names itself.
There is no fixed limit on `page_size`, but the combined size of a requested page's values is checked against `MAX_LIST_PAYLOAD_SIZE` (see [Quotas](#quotas)) before any row's value is read from the database — a page that would exceed it is rejected rather than silently truncated, so reduce `page_size` and retry if that happens.
Backend `list()` decodes every entry's value unconditionally. The REST API backing the frontend's `list()` call is more restrictive: an entry written with a codec that isn't safe to decode over the wire (e.g. `pickle`) comes back with `value: null` (its `codec` is still reported) rather than being decoded.
### Enumerating and Managing Storage from the Backend
`ctx.storage.persistent.list()` covers the common case of listing an extension's own entries. Backend extension code that needs lower-level bulk operations — for example, a cleanup routine that removes storage linked to resources that no longer exist, using filters `list()` doesn't expose — can use `superset_core.extensions.storage.dao.ExtensionStorageDAO` instead:
```python
from superset_core.extensions.storage.dao import ExtensionStorageDAO
# All entries this extension has linked to a given resource type
entries = ExtensionStorageDAO.filter_by(resource_type="my-resource-type")
# Bulk-delete rows for resources that no longer exist
orphaned = [e for e in entries if e.resource_uuid not in my_extension_live_resource_ids()]
ExtensionStorageDAO.delete(orphaned)
```
`ExtensionStorageDAO` is automatically scoped to the calling extension's own rows — `extension_id` is never a parameter you pass, so there is no way to reach another extension's storage through this API. It supports the standard DAO operations `find_all`, `find_one_or_none`, `filter_by`, `query`, `update`, and `delete`. `create()` is not supported — it's a raw insert with no upsert dedup, quota check, or locking against the key `.set()` writes to, so use `ctx.storage.persistent.set()` to write a value instead.
### When to Use Tier 3
- Extension configuration that must survive restarts
- User-specific saved artifacts that need to roam across devices and browsers (e.g. saved SQL snippets, annotations)
- Any data where loss is unacceptable
### Limitations
- Higher latency than Tiers 12 (database round-trip per operation)
- Subject to the per-value size limit configured via `EXTENSIONS_PERSISTENT_STORAGE.MAX_VALUE_SIZE` (see [Quotas](#quotas))
- Subject to the per-extension quota configured via `EXTENSIONS_PERSISTENT_STORAGE.QUOTA_PER_EXTENSION` (see [Quotas](#quotas))
- Requires a database migration when first deployed
## Working with Binary Data
Tiers 2 and 3 accept a `codec` option describing how `value` is encoded — `"json"` (the default) for JSON-native values, or `"binary"` for raw bytes. `list()` and `get()` report which codec an entry was written with via `entry.codec`.
### Backend
Backend code talks to the storage DAO directly and works with real Python `bytes` — no extra flag is needed:
```python
from superset_core.extensions.context import get_context
from superset_core.extensions.storage.persistent import PersistentSetOptions
ctx = get_context()
with open('logo.png', 'rb') as f:
png_bytes = f.read()
ctx.storage.persistent.set('logo', png_bytes, PersistentSetOptions(codec='binary'))
stored = ctx.storage.persistent.get('logo') # raw bytes back, unchanged
```
### Frontend
The frontend SDK talks to the backend over a JSON REST API, which has no byte type — a binary value must be base64-encoded to travel in a JSON request body. There is no way for the SDK to infer this from a JS value's type, so it is never detected automatically: set `isBinary: true` explicitly whenever `value` is binary.
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
const pngBytes = new Uint8Array([0x89, 0x50, 0x4e, 0x47 /* ... */]);
await ctx.storage.persistent.set('logo', pngBytes, { isBinary: true });
```
`get()` returns a binary entry's value exactly as stored on the wire — a base64 string, not a `Uint8Array` — so decode it yourself:
```typescript
const base64 = await ctx.storage.persistent.get<string>('logo');
const bytes = Uint8Array.from(atob(base64), c => c.charCodeAt(0));
```
`list()` entries report `isBinary` alongside `codec`, so you know which entries need decoding before you read `value`:
```typescript
const result = await ctx.storage.persistent.list({ page: 0, pageSize: 10 });
result.entries.forEach(entry => {
const value = entry.isBinary
? Uint8Array.from(atob(entry.value as string), c => c.charCodeAt(0))
: entry.value;
});
```
## Key Patterns
All storage keys are automatically namespaced:
| Scope | Key Pattern |
| ---------------------- | -------------------------------------------------- |
| User-scoped | `superset-ext:{extension_id}:user:{user_id}:{key}` |
| Shared (Tiers 2 and 3) | `superset-ext:{extension_id}:shared:{key}` |
This ensures:
- Extensions cannot accidentally access each other's data
- Users cannot see other users' data (by default)
- Clean prefix-based deletion on uninstall
## Configuration
### Tier 2: Ephemeral Storage
Administrators can configure the server-side cache backend in `superset_config.py`:
```python
EXTENSIONS_EPHEMERAL_STORAGE = {
# Use Redis for better performance in production
"CACHE_TYPE": "RedisCache",
"CACHE_REDIS_URL": "redis://localhost:6379/2",
"MAX_TTL": 604800, # 7 days maximum TTL
"MAX_VALUE_SIZE": 100 * 1024, # 100 KB maximum per stored value
}
```
For development, the default `SupersetMetastoreCache` stores data in the metadata database.
### Tier 3: Persistent Storage
Tier 3 values are stored in the `extension_storage` database table. Values are stored unencrypted by default; encryption is opt-in per write, available via a `PersistentSetOptions`-shaped `encrypt` option:
```typescript
import { extensions } from '@apache-superset/core';
const ctx = extensions.getContext();
await ctx.storage.persistent.set('api_token', 'sk-...', { encrypt: true });
const token = await ctx.storage.persistent.get('api_token');
```
```python
from superset_core.extensions.context import get_context
from superset_core.extensions.storage.persistent import PersistentSetOptions
ctx = get_context()
ctx.storage.persistent.set('api_token', 'sk-...', PersistentSetOptions(encrypt=True))
token = ctx.storage.persistent.get('api_token')
```
Encryption reuses Superset's existing `EncryptedType` (from `sqlalchemy-utils`) rather than a separate mechanism — the same infrastructure used for database connection credentials. The key is not configured directly: user-scoped values are encrypted with a key derived per-user via HMAC-SHA256 from the deployment's `SECRET_KEY`, so ciphertext for one user cannot be decrypted as another's; shared/global values (written via `.shared`) use `SECRET_KEY` directly. The encryption engine (AES-CBC by default) can be changed for the whole deployment via the existing `SQLALCHEMY_ENCRYPTED_FIELD_ENGINE` config, the same setting that controls encryption for database credentials elsewhere in Superset — there is no separate key list or rotation mechanism specific to extension storage. When `SECRET_KEY` is rotated, extension storage rows are re-encrypted by the existing `SecretsMigrator` tooling alongside database credentials, with no extension-specific steps required.
### Quotas
Persistent storage is subject to per-value, per-list-page, and per-extension size limits, configured in `superset_config.py`:
```python
EXTENSIONS_PERSISTENT_STORAGE = {
# Maximum total bytes an extension may store across all of its rows
# (global, every user's, and shared). Defaults to 100 MB.
"QUOTA_PER_EXTENSION": 100 * 1024 * 1024,
# Maximum size (in bytes) of a single stored value. Defaults to 1 MB.
"MAX_VALUE_SIZE": 1024 * 1024,
# Maximum combined value size (in bytes) that a single list() page may
# return. Defaults to 10 MB.
"MAX_LIST_PAYLOAD_SIZE": 10 * 1024 * 1024,
}
```
A write that would push the extension's total stored size over `QUOTA_PER_EXTENSION` is rejected — the frontend SDK's `set()` call rejects with an HTTP 413 response, and backend code calling `ctx.storage.persistent.set()` sees the corresponding exception raised. Overwriting an existing key nets out that key's own current size against usage first, so replacing a value with one of the same or smaller size never spuriously fails even when the extension is already near its quota.
A write whose encoded value exceeds `MAX_VALUE_SIZE` is rejected outright, independently of the total quota.
A `list()` call whose requested page would return more than `MAX_LIST_PAYLOAD_SIZE` combined bytes of value data is rejected — reduce `page`/`pageSize` and retry. This check runs against the page's stored sizes before any value is read from the database, so an oversized request fails fast. Because `list()`'s response is consumed as JSON in the browser, raising `MAX_LIST_PAYLOAD_SIZE` substantially above the default risks client-side memory and parse-time issues — treat the default as already generous rather than a floor to build up from.