a9s Search Resources Considerations

This document describes concerns and limitations to be considered when allocating resources for your a9s Search plans.

Backup Storage

a9s Search uses the plugins repository-s3 and repository-azure for backup and restore. Those are the only backend storages supported for a9s Search and currently they do not support SSL with unknown certificate issuers.

Known Limitations of a9s Search's Backups

• The backup process is based on OpenSearch's Snapshot functionality, this means that:
  • Backups are not encrypted.
    • Due to this, the backup_manager_encryption_key property cannot be set.
  • Backups cannot be downloaded.
  • The storage backend is limited to AWS S3 and Azure; generic S3 API services are not supported.
• OpenSearch's default filter plugins are not available.

Reserved Memory

OpenSearch runs in a JVM, a9s Search configures the maximum and minimum heap size (-Xms and -Xmx) with 46% of the total memory. Keep in mind that memory is used for other essential operations, so it is normal for an OpenSearch process to report a higher memory usage than this. The remaining available memory is shared between other internal OpenSearch processes, side processes, and OS.

a9s Search does not limit the size of memory that can be locked, as soon as OpenSearch starts, the heap size is locked from the total memory, making it unavailable for other processes.

File Descriptors

OpenSearch uses a lot of files. It can happen that in a given moment, a vast number of files are opened by the OpenSearch processes. a9s Search configures the service instances with a maximum number open files of 65536 (ulimit -n 65536). Bear this in mind in case your usage of the cluster favors opening a vast number of files.

If the OpenSearch process runs out of file descriptors, data loss might happen.

Number of Threads

OpenSearch uses a thread pool for multiple operations, a9s Search configures the maximum number of processes for OpenSearch as 4096 (ulimit -u 4096).