Managing metadata performance
You can update the metadata configuration of your FSx for Lustre file system without any disruption to your end users or applications by using the Amazon FSx console, Amazon FSx API, or AWS Command Line Interface (AWS CLI). The update procedure increases the number of provisioned Metadata IOPS for your file system.
Enhanced metadata is available only for 2.15 file systems. You can increase metadata performance only on FSx for Lustre file systems created with the Persistent 2 deployment type and a metadata configuration specified. You cannot add or update the metadata configuration for an FSx for Lustre file system if the metadata configuration is not specified at the time of file system creation. This also applies to file systems restored from backups of 2.12 file systems which did not support enhanced metadata performance, or from 2.15 file systems which had no metadata configuration specified.
The increased metadata performance of your file system is available for use within minutes. You can update the metadata performance at anytime, as long as metadata performance increase requests are at least 6 hours apart. While scaling metadata performance, the file system may be unavailable for a few minutes. File operations issued by clients while the file system is unavailable will transparently retry and eventually succeed after metadata performance scaling is complete. You will be billed for the new metadata performance increase after it becomes available to you.
You can track the progress of a metadata performance increase at any time by using the Amazon FSx console, CLI, and API. For more information, see Monitoring metadata configuration updates.
Topics
Lustre metadata performance configuration
The number of provisioned Metadata IOPS determines the maximum rate of metadata operations that can be supported by the file system.
When you create the file system, you choose a metadata configuration mode:
-
Choose Automatic mode if you want Amazon FSx to provision and scale Metadata IOPS for you. For SSD file systems, the number is based on storage capacity. For Intelligent-Tiering file systems, it is based on throughput capacity.
-
Choose User-provisioned mode if you want to specify the number of Metadata IOPS to provision for your file system.
For both SSD and Intelligent-Tiering file systems, valid Metadata IOPS values are 1500, 3000, 6000, 12000, and multiples of 12000 up to a maximum of 192000.
Intelligent-Tiering file systems created before November 5, 2026, support only User-provisioned mode, with valid Metadata IOPS values of 6000 and 12000. A file system restored from a backup inherits the supported metadata modes, Metadata IOPS values, and minimum Metadata IOPS requirement of the file system the backup was created from.
You can't provision fewer Metadata IOPS than Automatic mode provisions for your file system, and you can switch to Automatic mode only if your current Metadata IOPS is less than or equal to that number. For the values that Automatic mode provisions, see Metadata IOPS in Automatic mode for SSD file systems and Metadata IOPS in Automatic mode for Intelligent-Tiering file systems.
If the metadata performance of your workload exceeds the number of Metadata IOPS provisioned in Automatic mode, you can use User-provisioned mode to increase the Metadata IOPS value for your file system.
You can view the current value of the file system's metadata server configuration as follows:
-
Using the console – On the Summary panel of the file system details page, the Metadata IOPS field shows the current value of the provisioned Metadata IOPS and the current metadata configuration mode of the file system.
-
Using the CLI or API – Use the describe-file-systems CLI command or the DescribeFileSystems API operation, and look for the
MetadataConfigurationproperty.
Considerations when increasing metadata performance
Here are a few important considerations when increasing your metadata performance:
Metadata performance increase only – You can only increase the number of Metadata IOPS for a file system; you cannot decrease the number of Metadata IOPS.
Specifying Metadata IOPS in Automatic mode not supported – You can't specify the number of Metadata IOPS on a file system that is in Automatic mode. You'll have to switch to User-provisioned mode and then make the request. For more information, see Changing the metadata configuration mode.
Metadata IOPS for data written before scaling – When scaling Metadata IOPS beyond 12000, FSx for Lustre adds new metadata servers to your file system. New metadata is automatically distributed across all servers for improved performance. However, existing metadata and subdirectories created before scaling remain on original servers, with no increase in Metadata IOPS.
Time between increases – You can't make further metadata performance increases on a file system until 6 hours after the last increase was requested.
Throughput capacity increases on Intelligent-Tiering file systems – Increasing throughput capacity can increase the minimum Metadata IOPS for your file system. For more information, see Considerations when updating throughput capacity.
Concurrent metadata performance and SSD storage increases – You cannot scale metadata performance and file system storage capacity concurrently.
When to increase metadata performance
Increase the number of Metadata IOPS when you need to run workloads that require higher levels of metadata
performance than is provisioned by default on your file system. You can monitor your metadata performance
on the AWS Management Console by using the Metadata IOPS Utilization graph which provides the
percentage of provisioned metadata server performance you are consuming on your file system.
You can also monitor your metadata performance using more granular CloudWatch metrics. CloudWatch metrics include
DiskReadOperations and DiskWriteOperations, which provide the volume of metadata
server operations that require disk IO, as well as granular metrics for metadata operations
including file and directory creation, stats, reads, and deletes. For more
information, see FSx for Lustre metadata metrics.