Performance: Tips & Tricks to Optimize the Speed of Your Mission-Critical Database
summary
Concrete YugabyteDB tactics: pick primary keys that avoid extra RPCs (make email the PK instead of a serial id), prefer high-cardinality hash keys or explicit range sharding for timestamps, and size tablets ~6–8 per device. Use YB_ENABLE_EXPRESSION_PUSH_DOWN and follower reads (rf3/rf5 with configurable staleness) for fewer round trips, create covering and partial indexes (index-only scans limited to 32 columns), pack infrequent fields in JSONB, and enable packed rows (v2.15+) to cut write amplification and read latency.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 0:21Speed Costs Money
- 1:07Areas of Focus
- 1:22Data Modeling
- 1:53Key Organization
- 2:59Primary Key
- 5:50Pass Splitting
- 8:39Range Splitting
- 10:17Change Primary Keys
- 12:15Use Less Reference Columns
- 14:04Use the Right Number of Tablets
- 15:48Avoid Common Mistakes
- 17:49Mask Tablespaces
- 19:05Example
- 20:08Performance Structures
- 21:21Partial Indexes
- 22:39Duplicate Indexes
- 23:49Session tunables
- 25:14Followers
- 26:24Session Tunable
- 27:41Packed Rows
- 29:16Summary
related talks
29mData Modeling for Distributed SQL
33mMigrating from Oracle to
26mHow to Assess Your Cloud Native Workloads for Migration to a Distributed SQL Database
26mEfforts Toward Adoption and Uncovering Hidden Features
26mUnderstanding the Query Execution in YugabyteDB
18mBulletproofing your YugabyteDB Infrastructure to Reduce Risk
30mTechnical Comparison of YugabyteDB with Oracle and PostgreSQL
30mBuilding Cloud Native Applications on Distributed SQL Platform