http://starrocks.io logo
Join Slack
Powered by
<@U0A71G31CDV> we are open-source starrocks users ...
# questions-and-troubleshooting
j
@Rocky we are open-source starrocks users interested in a number of features showcased in the phoenixAI 26.2 release, when / in what version can we expect these in starrocks?
r
PhoenixAI (formerly CelerData) is the company founded by the creators of StarRocks, and their PhoenixAI 26.2 release (October 2026) corresponds to features being rolled out in the StarRocks 4.x release cycle. Key Feature: Row-Level Incremental Materialized Views The flagship feature in PhoenixAI 26.2 is the transition of Incremental Materialized Views (IMV) from partition-level refresh to row-level refresh. * StarRocks 3.3/3.4: Support partition-level incremental refresh, where an update to a single row requires re-calculating the entire affected partition. * StarRocks 4.x (Current): Introduced row-level IMV (sometimes referred to as IVM or Incremental View Maintenance). This allows the engine to refresh only the specific rows that changed, significantly reducing compute costs and refresh latency (often down to 1–2 seconds even on billion-row tables). When to expect them in StarRocks OSS? Since PhoenixAI Enterprise and Cloud are built on top of the open-source StarRocks core, these features generally arrive in the community edition according to the following pattern: 1. Core Engine Features (IMV/IVM): These are part of the StarRocks 4.0 and 4.1 releases. The 
refresh_mode = "INCREMENTAL"
property is already available in the 4.x documentation and allows for row-level maintenance on Primary Key-based materialized views. 2. AI-Specific Features (MCP Server, Vector Enhancements): Features like the Model Context Protocol (MCP) server and enhanced Vector Indexes (IVFPQ/HNSW) were introduced in StarRocks 3.4 and 4.0 to support the AI agent workloads showcased in PhoenixAI 26.2. 3. Managed/Enterprise Features: Features such as the PhoenixAI Cloud API, automated multi-warehouse isolation, and row/column access policies remain specific to the PhoenixAI (CelerData) commercial offerings. Recommendation for OSS Users If you are currently on StarRocks 3.x, you should look toward StarRocks 4.1 to get the stabilization of the row-level incremental refresh and the latest "Fast for Agents" optimizations. For real-time updates on specific PRs or to see the progress of the 
IVMBasedMVRefreshProcessor
, you can track the StarRocks Roadmap 2026 (GitHub Issue #67632) or join the StarRocks community Slack. If you require the fully managed automation or the enterprise-specific security features showcased in 26.2, those are available via CelerData Cloud. References • sql-reference/sql-statements/materialized_view/CREATE_MATERIALIZED_VIEW.md • using_starrocks/async_mv/use_cases/create_partitioned_materialized_view.md • using_starrocks/async_mv/use_cases/data_modeling_with_materialized_views.md • release_notes/release-3.3.md
j
@Rocky these features specifically • Iceberg 
UPDATE
 • Iceberg v3 deletion vectors • Incremental materialized views on cloud-native tables • JDBC aggregate pushdown • better Cardinality estimation
r
Many of the features showcased in PhoenixAI 26.2 (the enterprise offering from CelerData) are part of the StarRocks 4.1 release cycle. Since PhoenixAI is built on the open-source StarRocks core, you can find these specific features in the following StarRocks versions: Feature Availability Guide Feature | StarRocks Version | Notes Iceberg 
UPDATE
 | v4.1+ | Supports 
UPDATE
(and 
DELETE
) for Iceberg V2 tables using the Merge-on-Read model. Iceberg v3 Deletion Vectors | v4.1+ | Part of the full Iceberg v3 spec support introduced in the 4.x line to improve read performance for mutable tables. Incremental MV on Cloud-Native Tables | v4.0 / v4.1 | Introduced row-level incremental refresh (
refresh_mode = "INCREMENTAL"
), which allows sub-second updates for shared-data (cloud-native) tables. JDBC Aggregate Pushdown | v4.1+ | Controls the ability to push 
GROUP BY
and aggregate functions (like 
SUM
, 
COUNT
) down to the source database. Enabled via the 
enable_jdbc_agg_push_down
variable. Better Cardinality Estimation | v3.5 & v4.0 | Recent versions added Multi-column Joint NDV (3.5), partition-level cardinality estimation, and per-bucket distinct count estimation for histograms (4.0). Summary for Open-Source Users * StarRocks 4.1 is the primary version to target for the majority of these "Lakehouse" and "Real-time" enhancements. * Incremental Materialized Views: While 3.x supported partition-level refresh, the row-level (IVM) capabilities required for the performance shown in PhoenixAI 26.2 are part of the 4.x engine architecture. * Iceberg DML: v4.1 marks a major milestone for Iceberg as a primary storage format, moving beyond "read-only" or "append-only" to fully mutable operations. To track the progress of these features, you can monitor the StarRocks 4.1 Release Notes or the StarRocks Roadmap 2026 on GitHub. For managed automation and enterprise-specific security isolation mentioned in the 26.2 announcement, please refer to CelerData Cloud. References • using_starrocks/accelerate-cardinality-preserving-joins.md • release_notes/release-3.5.md
The features you saw in the PhoenixAI 26.2 release (October 2026) are primarily delivered in the StarRocks 4.1 and 4.2 open-source release cycle. Since PhoenixAI is the commercial distribution of StarRocks, features typically land in the open-source version around the same time or in the immediately following minor release. Here is the status of those specific features in StarRocks OSS: 1. Iceberg 
UPDATE
 * Status: Introduced in StarRocks 4.2. * Details: Building on the native 
DELETE
support added in v4.1, StarRocks 4.2 adds native SQL 
UPDATE
support for Iceberg v2 tables using the Merge-On-Read (MoR) model. This allows for atomic updates without full-table rewrites. * Note: Currently limited to Parquet-formatted Iceberg v2 tables. 2. Iceberg v3 Deletion Vectors * Status: Reading support began in StarRocks 4.1. * Details: StarRocks 4.1 added foundational support for Iceberg v3 capabilities, specifically the ability to read tables utilizing Deletion Vectors (bitmaps that mark rows as deleted). This significantly improves read performance for tables with high churn compared to traditional position delete files. Full write-back support for deletion vectors is on the 4.2+ roadmap. 3. Incremental Materialized Views on Cloud-Native Tables * Status: Available in StarRocks 4.1. * Details: The 
refresh_mode = "INCREMENTAL"
(Incremental View Maintenance or IVM) is now supported for cloud-native (shared-data) internal tables and Iceberg tables. * Benefit: Unlike partition-level refresh (which recomputes an entire partition), row-level incremental refresh scales with the size of the change, allowing for sub-minute freshness even on massive datasets. 4. JDBC Aggregate Pushdown * Status: Expanded in StarRocks 4.1. * Details: While predicate pushdown (WHERE clauses) has long been supported, StarRocks 4.1 expanded pushdown capabilities for JDBC Catalogs to include Aggregate Pushdown (e.g., 
SUM
, 
COUNT
, 
MAX
, 
MIN
). * Benefit: This allows StarRocks to offload the initial aggregation to the source database (like MySQL or PostgreSQL), drastically reducing the amount of data transferred over the network. 5. Better Cardinality Estimation * Status: Core engine improvement in StarRocks 4.0/4.1. * Details: The Cost-Based Optimizer (CBO) received several major enhancements: * Joint NDV Statistics: Improved estimation for correlated columns (e.g., 
city
and 
zip_code
). * Function-Aware Estimation: Improved MCV (Most Common Values) and null-fraction propagation for 
CASE WHEN
, 
date_trunc
, and 
COALESCE
. * Iceberg Stats: Enhanced partition-level cardinality estimation for external Iceberg tables. Summary Table Feature | StarRocks OSS Version Iceberg 
UPDATE
 | v4.2 Iceberg v3 Deletion Vectors | v4.1 (Read) / v4.2+ (Write) Row-Level Incremental MV | v4.1 JDBC Aggregate Pushdown | v4.1 Advanced CBO Estimation | v4.0 & v4.1 For the most accurate timeline on the StarRocks 4.2 release, I recommend checking the StarRocks GitHub Releases or the #announcement channel in the community Slack. If you need the automated management of these features (such as the AI-driven warehouse scaling mentioned in PhoenixAI 26.2), those remain exclusive to CelerData Cloud. References • data_source/feature-support-data-lake-analytics.md • data_source/catalog/iceberg/DML.md • data_source/catalog/deltalake_catalog.md • sql-reference/sql-statements/table_bucket_part_index/DELETE.md
j
is there anything happening related to table stats being stale?
Open in Slack
PreviousNext
Powered by