New Features in MMCloud Fire Island 2.3.2 and 2.3.3 Patch Releases

Fire Island 2.3.2 released on 09-20-2023.

Fire Island 2.3.3 released on 10-16-2023.

• URL for registering in order to obtain an OpCenter license changes from account.memverge.com to www.mmcloud.io.
• If a spot instance is reclaimed, the default behavior for OpCenter is to start a new spot instance of the same type. In earlier releases, if that instance type is unavailable, the job fails. In 2.3.2, OpCenter searches for an alternative instance if the first choice is unavailable.
• Bug fixes
  • A spot reclaim notification, received while a job migration is in progress, no longer causes a failure.
  • Ghost files are files that are automatically marked as hidden, often because they are system-related. If an application uses a ghost file, a checkpoint must include the ghost file. Release 2.3.2 fixes an issue in earlier releases where the size of ghost files was not included when calculating the storage requirements for a checkpoint.
  • Fixes the problem where a (non-zero) billing statement was generated upon completion of a free trial.

• Every batch job submitted to OpCenter must specify a data volume if the job requires any file system I/O. In 2.3.3, JuiceFS, a high-performance, open-source, distributed file system, is added as an option when specifying a data volume for a job.
• The MMCloud CLI has a command to add users to the OpCenter's built-in authentication database. In earlier releases, the new user's user id (uid) is generated automatically. In 2.3.3, the option to specify the uid is added.
• In earlier releases, a user's uid is cannot be changed once the account is added. In 2.3.3, the uid can be changed after the account is added.
• OpCenter provides Job Templates to simplify the procedures for starting specialized servers, for example, servers providing RStudio or Jupyter sessions. A template to start a Nextflow host is now available. After the Nextflow host is running, you can ssh into the server and submit a Netxtflow pipeline in the usual way.
• Access to some services, for example, AWS S3, requires an [access_key, secret_key] pair. In earlier releases, the [access_key, secret_key] pair is included as plaintext in float commands. To improve security, 2.3.3 includes a secret key manager, that is, an encrypted store of [access_key, secret_key] pairs managed by the OpCenter. CLI commands can now include an [access_key, secret_key] pair by referring to the secret manager, rather than displaying the plaintext.
• Bug fixes
  • Fixed condition where job state shows "Executing" even though checkpoint failed.
  • In earlier releases, setting the VM creation policy to onDemand=true disabled further attempts if the initial attempt at starting an on-demand instance failed. In 2.3.3, the retry policy for on-demand instances is similar to the policy for spot instances.