Practical SQL Server diagnostics: SQL Server Performance Office Hours, Episode 81
Erik Darling · September 17 · Performance and troubleshooting
Key takeaway: The most useful section explains how worker-thread starvation appears operationally: rising THREADPOOL waits, gaps where monitoring queries cannot obtain workers, followed by a large wait-stat spike. The episode also distinguishes latch waits from lock waits and explains why Query Store timings may disagree with user-visible latency.
Client use: Collect wait-stat deltas externally or continuously. A diagnostic query launched after the server becomes unresponsive may itself be unable to run.
Official Microsoft update: Azure SQL updates for mid-September 2026
Microsoft Azure · September 16 ·
Key takeaway: Azure SQL is previewing configurable soft deletion for logical servers. A deleted server enters a recoverable state for the configured retention period instead of becoming immediately unrecoverable.
Client use: This adds protection against administrative mistakes, but it is not a substitute for database backup retention. Test server restoration, document the retention period, restrict purge capability and ensure infrastructure automation handles soft-deleted names correctly.
Cross-platform availability: Group Replication Beyond a Single Cluster: DC-DR with Percona MySQL Operator
Anil Joshi, Percona · September 17 · MySQL disaster recovery
Key takeaway: The walkthrough builds an InnoDB ClusterSet in which each site uses synchronous Group Replication locally while asynchronous replication connects the DR site. For large databases, pre-seeding from backup can avoid the donor load and transfer time of cloning the entire dataset.
Client use: Automation can create topology, but it does not choose acceptable RPO, validate network capacity or prove failover behavior. Those still require explicit design and rehearsals.
Patch action: KB5104824, SQL Server 2022 CU27
Microsoft Support · September 15 · Cumulative update and security
Key takeaway: Build 16.0.4295.3 contains 41 fixes, including a rare corruption defect during transaction rollback when ADR is disabled, a slow VSS restore issue on servers with hundreds of databases, an AG join failure with 64-character group names, and multiple security fixes. It retains documented issues involving parallel SESSION_CONTEXT, MSDASQL linked servers and monitoring during database recovery.
Client use: Prioritize estates using managed backup, VSS, replication, contained AGs or ADR-disabled databases. Regression testing should cover recovery-time monitoring and linked servers, not merely service startup.
Additional deep practical analysis: Recovering TPS After a Cross-Database Migration
Vinay Kumar Dumpa, Microsoft PostgreSQL Engineering · September 16 · Migration performance
Key takeaway: Scaling compute barely helped a CPU-saturated migration. Rewriting a high-frequency NOT IN query as NOT EXISTS enabled an anti-join; only then did an index on the foreign-key column become useful. The reported workload improved from 15.5 to 12,963 TPS.
Client use: After a heterogeneous migration, rank statements by total resource consumption and inspect plan shape before recommending larger hardware or adding indexes.
