Monday, September 28, 2026

Chris van Eijk benchmarks five index configurations across 1,000 tenants in PostgreSQL 17 to determine optimal column ordering for multi-tenant workloads. The analysis reveals that placing tenant_id as the leading column is critical for query performance, particularly when Row Level Security (RLS) alters execution plans. The study also evaluates when table partitioning provides sufficient payoff to justify its complexity.

• Leading with tenant_id ensures efficient index scans for per-tenant queries in multi-tenant schemas.
• RLS can significantly change query plans; verify execution paths after enabling security policies.
• Partitioning only pays off at scale; measure overhead before adopting for small tenant counts.
• PostgreSQL 17 optimizations were tested, but column order principles remain consistent across versions.

Chris van Eijk benchmarks five strategies for generating gapless, unique invoice numbers per tenant in PostgreSQL 17, including sequences, max()+1, SERIALIZABLE transactions, and counter rows. The analysis specifically measures performance, gap occurrence, and duplicate resistance under concurrent load. This provides empirical data to help engineers choose the right concurrency control model for high-throughput billing systems.

• Sequences offer speed but create gaps under high concurrency; assess if gaps are acceptable.
• SERIALIZABLE transactions guarantee no duplicates or gaps but significantly reduce throughput.
• Max()+1 is simple but prone to race conditions without strict locking or serialization.
• Counter row approach balances consistency and performance for multi-tenant isolation.
• PostgreSQL 17 handles these contention patterns efficiently, but locking remains the bottleneck.

Core contributor Robert Haas is hosting a hacking workshop in October 2026 focused on his presentation regarding pg_plan_advice. The session targets plan stability and user-level planner control for PostgreSQL. Participants must sign up via a provided form to receive a session invite.

• Workshop focuses on pg_plan_advice for managing plan stability.
• Target audience includes engineers interested in user planner control.
• Attendance requires registration through the official sign-up form.
• Event is scheduled for October 2026 with specific session invites.
• Led by Robert Haas, a key figure in PostgreSQL core development.

AWS has made PostgreSQL 19 Beta 4 available in the RDS Database Preview Environment, allowing users to evaluate the pre-release version. This build refines query performance and autovacuum management capabilities introduced in earlier betas, alongside stability improvements. Key updates include bug fixes for the pg_stat_autovacuum_scores view and corrected freeze-score scaling for better tuning.

• PostgreSQL 19 Beta 4 is now available for evaluation in the RDS Database Preview Environment
• Autovacuum management improvements include rebalanced cost limits for parallel workers
• Fixes address pg_stat_autovacuum_scores reporting for TOAST tables and freeze-score scaling
• Users can now tune autovacuum prioritization more effectively with corrected scaling metrics

PgBouncer 1.26.0 addresses three critical denial-of-service vulnerabilities triggered by unauthenticated clients or malicious servers, specifically involving SCRAM nonce handling, integer overflows, and unbounded iteration counts. The release introduces functional improvements including tracking of search_path and default_transaction_read_only, new pool_idle_timeout configuration, and per-user/query_wait_timeout settings. Additionally, it adds Meson build support and removes the deprecated online restart (-R) functionality.

• Patch immediately to fix DoS risks from unauthenticated SCRAM and integer overflow attacks.
• Migrate from Make to Meson build system for future compatibility.
• Utilize new pool_idle_timeout and granular query_wait_timeout settings.
• Remove deprecated -R flag usage and verify search_path tracking behavior.
• Monitor for malicious server-side SCRAM iteration count attacks.

An audit of a production text-to-SQL pipeline reveals that the deployed gpt-4o-mini judge performs poorly, agreeing with human annotators at only 0.04 Cohen's kappa on a disagreement-enriched dataset. The primary failure mode is identified as GRADE-HALLUCINATION, causing the model to over-flag 77.1% of human-verified faithful cases. Replacing the judge with a self-hosted Qwen3.6-27B model improves agreement to 0.72 kappa while reducing costs by a factor of 300 compared to premium alternatives.

• Deployed LLM judges require empirical validation; assumed accuracy can be dangerously low.
• gpt-4o-mini showed near-random agreement (kappa 0.04) on enriched disagreement sets in this context.
• GRADE-HALLUCINATION is a specific failure mode causing excessive false positives in faithfulness checks.
• Self-hosted Qwen3.6-27B matches Claude Opus 4.7 quality (kappa ~0.71) at 1/300th the cost.
• Ensembling strategies were noted but the excerpt cuts off before detailing their efficacy.

This paper critiques existing KV cache eviction methods like SnapKV for relying too heavily on mean attention scores. It introduces a unified scoring function that incorporates attention dispersion and redundancy penalties similar to maximal marginal relevance, all without requiring additional forward passes. The authors further demonstrate that optimal eviction parameters should vary by network depth, using a reparameterized search to find ideal profiles.

• Adds attention dispersion and redundancy to standard mean attention scores for better token ranking.
• Implements MMR-like penalties for token similarity without incurring extra forward pass costs.
• Shows that eviction coefficients should vary by layer depth rather than using fixed global values.
• Validated on 16 English LongBench datasets using Mistral-7B with constrained compute budgets.

Researchers introduce HARDEN, a method that uses constrained evolutionary search to make existing evaluation benchmarks more difficult without changing the expected answers. By mutating inputs along domain-specific complexity axes while enforcing feasibility constraints like semantic preservation, the technique creates harder test cases. Tests on Qwen3.5 models across FinQA, PubMedQA, and ContractNLI show significant accuracy drops compared to standard baselines, revealing deeper model weaknesses.

• HARDEN adapts existing eval inputs to be harder while keeping ground-truth answers fixed.
• Feasibility constraints ensure new cases remain realistic and semantically valid.
• Qwen3.5 models saw up to 49.9% accuracy drop vs single-pass baselines.
• Current curated benchmarks likely underrepresent real enterprise deployment complexity.

Amazon SageMaker HyperPod Inference Gateway is a Kubernetes-native add-on that replaces round-robin balancing with real-time inference-signal routing. It deploys on existing HyperPod infrastructure without application changes, using an Envoy endpoint and model-aware router. This approach reduces first-token latency by up to 82% and drastically lowers p99 TTFT in mixed-hardware scenarios.

• Deploys as an EKS managed add-on with zero client-side changes
• Routes traffic based on model name in the request body
• Reduces first-token latency by up to 82%
• Cuts p99 TTFT by 97-98% in burst traffic
• Serves multiple models from a single private endpoint

Amazon Bedrock Managed Knowledge Base now includes native connectors for Salesforce and Zendesk, eliminating the need for custom ingestion pipelines. Users can sync knowledge articles, community posts, and metadata by simply providing instance credentials. The service handles crawling and incremental synchronization automatically, enabling AI agents to leverage existing support data.

• Native connectors replace custom ETL pipelines for Salesforce and Zendesk data.
• Supports incremental sync of articles and community posts to keep RAG grounded.
• Metadata extraction is automated, simplifying agent context management.
• Reduces engineering overhead for building support-focused AI assistants.

Retrieval-augmented generation is standard for verifying LLM claims in clinical settings, yet aggregate metrics like F1 hide specific failure modes. This paper introduces two comprehensive taxonomies to diagnose errors, moving beyond opaque scoring. Grounded in case studies on MedExpert and other datasets, the framework decomposes failures into retrieval-stage errors across five quality dimensions.

• Aggregate F1 scores obscure the root causes of hallucinations in medical RAG systems.
• New taxonomies decompose retrieval failures into five distinct quality dimensions.
• Diagnostics work without gold answers or annotated evidence, enabling scalable auditing.
• Framework enables precise debugging of retrieval-stage errors in high-stakes clinical AI.

A new study of 1,200 agent trajectories reveals that coding agents frequently exhibit three cost-inefficient behaviors: subsumed retrieval, similar script generation, and test re-execution. These patterns impact up to 98% of tasks, driving unnecessary token consumption. The authors evaluate mitigation strategies like structure-aware retrieval and synthesized skills across 10,000 held-out trajectories to reduce this overhead.

• Subsumed retrieval, similar script generation, and test re-execution are the primary drivers of cost inefficiency.
• These wasteful behaviors affect between 79% and 98% of all coding agent tasks analyzed.
• Structure-aware retrieval and agent-synthesized skills offer viable paths to reduce token spend.
• Optimizing agent behavior is critical for scaling coding assistants without exponential cost growth.

Current LLM-as-a-judge evaluations often fail in production because they treat retrieved reference answers as literal targets, penalizing correct procedural actions that differ only in specific entity details like IDs or dates. This phenomenon, termed reference-instance divergence (RID), causes valid agent behaviors to be flagged as hallucinations. CARGO addresses this by treating references as procedural guides rather than exact templates, grounding judgments in the live instance's actual context.

• Solves reference-instance divergence where valid actions are penalized for differing identifiers.
• Shifts evaluation from literal matching to procedural correctness based on live context.
• Uses three-way claim status: supported, contradicted, or unverifiable.
• Only contradictions are penalized, reducing false positives in agentic AI monitoring.

This paper identifies a pattern termed 'LLM Parkinsonism,' where autonomous agents continue executing low-value actions even after their primary objective is satisfied. The authors argue this stems from a structural flaw where proposal generation, scope interpretation, and stopping authority are concentrated in a single self-conditioned loop. To address this, the work proposes an uncertainty-aware global executive control architecture designed to decouple these functions and enforce proper termination.

• Agents often fail to stop, producing redundant refinements or repairs to self-created complexity.
• The root cause is architectural: self-conditioned loops lack external executive oversight.
• Next-token prediction alone does not explain this persistence; control flow is key.
• A proposed global executive control architecture uses uncertainty awareness to manage scope.
• Decoupling proposal generation from stopping authority is critical for reliable autonomy.

Cartograph addresses the scalability bottleneck of the Model Context Protocol by replacing linear catalog traversal with progressive disclosure. It uses operator-attested capability cards signed via Ed25519 to ensure authenticity without relying on publisher rankings. The system employs a three-layer confusable-cluster analysis called Rift and a two-stage retrieval process to rank servers before individual tools, significantly reducing overhead.

• Shifts MCP discovery from O(n) to O(k) via progressive disclosure.
• Uses Ed25519 operator-attested cards for secure, decentralized tool descriptions.
• Rift mechanism clusters confusable tools using density and query-margin analysis.
• Two-stage retrieval ranks servers first, then tools, optimizing lookup paths.

A new architectural approach uses the Model Context Protocol to resolve conflicts between probabilistic LLM interactions and policy-driven data infrastructures. By implementing a mediation layer called Eunomia Agent, data space capabilities are translated into structured, schema-driven tools that AI agents can safely discover and invoke. This prototype validates end-to-end interaction while preserving strict governance constraints across organizational boundaries.

• MCP acts as a translator between LLMs and governed data spaces, ensuring policy compliance.
• Eunomia Agent exposes data capabilities as structured tools rather than raw text or APIs.
• Schema-driven discovery allows agents to invoke services without violating sovereignty rules.
• Prototype confirms end-to-end feasibility of mediated agent-data space interactions.

This paper introduces Learning What to Skip (LW2S), a method that optimizes multi-agent LLM workflows by selectively omitting redundant or harmful steps like verification or summarization. The approach treats component omission as a counterfactual credit assignment problem, using full-workflow logs and controlled skip interventions to learn action-specific safety models. By combining held-out calibration with domain-native guards, the controller decides whether to execute a step or skip it, preserving correct intermediate answers and saving computation.

• Reduces latency and token costs by skipping unnecessary workflow stages like verification.
• Uses counterfactual credit assignment to quantify the value of each component dynamically.
• Learns safety models from interventions, allowing the system to skip safely when appropriate.
• Prevents overwriting correct intermediate answers that might be lost by forced execution.
• Integrates with domain-native guards for robust decision-making in complex agent chains.

Spotify addresses the cold-start challenge in conversational recommendation by generating multi-turn synthetic data from single-turn prompts. This pipeline allows for systematic evaluation of agent planning before real user interactions are available. The system also employs a self-improvement loop to refine tool selection and sequencing over time.

• Synthetic data bridges the cold-start gap for conversational AI without waiting for real user logs.
• Multi-turn conversation generation enables robust planning evaluation prior to production launch.
• Self-improvement loops help optimize tool invocation strategies through variance-based contrastive learning.
• This approach accelerates deployment of complex intent-based recommendation systems.

Release 1.7.4 of langchain-anthropic introduces profile augmentations for new model variants including Opus 5.5 and GPT-6. It also enables tool changes on SystemMessage during active conversations, affecting both Anthropic and OpenAI integrations. Additionally, integration test cassettes have been corrected to ensure stable CI pipelines.

• New Opus 5.5 and GPT-6 model profiles are now available for selection.
• SystemMessage can now modify tools mid-conversation for dynamic agent behavior.
• Fixes integration test cassettes to prevent CI flakiness.
• Upgrade to access latest model support and conversation flexibility.

The PostgreSQL Global Development Group has updated their repository RPMs to track the specific minor version of RHEL, Rocky Linux, and AlmaLinux rather than just the major release. Previously, running dnf update on an older minor release would upgrade the repo configuration to the latest minor version, causing the system to remain on that newer package set even after the OS was upgraded. This fix ensures that the repository metadata aligns with the actual OS minor version installed on the host. Administrators are advised to update their repository RPMs to restore correct behavior.

• Fixes issue where dnf update locked repo to latest minor version regardless of OS state
• PostgreSQL YUM repo RPMs now match the specific minor version of RHEL/Rocky/AlmaLinux
• Update repo RPMs immediately to ensure correct package availability after OS upgrades
• Applies to RHEL 9.6+ and compatible downstream distributions

The author argues that hardcoding GitHub references in Go applications creates unnecessary coupling and reduces portability. By abstracting version control systems, teams can switch providers or manage credentials more flexibly without refactoring core logic. This approach supports better separation of concerns and easier testing.

• Avoid hardcoding GitHub URLs or tokens directly in application logic.
• Abstract VCS interactions behind interfaces for easier swapping of providers.
• Simplifies local development and CI/CD pipelines by decoupling infrastructure.
• Reduces risk of vendor lock-in and improves code portability.

Amazon DynamoDB Global Tables with Multi-Region Strong Consistency (MRSC) now supports five new regions, including Mumbai and Singapore, bringing the total to 15. Users can configure global tables with either three active replicas or two replicas plus one witness region across any supported combination. This expansion enables zero recovery point objective (RPO) architectures spanning North America, Europe, and Asia Pacific.

• MRSC now covers 15 regions, adding Canada, Stockholm, Spain, Mumbai, and Singapore
• Supports three-replica or two-replica-plus-witness configurations across any supported regions
• Enables zero RPO data consistency for applications spanning multiple continents
• Allows routing traffic to geographically closer endpoints for lower latency

Amazon RDS for SQL Server now allows Multi-AZ deployments with Always On Availability Groups for the Developer Edition engine. This feature provisions a synchronous standby replica in a separate Availability Zone, enabling automatic failover during infrastructure outages. It permits teams to validate high-availability architectures and test failover behaviors in non-production environments without paying Enterprise Edition licensing fees.

• Enable Multi-AZ on SQL Server Developer DB instances to mirror production HA setups.
• Validate Always On Availability Group failover logic in dev/test before production rollout.
• Eliminate Enterprise license costs while testing synchronous standby replication.
• Use this setup to stress-test infrastructure failure scenarios safely.

Fleet impact: You can now validate production-equivalent HA and failover behaviors for SQL Server workloads using zero-cost Developer Edition instances. Ensure your CI/CD pipelines include Multi-AZ failover tests for Developer DBs to catch configuration drift early.

A 2026 review of Rust's SIMD ecosystem highlights significant maturation in standard library support and cross-platform consistency. The article details how modern intrinsics reduce the friction of writing vectorized code while maintaining safety guarantees. It also benchmarks current performance against previous iterations, showing measurable improvements in data processing workloads.

• Cross-platform SIMD support is now more consistent across major architectures.
• Standard library intrinsics reduce boilerplate for common vector operations.
• Performance benchmarks show tangible gains over 2024-2025 baseline versions.
• Safety guarantees remain intact even with low-level vector manipulation.