DolphinDB V3.00.3 & 2.00.16 – Powering the Future of Real-Time Enterprise Intelligence

DolphinDB
2026-05-28

As real-time data becomes central to every business decision, enterprises are under pressure to reimagine how they store, process, and act on information. With version 3.00.3 & 2.00.16, DolphinDB delivers a major leap forward in real-time computing—bringing greater speed, flexibility, and intelligence to mission-critical workloads.

At the heart of this release is ORCA, our new enterprise-grade real-time computing platform. Designed for scale and simplicity, Orca enables teams to build powerful streaming pipelines with declarative APIs, multi-cluster orchestration, and intelligent task scheduling—making real-time development as intuitive as drawing a flowchart.

This version also introduces GPU-accelerated computational graphs, supercharging the performance of user-defined functions and unlocking faster, more complex analytics.

Streaming capabilities are further enhanced with support for event-driven workflows, threshold-based triggers, and session-aware time series processing. Additional highlights include FICC pricing suite, expanded time series modeling, reinforced enterprise security, and smarter infrastructure management.

Whether you're building alpha strategies, scaling IoT, or running global analytics, DolphinDB delivers performance, control, and real-time agility.

This article outlines the key features and enhancements in DolphinDB V3.00.3 & 2.00.16.

ORCA: Redefining Enterprise-Grade Real-Time Computing

As enterprises elevate data as their core productivity driver, they demand computing platforms that deliver real-time intelligence with unified architecture and bulletproof availability. DolphinDB ORCA redefines both system architecture and development models, equipping industries like finance, energy, and manufacturing with streaming capabilities built for mission-critical, real-time decision-making.

ORCA goes beyond traditional streaming—it's a new design philosophy that makes real-time computing as modular as building blocks , as intelligent as autonomous orchestration , and as intuitive as business logic. Anchored by three core principles—unification, instant response, and full automation —ORCA bridges every enterprise data source. From historical archives to live streams, high-velocity market feeds to periodic reports, ORCA delivers consistent, instant access across all business units, dramatically accelerating decision-making speed and precision at enterprise scale.

From Imperative to Declarative: A New Developer Experience

ORCA's DStream API transforms streaming development with clean, chainable syntax. The system automatically handles schema inference, engine deployment, and task scheduling. For example, a real-time minute-level factor computation task now can be expressed in just a few lines:

g = createStreamGraph("factor")
baseStream = g.source("snapshot",  1024:0, defs.name, defs.typeString)
  .parallelize(`symbol, 2)
  .timeSeriesEngine(60*1000, 60*1000, agg_metrics, `timestamp, false, `symbol)
  .reactiveStateEngine(factors, `symbol)
  .sync()
  .sink("basic_values")
g.submit()

No need to manually define stream tables, choose deployment nodes, or manage subscriptions—ORCA abstracts complexity into an intuitive, business-aligned model.

Decoupled Architecture with Intelligent Scheduling

ORCA introduces a Master-Worker architecture that cleanly separates task scheduling from execution:

  • Stream Master – A stateless global coordinator responsible for task scheduling and lifecycle management
  • Stream Worker – Executes streaming jobs with built-in fault tolerance and runtime state reporting

A built-in resource scoring system—based on CPU, memory, and disk usage—drives a greedy scheduling algorithm that dynamically routes tasks to the most suitable nodes. The result: fully automated, resource-aware scheduling with minimal manual overhead.

Enterprise-Ready: Resilient, Secure, and Built for Real-Time

Every streaming task is governed by a robust state machine. The system detects failures, restarts tasks, and ensures consistency — No downtime, no data loss, no disruptions.

ORCA also enforces security through compute group isolation and granular access controls , enabling safe, multi-user collaboration while protecting the integrity of underlying resources.

ORCA delivers more than speed—it's your central hub for real-time processing :

  • Unified data catalog with transparent cross-cluster access
  • Zero data duplication or migration requirements
  • Single integration point for data engineering, analytics modeling, and risk control

Real-time computing, reimagined—faster, simpler, more reliable, completely under your control.

ORCA beta is officially released in DolphinDB V3.00.3, with support for stream processing and a global data catalog. More features are on the way.

SharkGraph: Next-Level GPU Acceleration

Shark's latest release unveils SharkGraph —a new computing module that transforms DolphinDB user-defined functions into GPU-accelerated powerhouse. By transforming functions into optimized computational graphs , SharkGraph delivers substantial performance gains for complex analytical workloads.

A computational graph is essentially a directed acyclic graph (DAG) that represents the dependencies between operations in a computation. Unlike traditional step-by-step CPU-GPU task submission, which incurs frequent overhead, computational graphs batch operations into a single GPU task, maximizing parallelism and minimizing data transfer.

Shark intelligently analyzes your functions and automatically builds the optimal computation graph behind the scenes. The entire process runs natively on GPU, eliminating expensive CPU-GPU data transfers that traditionally slow down complex operations.

The usage is also designed to be intuitive. By simply adding a @gpu annotation to a function, users can offload it to GPU execution via the computational graph—without needing to write low-level code or understand GPU internals such as CUDA. For example:

@gpu
def myFunc(x){
  return min(mavg(x, 128), mavg(x, 64))
}
x=1..1000000
res = myFunc(x)

Streaming Engine Enhancements

Unlock greater flexibility, performance, and precision with powerful upgrades across our core streaming engines.

Threshold Engine: Trigger-Based Aggregation

The new threshold engine enhances streaming analytics by replacing fixed time windows with intelligent, data-driven triggers. It continuously tracks cumulative values across specified fields—such as trading volume or transaction counts—and generates aggregated results the moment your defined thresholds are crossed.

  • Smart threshold detection - Trigger aggregations based on actual data patterns, not arbitrary time intervals
  • Flexible grouping - Support for complex grouped aggregations across multiple dimensions
  • Cross-session continuity - Seamless accumulation that spans trading days and market sessions
  • Session-aware processing - Built-in trading session filters with automatic end-of-session completion

Perfect for high-frequency trading and event-driven analytics, where precision and responsiveness matter more than the clock.

CEP Engine: Smarter Complex Event Processing

The latest update simplifies the complex event processing workflow, making development and debugging more streamlined and efficient.

  • The new CEPMonitor class now unifies CEP functions, enhancing both code clarity and performance.
  • Added support for the name parameter and terminate() method enables more dynamic Listener management
  • Real-time interfaces now allow on-the-fly querying and debugging of monitors and listeners

Lookup Join Engine: Inner Join Support

The lookup join engine now features full equi-join (inner join) capabilities, easily activated through the isInnerJoin parameter for precise data matching scenarios.

  • On each incoming left-table record, the engine scans the right table for exact matches on the join key
  • Metrics are computed and emitted only when a match is found—unmatched data is ignored entirely

This feature enables accurate, noise-free aggregation in join-heavy streaming applications.

Daily Time Series Engine: Flexible Session Merging

Optimized for markets with complex trading hours, the enhanced daily time series engine introduces flexible control over session-based K-line merging.

Perfect for futures, forex, and other 24-hour markets where session boundaries matter as much as daily boundaries.

FICC Business: Advanced Bond Pricing Suite

The enhanced FICC bond pricing suite delivers comprehensive support for a broad spectrum of complex fixed income instruments, offering institutional-grade precision for both valuation and risk analysis.

  • Floating Rate Bonds (floatingRateBondDirtyPrice) – Calculates prices for floating-rate bonds based on a benchmark interest rate plus a fixed spread.
  • Callable/Putable Bonds (callableFixedRateBondDirtyPrice) – Valuates callable/putable bonds using the Hull-White model, incorporating optionality, volatility, and mean reversion effects.
  • Convertible Bonds (convertibleFixedRateBondDirtyPrice) – Prices convertible bonds via a binomial tree model, accounting for equity conversion options and call/put features.
  • Amortizing Bonds (amortizingFixedRateBondDirtyPrice) – Calculates prices for amortizing bonds with scheduled principal repayments. Utilizes discounted cash flow techniques, ideal for mortgage-backed and structured bond instruments.

Optimized Data Modeling & Analytics

Extended Time Series Modeling

To strengthen time series forecasting capabilities, two statistical models have been introduced:

  • ARMA Model – Combines autoregressive and moving average components for compact, efficient modeling of stationary time series with trends or seasonality.
  • ARIMA Model – Builds on ARMA by adding differencing to support non-stationary series, ideal for financial and macroeconomic time series forecasting.

Precision Interpolation & Fitting

New algorithms deliver superior modeling accuracy across diverse scenarios:

  • BVLS (BVLS) – Performs least-squares regression with boundary constraints, suited for constrained optimization scenarios.
  • Kernel Ridge (kernelRidge) – Captures complex, nonlinear relationships using kernel functions and ridge regularization. Effective in high-dimensional, low-sample contexts.
  • Cubic Hermite Spline (cubicHermiteSplineFit) – Generates smooth interpolated curves using derivative-informed cubic Hermite splines.
  • PCHIP Interpolation (pchipInterpolateFit) – Ensures monotonicity in interpolation using piecewise cubic Hermite methods, robust to noisy or irregular data.

New Temporal Func & Optimized SQL Execution

  • Smart Temporal Operations (temporalDeltas/datetimeDeltasM) – Precise interval calculations using natural days, business days, or custom trading calendars
  • Automatic Join Optimization – Dynamic cost estimation and row count analysis for optimal multi-table join ordering while preserving outer join constraints
  • Memory-Aware Join Retry Mechanism – Configurable retry limits (maxJoinTaskRetry) prevent system hang-ups during memory-intensive operations
  • NULL-Safe Join Control – Admin-level enableNullSafeJoin configuration for consistent NULL value handling across join operations

Comprehensive Security Upgrade

This release delivers comprehensive security hardening with advanced protection mechanisms that strengthen system integrity and operational resilience across enterprise touchpoints.

Authentication & Access Control

  • Secure Single Sign-On – Login tickets now include expiration and timestamps to prevent reuse. Only admin users can invoke generateUserTicket.
  • IP Access Control – Supports either a blacklist or whitelist mode, managing up to 65,535 IPs at the cluster level via persistent configuration APIs.
  • Session Management – Admins can configure session expiry with setSessionExpiredTime; users can check settings via getSessionExpiredTime.
  • Login Flow Hardening – Re-authentication required on re-login; automatic variable cleanup on logout or account switch prevents data leakage.
  • SCRAM Protocol – Adds support for SCRAM authentication, improving defense against replay, MITM, and password-based attacks.

End-to-End Data Protection

  • Transport Security (SSL/TLS) – Full encryption across RPC, distributed queries, cross-cluster access, and replication.
  • Transparent Data Encryption (TDE) – Two-tier key encryption for storage and backup, ensuring data confidentiality at rest.

Safer Queries & Audit Logging

  • Data Masking – Allows column-level masking for sensitive data; non-privileged users only see obfuscated values.
  • Parameterized Queries – Prevents SQL injection by decoupling query logic from input data using placeholders.
  • Tamper-Resistant Audit Logs – Now supports structured binary audit logs for improved traceability and integrity.

Infrastructure & Ops Optimization

The latest release delivers major improvements in operations, performance, and system reliability, enabling more efficient and secure data management.

Enhanced Monitoring

  • Audit Logging – Tracks key operations like login/logout, permission changes, and config updates, supporting compliance and troubleshooting
  • User Traffic Monitoring – Identifies high-resource users and logs data transfer volumes for better network resource management
  • IMOLTP Permissions – Adds fine-grained DML/DDL control with roles like DB_OWNER, DB_MANAGE, DB_READ, etc., enhancing production security

Performance & Stability

  • Query Optimization – Boosted performance for delete operations with enhanced partition pruning and TSDB handling.
  • Memory Protection – Better cache flushing and transaction size limits to prevent memory overflow
  • Resource Management – Configurable query timeouts and per-user limits on shared variables and partition counts ensure system stability

Storage-Compute Separation Upgrades

  • Cache Visibility – getComputeNodeCacheDetails provides insights into memory and disk usage by database/table/partition
  • Cache Management – New functions (warmupComputeNodeCache, getComputeNodeCacheWarmupJobStatus) support cache prewarming; cache now supports fine-grained clearing, LZ4 compression, and checksums
  • Performance Analysis – Extended HINT_EXPLAIN reveals whether queries hit compute node cache or fall back to data nodes

Targeted Feature Additions

  • IoT Feature Extension – New function getLocalIOTDBStaticTable allows retrieval of static table of an IOTDB database stored on the local node, such as measurement point statistics.
  • Plugin Flexibility – Dynamic logging control (setLogLevel/getLogLevel) and keyword parameter support for flexible plugin architecture