Object Storage Service:OSS Connector configuration

Prerequisites

OSS Connector for AI/ML is installed. For more information, see Install OSS Connector for AI/ML.

Configure access credentials

1. Create an access credentials configuration file named credentials.

   mkdir -p /root/.alibabacloud && touch /root/.alibabacloud/credentials

2. Configure parameters and save the configuration file.

   • Example:

     {
       "AccessKeyId": "<Access-key-id>",
       "AccessKeySecret": "<Access-key-secret>",
       "SecurityToken": "<Security-Token>",
       "Expiration": "2024-08-02T15:04:05Z"
     }
     

     The following table describes the preceding parameters.

     Parameter	Required	Example	Description
     AccessKeyId	Yes	STS.L4aB******************	The AccessKey ID and AccessKey secret of your Alibaba Cloud account or a RAM user. If you use a security token for authentication, set these parameters to the AccessKey ID and AccessKey secret in the temporary access credentials.
     AccessKeySecret	Yes	At32************************
     SecurityToken	No	STS.6MC2***************************************	The security token. This parameter is required when you use the temporary access credentials obtained from Security Token Service (STS) to access OSS. If you use the AccessKey ID and AccessKey secret of an Alibaba Cloud account or a RAM user for authentication leave this parameter empty.
     Expiration	No	2024-08-02T15:04:05Z	The expiration time of the authentication information. After the expiration time, the OSS connector re-reads the authentication information. If you do not specify the Expiration parameter, the authentication information does not expire. If you use a security token for authentication, we recommend that you specify this parameter. If you use the AccessKey ID and AccessKey secret of an Alibaba Cloud account or a RAM user for authentication, leave this parameter empty.

   • Use AccessKey ID and AccessKey secret:

     Replace <Access-key-id> and <Access-key-secret> in the example with the AccessKey ID and AccessKey secret of a RAM user. For more information about how to create an AccessKey ID and AccessKey secret, see Create an AccessKey pair.

     {
       "AccessKeyId": "LTAI************************",
       "AccessKeySecret": "At32************************"
     }
     

   • Use temporary access credentials:

     Note

     To ensure data security in scenarios in which access credentials are used in the production environment for a long period of time, we recommend that you use temporary access credentials to prevent the AccessKey ID and the AccessKey secret from being leaked. If you want to authorize temporary access, you must obtain temporary access credentials. For more information, see Use temporary credentials provided by STS to access OSS. After you obtain the temporary access credentials, replace <Access-key-id>, <Access-key-secret>, and <Security-Token> with the AccessKey ID, AccessKey secret, and security token.

     {
       "AccessKeyId": "STS.L4aB******************",
       "AccessKeySecret": "wyLTSm*************************",
       "SecurityToken": "************",
       "Expiration": "2024-08-15T15:04:05Z"
     }
     

3. Run the chmod 400 /root/.alibabacloud/credentials command to grant read-only permissions on the credentials file to ensure the security of the AccessKey ID and AccessKey secret.

Configure OSS Connector

1. Create a configuration file named config.json for OSS connector.

   mkdir -p /etc/oss-connector/ && touch /etc/oss-connector/config.json

2. Configure parameters and save the configuration file.

   In most cases, you can use the default configurations.

   {
       "logLevel": 1,
       "logPath": "/var/log/oss-connector/connector.log",
       "auditPath": "/var/log/oss-connector/audit.log",
       "expireTimeSec": 120,
       "prefetch": {
           "vcpus": 16,
           "workers": 16,
           "maxCacheAdviseGB": -1
       },
       "datasetConfig": {
           "prefetchConcurrency": 24,
           "prefetchWorker": 2
       },
       "checkpointConfig": {
           "prefetchConcurrency": 24,
           "prefetchWorker": 4,
           "uploadConcurrency": 64
       }
   }
   

   The following table describes the parameters. Read the instructions in the table carefully before you change the configurations.

   Parameter		Required	Example	Description
   logLevel		No	1	The log level. The default value is 1. We recommend that you set the parameter to 2. Valid values: 0, 1, 2, and 3. 0 specifies Debug, 1 specifies INFO, 2 specifies WARN, and 3 specifies ERROR.
   logPath		No	/var/log/oss-connector/connector.log	The path of the OSS Connector for AI/ML log. Default value: /var/log/oss-connector/connector.log.
   auditPath		No	/var/log/oss-connector/audit.log	The path of the OSS Connector for AI/ML audit log, which records read and write requests that have a latency of greater than 100 milliseconds. Default value: /var/log/oss-connector/audit.log.
   expireTimeSec		No	120	The interval, in seconds, for the connector to re-read the access credentials file and refresh them if they have expired. Default: 120.
   prefetch	vcpus	No	16	The number of vCPUs for prefetching in inference scenarios. Default: 16.
   	workers	No	16	The number of worker threads for prefetching in inference scenarios. Default: 16.
   	maxCacheAdviseGB	No	-1	The maximum size of the memory cache, in GB, that can be used for prefetching. The default value of -1 indicates no limit. If the model file size exceeds the node's available memory, set this parameter to prevent Out of Memory (OOM) errors. You can also set this value by using the CONNECTOR_MAX_CACHE_ADVISE_GB environment variable, which takes precedence over this setting.
   DatasetConfig	prefetchConcurrency	No	24	The number of concurrent download tasks when you use a dataset to prefetch data from OSS. Default value: 24.
   	prefetchWorker	No	2	The number of available vCPUs when you use a dataset to prefetch data from OSS. Default value: 2.
   checkpointConfig	prefetchConcurrency	No	24	The number of concurrent download tasks when you use checkpoint read to prefetch data from OSS. Default value: 24.
   	prefetchWorker	No	4	The number of available vCPUs when you use checkpoint read to prefetch data from OSS. Default value: 4.
   	uploadConcurrency	No	64	The number of concurrent upload tasks when you use checkpoint write to upload data to OSS. Default value: 64.

Configure environment variables

You can also configure OSS Connector for AI/ML using environment variables. These variables override the corresponding settings in the configuration file.

References

After installing and configuring OSS Connector for AI/ML, you can:

• Use OssMapDataset to build a map dataset suitable for random reading. For more information, see Use OSS data to build an OssMapDataset dataset for random reading.

• Use OssIterableDataset to build an iterable dataset suitable for sequential streaming reading. For more information, see Use data in OSS to build an iterable dataset suitable for sequential streaming reading.

• Use OssCheckpoint to perform read and write operations on checkpoints in OSS. For more information, see Store and access checkpoints in OSS.

• Use OSS Connector for AI/ML to improve model deployment efficiency in inference scenarios. For more information, see Use OSS Connector for AI/ML to improve model deployment efficiency.

• Use the model broadcasting feature to efficiently distribute model data to multiple inference instances. For more information, see Model broadcasting.

• Enable the connector feature in Kubernetes for model inference deployment. For more information, see Enable the connector in Kubernetes.