Dremio Support Advisory, November 18, 2026
***Dremio Enterprise 27.0 Breaking Changes***
Dear Dremio Enterprise Customer,
Please be advised that Dremio Enterprise 27.0, targeted to ship mid December 2026, contains
breaking changes that may require action before you upgrade. Review the list to identify
changes that impact you as not all changes will impact every customer.
Important notice: Dremio Enterprise 27.0 will not be available for the Dremio AWS Edition
(AWSE) on the AWS Marketplace, which is deprecated. If you have not done so, please consult
your Dremio Account Executive about migrating to another supported Dremio platform.
Breaking changes for Dremio Enterprise 27.0:
| Product Area | Change | What You Will See in 27.0 | Required Steps |
| Platform | Dremio Enterprise is supported only on Kubernetes | VM-based deployments are no longer supported. Dremio Enterprise is supported only on Kubernetes, deployed with Dremio’s Helm chart. | Move VM-based deployments to Kubernetes using Dremio’s Helm chart before upgrading. If your deployment uses a JDBC driver that you added to Dremio nodes manually, such as the Teradata driver, you must include it in a custom image based on Dremio’s image. |
| Platform | Classic engines are no longer supported | Classic engines no longer work in Dremio Enterprise 27.0. This change was announced in the documentation with Dremio 26.0. | Migrate to Managed Engines before upgrading. |
| Platform | Some support keys have been removed | These support keys are no longer supported: exec.enable_union_type, planner.use_legacy_decorrelator, exec.operator.vectorized_spill.ndv, exec.operator.vectorized_spill.varchar, exec.operator.vectorized_nospill.varchar_ndv, exec.operator.sort.external.direct_write, exec.errors.verbose, reflection.manager.avoid_redundant_refresh, search.v2.enabled, search.job_ingest.enabled, search.scheduled_reconciliation.enabled, store.hive.orc.use_direct_memory, store.hive.orc.vectorize, store.hive.orc.vectorize.enable_filter_pushdown. Setting a removed key with ALTER SYSTEM or ALTER SESSION returns "The option does not exist." Values already stored are ignored. | Remove these keys from use. |
| Platform | Dremio OSS source code and Maven artifacts are no longer updated or released | Dremio Enterprise 27.0 is available only as Enterprise Edition. Dremio no longer updates or releases open source (OSS) source code or Maven artifacts. Custom Java user-defined functions (UDFs) and custom source plugins built against the OSS codebase are no longer supported (SQL UDFs are unaffected). | Before upgrading, replace custom Java UDFs and custom source plugins with supported alternatives (e.g., SQL UDFs or built-in sources). |
| Catalog | OPTIMIZE on Open Catalog tables requires SELECT and WRITE | OPTIMIZE now requires both SELECT and WRITE instead of USAGE. | Grant SELECT and WRITE to users and roles that currently rely on USAGE before upgrading. |
| Catalog | Deleting an Open Catalog folder requires DROP | ALTER no longer authorizes Open Catalog folder deletion. DROP is required. | Grant DROP to affected users and roles before upgrading. |
| Catalog | Privilege audit events change type | Privilege audit events use the type PRIVILEGE instead of PERMISSION. | Update rules, audit queries, dashboards, and alerts that match on PERMISSION. |
| Catalog | Deleting a non-empty Open Catalog namespace returns HTTP 409 | Deleting a non-empty Open Catalog namespace now returns HTTP 409 Conflict instead of HTTP 500. | Update API clients and automation that expect HTTP 500. |
| Datalake / Iceberg | Apache Hive 2 support removed | Hive 2 reached the end of life in May 2024. During the upgrade to Dremio Enterprise 27.0, any remaining Hive 2 sources are automatically deleted from the catalog. Views and Reflections that depend on them will fail. Hive 3 sources are unaffected. | Migrate Hive 2 sources to Hive 3 before upgrading. |
| Datalake / Iceberg | NAS source mount paths are restricted | NAS sources can only be mounted under the /mnt and /volumes folders (or any folder beneath them), which prevents attaching a NAS source to a node's root path. The allowed folders are configurable per deployment. | If any NAS source uses a path outside /mnt or /volumes, move it under an allowed folder or add the folder to the allowed list for your deployment. |
| Datalake / Iceberg | Microsecond and nanosecond timestamp precision | Timestamps now support microsecond and nanosecond precision (Apache Iceberg v3). Dremio previously truncated microseconds to milliseconds, so comparisons and aggregations on microsecond data become more exact. Operations that rely on millisecond-precision fields should not be affected. | No action for most workloads. Review comparisons or aggregations that depended on microsecond values being truncated. |
| Datalake / Iceberg | wasb:// and wasbs:// locations use the Azure Blob File System (ABFS) driver | Hadoop is upgraded to 3.5.0, which removes the native Windows Azure Storage Blob (WASB) driver. wasb:// and wasbs:// locations are now accessed through the ABFS driver. | Review and test Azure Blob Storage and Hive sources that use wasb:// paths, especially storage accounts without hierarchical namespace enabled, before upgrading. |
| Monitoring / Observability | dremio-telemetry.yaml is no longer supported | Dremio no longer reads the dremio-telemetry.yaml file. This file has not been included in Dremio’s Helm chart or documented since Dremio 26.0. If you added it manually, its settings are ignored after the upgrade. Micrometer metrics remain available through the Prometheus /metrics endpoint with no configuration required. | No action unless you added dremio-telemetry.yaml manually. If you did, remove it and collect metrics from the Prometheus /metrics endpoint. |
| Monitoring / Observability | Dropwizard (Codahale) metrics have been replaced by Micrometer | Dropwizard (Codahale) metrics are removed. Each has an equivalent Micrometer metric. Some metric names and labels differ from Dremio 26.0+, including JVM GC, memory, buffer-pool, and thread metrics. The JMX metrics endpoint has also been removed; metrics are available only through the Prometheus /metrics endpoint. | Update Prometheus and Grafana dashboards and alerts that reference Dropwizard metric names to use the Micrometer equivalents before upgrading. JMX metrics are no longer available, so switch any JMX-based monitoring to the Prometheus /metrics endpoint. |
| Monitoring / Observability | OpenTracing and Jaeger client support removed | Distributed tracing is provided only by the OpenTelemetry Java agent, configured with standard OTEL_* environment variables. Custom extensions that use Dremio's TracerFacade or OpenTracing APIs are no longer supported. | Migrate custom tracing extensions to OpenTelemetry before upgrading. |
| Monitoring / Observability | Timestamp metrics removed or renamed | Timestamp precision metrics are no longer emitted, and one metric has a new name: query_operator_count_total. | Update dashboards and alerts that use these metrics, including any that reference the renamed metric, to use query_operator_count_total. |
| Monitoring / Observability | Logback library updated | The Logback logging library is upgraded to version 1.6.x. Some existing logging configuration files may not be compatible with this version. | Review custom Logback configuration files and update any that are not compatible with Logback 1.6.x before upgrading. |
| Reflections | Reflection names must be unique within a project | When coordinators first start on Dremio Enterprise 27.0, a one-time migration renames duplicate Reflection names to <originalName>_<datasetPath>_<type>_<index>. The migration cannot be disabled. Reflections are not deleted or disabled, and anything that identifies a Reflection by its ID, including SQL, scripts, and APIs, continues to work. Creating a Reflection with an existing name now fails. Renamed Reflections keep their new names if you roll back. | Update scripts and automation that look up Reflections by an old duplicate name, or that create Reflections with duplicate names. |
| Security | Ranger script-based policy conditions disabled by default | Ranger script-based custom policy conditions on Hive sources are disabled by default (CVE-2025-59059), and the Ranger Hive plugin is upgraded to 2.8.0. Policies that depend on script conditions are no longer evaluated. | Review Ranger policies that use script conditions. If needed, re-enable them in the source's advanced properties. |
| Sources | Snowflake login/password authentication removed | You can no longer authenticate to a Snowflake source with a login and password, following Snowflake's deprecation of this method. | Switch Snowflake sources to a supported authentication method before upgrading. |
| SQL / Planner | IS [NOT] DISTINCT FROM no longer converts VARCHAR to date/time | Comparing a VARCHAR column to a DATE, TIME, or TIMESTAMP column with IS [NOT] DISTINCT FROM no longer converts types automatically. The previous behavior could silently return wrong results when VARCHAR values were not valid dates. Queries and saved views that do this fail validation. Comparisons against string literals are not affected. | Add an explicit CAST to these comparisons before upgrading. |
| SQL / Planner | FLATTEN alias columns blocked in JOIN conditions | Columns produced by a FLATTEN alias can no longer be used in JOIN conditions, because such queries could return incorrect results. Queries and views that do this fail. | Rewrite affected queries to flatten in a subquery or CTE and join on the resulting column, or move the condition to a WHERE clause. |
| SQL / Planner | Redundant ORDER BY in subqueries is dropped | The planner removes ORDER BY clauses in subqueries and views where they do not affect the final result, in line with the SQL standard. Queries that relied on an inner ORDER BY without an outer ORDER BY may return rows in a different order. | Add an explicit ORDER BY to the outermost query. |
| SQL / Planner | INSERT matches columns by position | INSERT statements no longer fall back to name-based column matching. Values are matched to columns by position so that data can be written to different columns than before. | Make the order of values match the table's column order, or list the target columns explicitly. |
| SQL / Planner | Default planning strategy for correlated subqueries changed | The default for planner.decorrelation.type changes from CALCITE_WITH_FORCE_VALUE_GENERATION to CALCITE_WITH_CONDITIONAL_VALUE_GENERATION. Plans for correlated subqueries can differ from Dremio 26.0+, and in edge cases, results can differ. | If you see a regression, set planner.decorrelation.type back to CALCITE_WITH_FORCE_VALUE_GENERATION. |
| AI | Legacy AI agent removed | The legacy AI agent has been removed and replaced by an updated AI agent interface. | No explicit action required. If underlying agent endpoints were in use, they would need to be updated. |
If you have further questions, please contact Dremio Support via the Support Portal.
Thank you for your continued support of Dremio.
Sincerely,
Dremio Support
Dremio is the Agentic Lakehouse: the only lakehouse built for agents and managed by agents. It is an open, high-performance data lakehouse platform built to accelerate AI and analytical workloads across all enterprise data.