Artificial Intelligence
Vector Database Protection using SnapCenter
Suggest changes
PDFs
Vector database protection using NetApp SnapCenter.
For example, in the film production industry, customers often possess critical embedded data such as video and audio files. Loss of this data, due to issues like hard drive failures, can have a significant impact on their operations, potentially jeopardizing multimillion-dollar ventures. We have encountered instances where invaluable content was lost, causing substantial disruption and financial loss. Ensuring the security and integrity of this essential data is therefore of paramount importance in this industry.
In this section, we delve into how SnapCenter safeguards the vector database data and Milvus data residing in ONTAP. For this example, we utilized a NAS bucket (milvusdbvol1) derived from an NFS ONTAP volume (vol1) for customer data, and a separate NFS volume (vectordbpv) for Milvus cluster configuration data. please check the here for the snapcenter backup workflow
-
Set up the host that will be used to execute SnapCenter commands.
-
Install and configure the storage plugin. From the added host, select "More Options". Navigate to and select the downloaded storage plugin from the NetApp Automation Store. Install the plugin and save the configuration.
-
Set up the storage system and volume: Add the storage system under "Storage System" and select the SVM (Storage Virtual Machine). In this example, we've chosen "vs_nvidia".
-
Establish a resource for the vector database, incorporating a backup policy and a custom snapshot name.
-
Enable Consistency Group Backup with default values and enable SnapCenter without filesystem consistency.
-
In the Storage Footprint section, select the volumes associated with the vector database customer data and Milvus cluster data. In our example, these are "vol1" and "vectordbpv".
-
Create policy for vector database protection and protect vector database resource using the policy.
-
-
Insert data into the S3 NAS bucket using a Python script. In our case, we modified the backup script provided by Milvus, namely 'prepare_data_netapp.py', and executed the 'sync' command to flush the data from the operating system.
root@node2:~# python3 prepare_data_netapp.py === start connecting to Milvus === === Milvus host: localhost === Does collection hello_milvus_netapp_sc_test exist in Milvus: False === Create collection `hello_milvus_netapp_sc_test` === === Start inserting entities === Number of entities in hello_milvus_netapp_sc_test: 3000 === Create collection `hello_milvus_netapp_sc_test2` === Number of entities in hello_milvus_netapp_sc_test2: 6000 root@node2:~# for i in 2 3 4 5 6 ; do ssh node$i "hostname; sync; echo 'sync executed';" ; done node2 sync executed node3 sync executed node4 sync executed node5 sync executed node6 sync executed root@node2:~# -
Verify the data in the S3 NAS bucket. In our example, the files with the timestamp '2024-04-08 21:22' were created by the 'prepare_data_netapp.py' script.
root@node2:~# aws s3 ls --profile ontaps3 s3://milvusdbvol1/ --recursive | grep '2024-04-08' <output content removed to save page space> 2024-04-08 21:18:14 5656 stats_log/448950615991000809/448950615991000810/448950615991001854/100/1 2024-04-08 21:18:12 5654 stats_log/448950615991000809/448950615991000810/448950615991001854/100/448950615990800869 2024-04-08 21:18:17 5656 stats_log/448950615991000809/448950615991000810/448950615991001872/100/1 2024-04-08 21:18:15 5654 stats_log/448950615991000809/448950615991000810/448950615991001872/100/448950615990800876 2024-04-08 21:22:46 5625 stats_log/448950615991003377/448950615991003378/448950615991003385/100/1 2024-04-08 21:22:45 5623 stats_log/448950615991003377/448950615991003378/448950615991003385/100/448950615990800899 2024-04-08 21:22:49 5656 stats_log/448950615991003408/448950615991003409/448950615991003416/100/1 2024-04-08 21:22:47 5654 stats_log/448950615991003408/448950615991003409/448950615991003416/100/448950615990800906 2024-04-08 21:22:52 5656 stats_log/448950615991003408/448950615991003409/448950615991003434/100/1 2024-04-08 21:22:50 5654 stats_log/448950615991003408/448950615991003409/448950615991003434/100/448950615990800913 root@node2:~# -
Initiate a backup using the Consistency Group (CG) snapshot from the 'milvusdb' resource
-
To test the backup functionality, we either added a new table after the backup process or removed some data from the NFS (S3 NAS bucket).
For this test, imagine a scenario where someone created a new, unnecessary, or inappropriate collection after the backup. In such a case, we would need to revert the vector database to its state before the new collection was added. For instance, new collections such as 'hello_milvus_netapp_sc_testnew' and 'hello_milvus_netapp_sc_testnew2' have been inserted.
root@node2:~# python3 prepare_data_netapp.py === start connecting to Milvus === === Milvus host: localhost === Does collection hello_milvus_netapp_sc_testnew exist in Milvus: False === Create collection `hello_milvus_netapp_sc_testnew` === === Start inserting entities === Number of entities in hello_milvus_netapp_sc_testnew: 3000 === Create collection `hello_milvus_netapp_sc_testnew2` === Number of entities in hello_milvus_netapp_sc_testnew2: 6000 root@node2:~# -
Execute a full restore of the S3 NAS bucket from the previous snapshot.
-
Use a Python script to verify the data from the 'hello_milvus_netapp_sc_test' and 'hello_milvus_netapp_sc_test2' collections.
root@node2:~# python3 verify_data_netapp.py === start connecting to Milvus === === Milvus host: localhost === Does collection hello_milvus_netapp_sc_test exist in Milvus: True {'auto_id': False, 'description': 'hello_milvus_netapp_sc_test', 'fields': [{'name': 'pk', 'description': '', 'type': <DataType.INT64: 5>, 'is_primary': True, 'auto_id': False}, {'name': 'random', 'description': '', 'type': <DataType.DOUBLE: 11>}, {'name': 'var', 'description': '', 'type': <DataType.VARCHAR: 21>, 'params': {'max_length': 65535}}, {'name': 'embeddings', 'description': '', 'type': <DataType.FLOAT_VECTOR: 101>, 'params': {'dim': 8}}]} Number of entities in Milvus: hello_milvus_netapp_sc_test : 3000 === Start Creating index IVF_FLAT === === Start loading === === Start searching based on vector similarity === hit: id: 2998, distance: 0.0, entity: {'random': 0.9728033590489911}, random field: 0.9728033590489911 hit: id: 1262, distance: 0.08883658051490784, entity: {'random': 0.2978858685751561}, random field: 0.2978858685751561 hit: id: 1265, distance: 0.09590047597885132, entity: {'random': 0.3042039939240304}, random field: 0.3042039939240304 hit: id: 2999, distance: 0.0, entity: {'random': 0.02316334456872482}, random field: 0.02316334456872482 hit: id: 1580, distance: 0.05628091096878052, entity: {'random': 0.3855988746044062}, random field: 0.3855988746044062 hit: id: 2377, distance: 0.08096685260534286, entity: {'random': 0.8745922204004368}, random field: 0.8745922204004368 search latency = 0.2832s === Start querying with `random > 0.5` === query result: -{'random': 0.6378742006852851, 'embeddings': [0.20963514, 0.39746657, 0.12019053, 0.6947492, 0.9535575, 0.5454552, 0.82360446, 0.21096309], 'pk': 0} search latency = 0.2257s === Start hybrid searching with `random > 0.5` === hit: id: 2998, distance: 0.0, entity: {'random': 0.9728033590489911}, random field: 0.9728033590489911 hit: id: 747, distance: 0.14606499671936035, entity: {'random': 0.5648774800635661}, random field: 0.5648774800635661 hit: id: 2527, distance: 0.1530652642250061, entity: {'random': 0.8928974315571507}, random field: 0.8928974315571507 hit: id: 2377, distance: 0.08096685260534286, entity: {'random': 0.8745922204004368}, random field: 0.8745922204004368 hit: id: 2034, distance: 0.20354536175727844, entity: {'random': 0.5526117606328499}, random field: 0.5526117606328499 hit: id: 958, distance: 0.21908017992973328, entity: {'random': 0.6647383716417955}, random field: 0.6647383716417955 search latency = 0.5480s Does collection hello_milvus_netapp_sc_test2 exist in Milvus: True {'auto_id': True, 'description': 'hello_milvus_netapp_sc_test2', 'fields': [{'name': 'pk', 'description': '', 'type': <DataType.INT64: 5>, 'is_primary': True, 'auto_id': True}, {'name': 'random', 'description': '', 'type': <DataType.DOUBLE: 11>}, {'name': 'var', 'description': '', 'type': <DataType.VARCHAR: 21>, 'params': {'max_length': 65535}}, {'name': 'embeddings', 'description': '', 'type': <DataType.FLOAT_VECTOR: 101>, 'params': {'dim': 8}}]} Number of entities in Milvus: hello_milvus_netapp_sc_test2 : 6000 === Start Creating index IVF_FLAT === === Start loading === === Start searching based on vector similarity === hit: id: 448950615990642008, distance: 0.07805602252483368, entity: {'random': 0.5326684390871348}, random field: 0.5326684390871348 hit: id: 448950615990645009, distance: 0.07805602252483368, entity: {'random': 0.5326684390871348}, random field: 0.5326684390871348 hit: id: 448950615990640618, distance: 0.13562293350696564, entity: {'random': 0.7864676926688837}, random field: 0.7864676926688837 hit: id: 448950615990642314, distance: 0.10414951294660568, entity: {'random': 0.2209597460821181}, random field: 0.2209597460821181 hit: id: 448950615990645315, distance: 0.10414951294660568, entity: {'random': 0.2209597460821181}, random field: 0.2209597460821181 hit: id: 448950615990640004, distance: 0.11571306735277176, entity: {'random': 0.7765521996186631}, random field: 0.7765521996186631 search latency = 0.2381s === Start querying with `random > 0.5` === query result: -{'embeddings': [0.15983285, 0.72214717, 0.7414838, 0.44471496, 0.50356466, 0.8750043, 0.316556, 0.7871702], 'pk': 448950615990639798, 'random': 0.7820620141382767} search latency = 0.3106s === Start hybrid searching with `random > 0.5` === hit: id: 448950615990642008, distance: 0.07805602252483368, entity: {'random': 0.5326684390871348}, random field: 0.5326684390871348 hit: id: 448950615990645009, distance: 0.07805602252483368, entity: {'random': 0.5326684390871348}, random field: 0.5326684390871348 hit: id: 448950615990640618, distance: 0.13562293350696564, entity: {'random': 0.7864676926688837}, random field: 0.7864676926688837 hit: id: 448950615990640004, distance: 0.11571306735277176, entity: {'random': 0.7765521996186631}, random field: 0.7765521996186631 hit: id: 448950615990643005, distance: 0.11571306735277176, entity: {'random': 0.7765521996186631}, random field: 0.7765521996186631 hit: id: 448950615990640402, distance: 0.13665105402469635, entity: {'random': 0.9742541034109935}, random field: 0.9742541034109935 search latency = 0.4906s root@node2:~# -
Verify that the unnecessary or inappropriate collection is no longer present in the database.
root@node2:~# python3 verify_data_netapp.py === start connecting to Milvus === === Milvus host: localhost === Does collection hello_milvus_netapp_sc_testnew exist in Milvus: False Traceback (most recent call last): File "/root/verify_data_netapp.py", line 37, in <module> recover_collection = Collection(recover_collection_name) File "/usr/local/lib/python3.10/dist-packages/pymilvus/orm/collection.py", line 137, in __init__ raise SchemaNotReadyException( pymilvus.exceptions.SchemaNotReadyException: <SchemaNotReadyException: (code=1, message=Collection 'hello_milvus_netapp_sc_testnew' not exist, or you can pass in schema to create one.)> root@node2:~#
In conclusion, the use of NetApp's SnapCenter to safeguard vector database data and Milvus data residing in ONTAP offers significant benefits to customers, particularly in industries where data integrity is paramount, such as film production. SnapCenter's ability to create consistent backups and perform full data restores ensures that critical data, such as embedded video and audio files, are protected against loss due to hard drive failures or other issues. This not only prevents operational disruption but also safeguards against substantial financial loss.
In this section, we demonstrated how SnapCenter can be configured to protect data residing in ONTAP, including the setup of hosts, installation and configuration of storage plugins, and the creation of a resource for the vector database with a custom snapshot name. We also showcased how to perform a backup using the Consistency Group snapshot and verify the data in the S3 NAS bucket.
Furthermore, we simulated a scenario where an unnecessary or inappropriate collection was created after the backup. In such cases, SnapCenter's ability to perform a full restore from a previous snapshot ensures that the vector database can be reverted to its state before the addition of the new collection, thus maintaining the integrity of the database. This capability to restore data to a specific point in time is invaluable for customers, providing them with the assurance that their data is not only secure but also correctly maintained. Thus, NetApp's SnapCenter product offers customers a robust and reliable solution for data protection and management.