2026.2 Series Release Notes¶

26.0.0-8¶

Fixed database status detection for stopped containers that retain a healthy Docker health status. Trove now identifies an exited container as SHUTDOWN before evaluating its health status.

MySQL replication instances now use the ROW binary logging format instead of MIXED. Previously, enabling replication changed binlog_format to MIXED on the replication source and replicas, which would override the default ROW value on existing instances. Existing replication setups are not modified. Rebuilding existing MySQL instance will change MIXED value to ROW.

Reduced PostgreSQL connection latency in some environments by removing the localhost pg_hba rule, avoiding reverse DNS lookups.

Removes character_set_client, character_set_connection, character_set_database, character_set_results, collation_connection, and collation_database from the shipped MariaDB configuration parameter validation rules. These variables are not valid MariaDB server option-file settings and can prevent an instance from starting when used in a configuration group.

Rules already loaded for existing datastore versions are not changed automatically. Cloud administrators should remove these parameters from each affected MariaDB datastore version with the management API, or delete and reload that version's complete validation rule set.

Database shutdown requests now use agent_call_high_timeout instead of agent_call_low_timeout. This provides more time for a database under load to stop before the guest agent call times out.

Introduced a new configuration option [DEFAULT]/cinder_client_version with default value set as 3. This allows operators to supply the intended version of Cinder API to interact with.

Adds the default_locality configuration option. Operators can use it to apply a Nova server group policy when an instance create request omits locality. The default is None for backward compatibility.

Simplifies the DevStack installation process for Trove.

The setup procedure is updated to produce a working environment out of the box, and the documentation now provides clearer step-by-step installation instructions.

Introduce additional fields "wal_size" and "wal_archive_size" for PostgreSQL datastore.

From now on it is possible to have the same backup image for series of datastore versions. Trove guest-agent will iterate over tagged versions for backup images from more to less specific ones attempting to find an image rather then fail if exact image is not present in the registry. For instance, in case you defined [mariadb]/backup_docker_image: registry/db-backup-mariadb and have only 10.4 in the registry, such image will cover whole 10.4 series, including 10.4.1, 10.4.2, 10.4.3, etc.

Improves instance list method by using Nova tags feature. Tags was introduced in Nova API version 2.26 in 2017. Querying Nova servers by tags allow to check relevant servers only instead of retrieving all of them, preventing possible 504 errors in large clouds.

Service credentials removal from trove-guestagent.conf. Trove guestagent no longer requires Keystone service credentials. Operators should remove this section from trove-guestagent.conf. Storing credentials on guest instances represents a security risk.

SSL/TLS certificate management for securing the client-server channel. The PKCS#12 container and credentials are stored in Barbican. Available modes are basic, enforced, and mTLS. See the documentation for more information.

Even after upgrading to this version, the existing guest-agent images will function normally alongside the existing backup images. However, to ensure compatibility with newly built guest images as well as the legacy ones, you must rebuild the backup images.

Instance rebuilds now apply updated cloud-init configuration when Nova API microversion 2.57 or later is used. Older Nova microversions continue to use the previous rebuild behavior.

Trove now uses blkid instead of dumpe2fs to detect existing ext2/3/4 filesystems and automatically attempts to repair recoverable filesystem inconsistencies before mounting a data volume.

Logic for fetching Cinder API version has been fixed, and it no longer relies on the [DEFAULT]/cinder_service_type value. This allows to have a cinder_service_type defined as block-storage, or any other name which is not versioned.

Fixed a typo in the default storage driver name (swfit instead of swift) in the task manager backup creation path. When a backup request arrived without storage_driver in its metadata, the Swift endpoint URL was not injected and the guest agent rejected the backup with "Missing swift_url in backup metadata", leaving the backup record stuck in the NEW state.

Guest log files and directories are now owned by database_service_uid:database_service_gid instead of using the uid for both the owner and the group. This makes a difference for custom Docker images where the database service gid differs from the uid.

Fixed guest log operations (e.g. listing logs) failing with "Please specify owner or group, or both." for MySQL/MariaDB and PostgreSQL instances when database_service_uid is not set in the datastore-specific configuration group. The log file owner now falls back to the database_service_uid value from the DEFAULT group, consistent with how the ownership of other database files is resolved.

Fixes MariaDB when datastore use official MariaDB dockerhub images. MariaDB starting from version 10.7 use different default path to configuration file.

Removed the legacy PostgreSQL max_connections workaround that persisted the setting in postgresql.auto.conf, allowing standard Trove configuration management to modify it. Rebuild now removes existing legacy max_connections entries from postgresql.auto.conf, while preserving support for custom configuration on individual replicas.

Showing an instance that is being deleted (or whose Nova server is shut down) no longer sends a get_filesystem_stats RPC to the guest agent. Previously the API would block for the full agent_call_low_timeout waiting for a reply from an agent that is no longer running, making the request slow and logging a spurious guest timeout traceback.

Trove now unmounts the data volume before rebuilding an instance to reduce the risk of data corruption.

Use pre-auth Swift client and remove Keystone re-auth. Stop using Keystone sessions and token re-scoping in backup and restore flows.

Backup container's swift client now respects swift_api_insecure config option in trove-guestagent.conf, instead of relying on hardcoded insecure=True.