返回资讯中心
外部精选
软件工程
#开发者实时流

PG Finds – PostgreSQL Articles and Tools from Hacker News

Filter by formatAll formatsArticleDocumentationGuideProjectResearchFilter by HN mention yearAll years2026202520242023

Hacker News 最新albedoweb2026/11/11 · 7 天前11 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

My reading list0Save a find for later. Stored only on this device.

Found in the threads. Made for the curious.

Search the collection

Filter by formatAll formatsArticleDocumentationGuideProjectResearchFilter by HN mention yearAll years2026202520242023

263 finds to explore

StackGres describes Query Routers, optional Citus entry-point nodes that hold metadata but no shards, intended to remove the single-coordinator bottleneck. It reports vendor-run pgbench results reaching about 1.43M tps with three routers.

PlanetScale describes TIN v1.0.6, its PostgreSQL full-text search index extension, which adds Snowball-based stemming for 18 languages and block-max plans for top-k queries with tiebreakers. It reports benchmark results against ParadeDB on an 85 GB Stack Exchange corpus, with throughput and latency figures attributed to PlanetScale.

Databricks describes its LTAP architecture, in which Spark builds valid Postgres heap and index pages in parallel and writes them to shared storage while the primary records only a manifest WAL record. It explains sandboxed binary-upgrade instances, frozen tuples, checksum validation and a helper-table index build, and reports vendor-measured load times.

The article describes a Postgres-backed escalation queue that used SAVEPOINT subtransactions and SKIP LOCKED, then suffered MultiXact LWLock contention under load. It reports that separating claiming from ticking, removing subtransactions, and reducing claim scans improved peak escalation throughput by 3.5x.

Databricks announces general availability of the lakebase_vector and lakebase_text extensions for Lakebase Postgres, presenting them as scalable vector and BM25 search built into the database. The post describes hierarchical IVF clustering, RaBitQ binary quantization and vendor-run benchmarks on a 100M-vector dataset.

The guide ranks pg_stat_statements entries by total time, mean time, calls, I/O and temp spill, shows reset and snapshot-delta approaches, and summarizes column changes from PostgreSQL 13 through 18. It finishes with EXPLAIN (ANALYZE, BUFFERS) interpretation and an ordered troubleshooting checklist tested on PostgreSQL 17.11.

The author describes implementing feature flags as rows in a single PostgreSQL key-value table instead of running a dedicated flag service. Lookups check a business-level override before the global value, and the override fails closed when the read fails rather than falling back to the global setting.

The article measures how RLS policy formulation changes query plans and latency on PostgreSQL 17.10, comparing setting-based predicates, helper-function volatility, membership subqueries and non-leakproof expressions. It reports that volatile PL/pgSQL helpers, IN (SELECT ...) membership checks and non-leakproof functions such as lower() can each prevent index use.

A Neon blog post introducing pg_redact, a demo support inbox that uses the Jev classification model to label PII candidate spans and stores them as JSONB beside each message body. It defines two SQL functions, pii_sensitivity() and redact(), which rank PII types and mask spans above a viewer role's clearance.

The article measures FIFO and per-tenant round-robin dequeue with FOR UPDATE SKIP LOCKED on PostgreSQL 17.10, reporting small-tenant wait times and a query-plan regression on a skewed backlog. It explains how a sub-select ordering choice drove dequeue latency from milliseconds to about ten seconds until a matching partial index was added.

Tomas Vondra examines the 12 features reverted from the PostgreSQL 19 development cycle and classifies each by whether the underlying issues were actually found by AI tooling. He compares revert counts and timing against releases since PostgreSQL 14 and argues that increased AI-driven CVE handling, not AI bug discovery in patches, better explains the stabilization delay.

ClickHouse engineers explain why Postgres's MemoryContext and palloc model plus PG_TRY/PG_CATCH longjmp error handling interact badly with C++ exceptions and object destructors in extensions. They describe the different isolation strategies used in pg_clickhouse, pg_re2, pg_chdb and pg_stat_ch, and state that routing via ereport(FATAL) is a mitigation rather than a crash-isolation guarantee.

A model-assisted selection, with sources you can inspect. Project status may have changed. Read the caveats ↗

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →