Infrastructure & tools

PlanetScale, summarized in your digest

Posts from PlanetScale on databases, MySQL, and Postgres. Snapbyte.dev reads each post as it is published and summarizes it alongside the rest of your sources.

About 4 posts a month

What it covers

The topics Snapbyte.dev most often assigns to PlanetScale posts.

Latest ranked stories

Recent PlanetScale posts

These stories are ranked from recent public source activity and shown as a preview of what a configured digest can deliver. Summaries are AI-generated; open the original source for full context.

01planetscale.com

Designing a reliable unplanned failover

Designing a reliable unplanned failover

Neki’s unplanned Postgres failover uses replica evidence rather than Admin’s health check alone. It declares a primary dead only when reachable replicas report broken replication without contradiction, then disconnects them to prevent the old primary from returning and to freeze WAL positions. It promotes the replica with the highest received, preferably fully replayed, WAL, reconfigures the shard, and tears down the old primary. Synchronous `ANY 1` durability helps establish write unavailability, while TCP and WAL receiver timeouts detect silent or frozen failures.

02planetscale.com

The hidden latency inside your database

The hidden latency inside your database

Network-attached storage is flexible and durable, but its network hop can hide significant latency inside cloud databases. SSDs made storage fast enough that this hop became a bottleneck, while locally attached NVMe avoids it and removes EBS IOPS caps. PlanetScale Metal combines local NVMe with primary-replica clusters across availability zones, cross-zone commit acknowledgements, and failover. It can improve read-heavy, cache-miss, or IOPS-bound workloads, though RAM-cached workloads may not need it.

03planetscale.com

Designing a planned switchover for minimum disruption

Designing a planned switchover for minimum disruption

Neki minimizes planned Postgres primary switchovers by validating candidates beforehand, establishing a final WAL position, preserving synchronous replication, and buffering primary-bound queries. Admin makes the old primary read-only, disconnects other replicas, waits for the target to replay all required WAL, prepares replication, and promotes it only when safety checks pass. Routers then replan buffered queries, so applications usually see delay rather than errors. A deadline enables rollback to the old primary, while acknowledged commits remain durable.

04planetscale.com

What was the turning point for Postgres?

What was the turning point for Postgres?

Postgres became today’s default through a sequence of moments rather than one decisive event, with Heroku’s 2011 decision to make hosted Postgres effortless arguably the pivotal catalyst. Its open governance and extensibility ensured survival; JSONB preserved relevance against MongoDB, while MySQL’s Oracle acquisition weakened its momentum. Postgres overtook MySQL in Stack Overflow’s 2023 survey (45.6% vs. 41.1%) and reached 86% of 32,955 LLM database recommendations, reinforced by cloud availability and pgvector.

05planetscale.com

Shard design doesn't start from your tables

Shard design doesn't start from your tables

Neki shard-group design should start from query and transaction patterns, not table structure. Tables sharing a key can be co-located across Postgres shards using a common shard index, keeping joins and transactions local and preserving full single-shard semantics. Reference tables are replicated to every shard, while differently keyed data can use separate groups. Global secondary indexes support lookups without shard keys. MoveTables and Reshard later change placement, shard keys, or shard counts as workloads evolve.

06planetscale.com

I don't trust ON DELETE CASCADE

I don't trust ON DELETE CASCADE

ON DELETE CASCADE can turn simple deletions into outages and replication failures as foreign-key trees grow. A cascading delete removes matching rows in one transaction, generating dead tuples, index updates, and WAL while holding locks; migrations may queue writes, unindexed keys can trigger scans, and logical replication can leave orphans or fail without replica identity. Soft deletes avoid this by marking one row, then reaping descendants in small background batches, keeping transactions short and bounded.

07planetscale.com

Neki update: $50 Metal, read-only replicas, Azure, and more

Neki update: $50 Metal, read-only replicas, Azure, and more

Neki, PlanetScale’s sharded Postgres, added read-only replicas, Metal nodes starting at $50/month, Azure regions, and broader Terraform, CLI, MCP, and configuration support during its platform preview. The query layer now supports more SQL, standard exports via COPY, and EXPLAIN (NEKI_PG_PLAN); distributed INSERT ... SELECT is 76x faster and COPY ingest 5x faster than launch. More than 650 changes have shipped, enabling evaluation for production workloads.

08planetscale.com

Sequences and other volatiles in Neki

Sequences and other volatiles in Neki

Neki implements PostgreSQL sequences in its sharded query planner by making sequence operations router-only and coordinating allocation through an authoritative shard. Because nextval() is volatile, Neki must evaluate it once per output row, even when rows arrive from multiple shards, while stable expressions such as now() are evaluated once per statement. Shard-local joins and immutable expressions can still be pushed down. For UPDATE, Neki locks and streams row IDs to the router, then sends parameterized per-row updates with globally unique sequence values.

09planetscale.com

PlanetScale is now available on Azure

PlanetScale is now available on Azure

PlanetScale is now available on Microsoft Azure, initially supporting PlanetScale Postgres and Neki databases in centralus, eastus2, and westeurope; Vitess is not yet supported. Azure deployments include high availability across three availability zones, Query Insights, schema recommendations, automated backups, and Metal clusters. PlanetScale built custom Kubernetes networking that assigns pod IPs in /28 blocks, increasing capacity 16x to over one million pod IPs per network. Existing databases can be migrated through support.

10planetscale.com

What gets TOASTed in Postgres

What gets TOASTed in Postgres

Postgres TOAST activates when a row exceeds 2,032 bytes: it compresses variable-length values or moves them into a `pg_toast` table in chunks. A 2,001-character `text` value therefore compresses with LZ4 while 2,000 characters remain uncompressed. Postgres 19 makes LZ4 the default where supported. LZ4 is generally the best choice for speed, while pglz can produce smaller repetitive JSON and better preview reads; `EXTERNAL` suits already-compressed data and random byte-range access.

11planetscale.com

TIN v1.0.6: 2-3x faster, with support for stemming

TIN v1.0.6: 2-3x faster, with support for stemming

TIN v1.0.6 adds language-specific stemming for 18 languages, supports heap-attribute tiebreakers in ranked top-k queries, and improves top-k performance without changing the index format. Benchmarks on 150 million Stack Exchange documents show TIN outperforming ParadeDB across conjunction, disjunction, phrase, and mixed queries: default scoring delivers 3.1–5.0x higher throughput and 2.7–4.4x lower p99 latency on x86-64. Existing indexes need no rebuild for the performance update; stemming requires rebuilding.

12planetscale.com

What today's software owes to Bronze Age clocks

What today's software owes to Bronze Age clocks

Leaky buckets, first used as ancient water clocks, became a low-cost software pattern for admission control: work is admitted when capacity exists and throttled as the bucket refills over time. Shopify uses query-cost buckets for GraphQL rate limiting. In 2026, PlanetScale brought similar decision gates to Postgres with Database Traffic Control, which estimates query costs, applies tagged resource budgets, and can warn or block queries based on server share, burst capacity, per-query limits, and concurrent workers.

Related pages

Continue comparing workflows, sources, and methodology.

Snapbyte.dev workflow

Get PlanetScale in your digest

Pick your topics, add PlanetScale, and receive one summarized digest on your schedule.