Upgrade to Pro — share decks privately, control downloads, hide ads and more …

TiDB: Overview, New Features, Multi-Cloud

Avatar for Kevin Xu Kevin Xu
November 07, 2018

TiDB: Overview, New Features, Multi-Cloud

This keynote talk was delivered with Li Shen, VP of Engineering at PingCAP, at Percona Live Europe 2018, to provide a technical overview of TiDB, new features in version 2.1, and multi-cloud deployment using a Kubernetes Operator.

Avatar for Kevin Xu

Kevin Xu

November 07, 2018
Tweet

More Decks by Kevin Xu

Other Decks in Technology

Transcript

  1. !2 What is Ti(tanium) DB? • NewSQL (RDBMS at Scale)

    • MySQL Compatible (5.7) • Hybrid OLTP + OLAP Workloads • Multi-Cloud
  2. !3 Product: the TiDB Platform • TiDB Platform: • TiDB

    (stateless SQL layer) • TiKV (distributed transactional key-value store on RocksDB) • TiSpark (Apache Spark plug-in to TiKV) • Open Sourced from Day 1 • GA 2.0: April 2018
  3. Community !4 • Stars: • TiDB: 15,600+ • TiKV: 3,800+

    • Contributors: • TiDB: 210+ • TiKV: 105+
  4. TiDB-DM !13 • Why? To support migrating data from MySQL/MariaDB

    to TiDB • Full data migration + binlog replication • Features: • Data migration job management • Monitoring system • Sharded table merge migration • Data transformation
  5. TiDB-Lightning !17 • Why? Enables fast, full import of SQL

    dump -> TiDB cluster • Fast: • Can import > 100 GB / hour on production grade hardware • Full Import: • Import the entire database once from scratch; not incremental import • Suitable for migrating an existing database to a new TiDB cluster • TiDB Cluster: • Lightning interacts with TiDB’s internal structures • Only works with TiDB cluster
  6. TiDB-Binlog !19 • Why? Provides real-time data synchronization and backup

    • Synchronizing downstream data platform in real-time to: • Hot backup • MySQL slave • OLAP data warehouses • Writing incremental backup files • Event subscription & trigger
  7. TiDB 2.1 - SQL Layer Smarter SQL optimizer • More

    efficient plan • Dynamic stats update based on query feedback Faster aggregation operator • New aggregation framework • Multi-thread hash aggregation operator 50% faster OLTP point query !23 Faster online DDL • Run DDL concurrently • Faster “Add Index” operation Improved debugging & diagnosis • More readable EXPLAIN result • “Explain Analyze” statement • Expose cluster information through HTTP API
  8. TiDB 2.1 - Storage Engine (aka TiKV) Raft learner •

    Improve cross-DC data availability Hotspot scheduling • Automatically distribute hotspots to more idle instances Batch region-splits • Avoid large region formation under heavy write workloads 2X read performance improvement !24
  9. TiDB As A Service !29 • Fully Managed by PingCAP

    • Upgrades, Patching, Online DDL with No Downtime • Backup and Restore • Multi-Cloud Deployment https://www.pingcap.com/tidb-cloud/
  10. See it to believe it !32 • PingCAP Booth #6

    • Demo by Greg Weber TODAY at 1:20PM