Skip to content

Commit 1cdebaf

Browse files
deeikeledeeikele
authored
Initial draft for soft delete
1 parent 2a9f014 commit 1cdebaf

File tree

1 file changed

+212
-0
lines changed

1 file changed

+212
-0
lines changed

‎articles/machine-learning/concept-soft-delete.md

Lines changed: 212 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,212 @@
1+
---
2+
title: 'Workspace soft-delete overview'
3+
titleSuffix: Azure Machine Learning
4+
description: Recover workspace data after accidental deletion with soft delete
5+
services: machine-learning
6+
ms.service: machine-learning
7+
ms.subservice: core
8+
ms.custom:
9+
ms.topic: conceptual
10+
ms.author: deeikele
11+
author: deeikele
12+
ms.date: 08/25/2022
13+
#Customer intent: As an IT pro, understand how to enable data protection capabilities, to protect against accidental deletion.
14+
---
15+
16+
# Recover workspace data after accidental deletion with soft delete (Preview)
17+
18+
Workspace soft delete provides a data protection capability to help attempt recovery of workspace data after accidental deletion of a workspace. Soft delete introduces a two-step approach in deleting a workspace. When a workspace is deleted, it is first soft deleted. While in soft-deleted state, you can choose to recover or permanently delete a workspace and its data during a data retention period.
19+
20+
## Soft-delete behavior
21+
22+
When a workspace is soft-deleted, data and metadata stored service-side get soft-deleted, but some configurations get hard-deleted. Below table provides an overview of which configurations and objects get soft-deleted, and which are hard-deleted.
23+
24+
> [!IMPORTANT]
25+
> Workspaces encrypted with customer-managed keys (CMK) are not enabled for soft delete and are always hard deleted.
26+
27+
Data / configuration | Soft-deleted | Hard-deleted
28+
---|---|---
29+
Run History | X |
30+
Models | X |
31+
Data | X |
32+
Environments | x |
33+
Components | X |
34+
Notebooks | X |
35+
Pipelines | X |
36+
Designer pipelines | X |
37+
AutoML jobs | X |
38+
Data labeling projects | X |
39+
Datastores | X |
40+
Queued or running jobs | | X
41+
Role assignments | | X*
42+
Internal cache | | X
43+
Compute instance | | X
44+
Compute clusters | | X
45+
Inference endpoints | | X
46+
Linked Databricks workspaces | | X*
47+
48+
\* Microsoft attempts recreation or re-attachment when a workspace is recovered
49+
50+
Soft delete is enabled on any workspaces for subscription that are enrolled for the soft-delete preview capability. During preview, workspaces with customer-managed keys are not supported for soft-delete. Soft deleted workspaces do not incur costs, since cost-incurring resources are hard deleted at time of workspace deletion.
51+
52+
Recently deleted workspaces can be queried through CLI/SDK/REST options and the Azure Portal, and can be recovered or permanently deleted during the data retention period. After expiry of the retention period, a soft deleted workspace automatically gets hard deleted. A data retention period of 14 days is the default, and can be set to a value between 1-14 as a property on the workspace through CLI/SDK/REST and template options.
53+
54+
## Enroll soft-delete on an Azure subscription
55+
56+
* Soft delete is the default behavior going forward on all workspaces, except for workspaces encrypted with a customer-managed key (CMK).
57+
* A data retention period of 14 days is the default, and can be set to a value between 1-14 as a property on the workspace through CLI/SDK/REST and template options. We are dependent on Azure Storage who has a retention period of deleted storage accounts of 14 days. This limitation may be solved with proposed HOBO workspace-storage.
58+
* Recently deleted workspaces can be queried through CLI/SDK/REST options and the Azure Portal, and can be recovered or permanently deleted during the data retention period. After expiry of the retention period, a soft deleted workspace automatically gets hard deleted.
59+
60+
## Manage soft-deleted workspaces
61+
62+
Recently deleted workspaces can be managed from the Azure Portal, CLI and SDK.
63+
64+
### Azure Portal
65+
66+
### CLI and SDK
67+
68+
#### Configure soft delete retention period
69+
70+
A data retention period for a soft deleted workspace may be specified on storage.
71+
72+
CLI:
73+
```cli
74+
> az ml workspace create --subscription {SUBSCRIPTION ID} -g {RESOURCE GROUP} -n {WORKSPACE NAME} -f soft-delete.yaml
75+
```
76+
77+
```yaml
78+
#source ../configs/workspace/soft-delete.yaml
79+
name: soft-delete-ws
80+
location: EastUS
81+
display_name: soft-delete-workspace
82+
description: A workspace with configured retention period for soft delete
83+
84+
data_retention:
85+
soft_delete_retention_days: 14
86+
87+
tags:
88+
purpose: testing
89+
team: ws-management
90+
```
91+
92+
SDK:
93+
```python
94+
workspace = Workspace.create(
95+
subscription_id="{SUBSCRIPTION ID}",
96+
resource_group="{RESOURCE GROUP}",
97+
name="soft-delete-ws",
98+
soft_delete_retention_days=14
99+
)
100+
```
101+
102+
> The default retention period is 14 days, but can be set to avalue between 1 and 14 days to meet data residency or other compliance requirements. CLI and SDK operations should fail if a value is provided outside of this range.
103+
104+
#### Soft delete a workspace
105+
106+
Soft delete is the default behavior when deleting a non-CMK workspace.
107+
108+
> Deletion of dependent resources is only possible in combination with permanently deleting a workspace, and fails in case a workspace is not permanently deleted to allow for best changes of recovering workspace data.
109+
110+
CLI:
111+
```cli
112+
az ml workspace delete --subscription {SUBSCRIPTION ID} -g {RESOURCE GROUP} -n {WORKSPACE NAME}
113+
```
114+
115+
SDK:
116+
```python
117+
workspace.delete(
118+
subscription_id="{SUBSCRIPTION ID}",
119+
resource_group="{RESOURCE GROUP}",
120+
delete_dependent_resources=False,
121+
permanently_delete=False,
122+
no_wait=False
123+
)
124+
```
125+
126+
#### Permanently delete a workspace
127+
128+
Optionally, you may choose to permanently delete a non-CMK workspace without going to soft delete state first by setting the `permanently_delete` flag. By default, this parameter is set to `false`. Permanently deleting a workspace allows recreation to accomodate for dev/test MLOps scenarios, or to immediately delete highly sensitive data if required.
129+
130+
CLI:
131+
```cli
132+
az ml workspace delete --subscription {SUBSCRIPTION ID} -g {RESOURCE GROUP} -n {WORKSPACE NAME} --permanently_delete
133+
```
134+
135+
SDK:
136+
```python
137+
workspace.delete(
138+
subscription_id="{SUBSCRIPTION ID}",
139+
resource_group="{RESOURCE GROUP}",
140+
delete_dependent_resources=False,
141+
permanently_delete=True,
142+
no_wait=False
143+
)
144+
```
145+
146+
#### List workspaces
147+
148+
Soft deleted workspaces should not show in az ml workspace list commands.
149+
150+
#### List soft-deleted workspaces
151+
152+
Soft-deleted workspaces can be listed by subscription, and by resource group. List by subscription allows customers to query soft deleted workspace for which the resource group was deleted as well.
153+
154+
> Design should be extensible to future interfaces provided by Azure Resource Manager. ARM will expose Microsoft.Resources/deletedResources API, which can be called to list soft deleted workspaces on subscription and resource group level.
155+
156+
Query results must provide subscription, original resource group and original name of the soft deleted resource.
157+
158+
CLI:
159+
```cli
160+
az ml workspace list-deleted --subscription {SUBSCRIPTION ID} --resource_group {RESOURCE GROUP}
161+
```
162+
163+
SDK:
164+
```python
165+
Workspace.list_deleted(
166+
subscription_id="{SUBSCRIPTION ID}",
167+
resource_group="{RESOURCE GROUP}", # Optional
168+
)
169+
```
170+
171+
#### Get a soft-deleted workspace
172+
173+
Regular `az ml workspace show` should return 404 for soft-deleted workspaces. Instead, soft-deleted workspaces can be shown via a show-deleted call, by providing subscription, original resource group and *original* name of the soft deleted resource. *Original*, since the resource group may have been deleted at this point. This aligns with ARMs future interface, where deleted resources can be queried by original resource URIs.
174+
175+
CLI:
176+
```cli
177+
az ml workspace show-deleted --subscription {SUBSCRIPTION ID} --resource_group --name {WORKSPACE NAME}
178+
```
179+
180+
SDK
181+
```python
182+
workspace.get_deleted(
183+
name="soft-delete-ws",
184+
subscription_id="{SUBSCRIPTION ID}",
185+
resource_group="{RESOURCE GROUP}"
186+
)
187+
188+
```
189+
190+
#### Recover soft-deleted workspace
191+
192+
To recover a deleted workspace, provide subscription id, original resource group and original name of the soft deleted workspace.
193+
194+
CLI:
195+
```cli
196+
az ml workspace recover --subscription {SUBSCRIPTION ID} --resource_group {RESOURCE GROUP} --name {WORKSPACE NAME}
197+
```
198+
199+
SDK:
200+
```python
201+
Workspace.recover(
202+
subscription_id="{SUBSCRIPTION ID}",
203+
resource_group="{RESOURCE GROUP}",
204+
name="soft-delete-ws",
205+
)
206+
```
207+
208+
> Since workspace is already deleted, it cannot be instantiated and recovery may be implemented as a static method.
209+
210+
## Next steps
211+
212+
+ [Create and manage a workspace](how-to-manage-workspace.md)

0 commit comments

Comments
 (0)