OLAP, OLTP, Zheap, and Tradeoffs in Postgres MVCC

We went through a brutal battle with Postgres bloat recently. Indexes were composed of 80% bloat, 20% useful data after years of heavy write activity. Vacuuming indexes was getting slower and slower, bringing us precariously close to transaction ID wraparound outages.

Glad that pg_repack exists, which is an extension that can be used to elegantly rebuild your indexes from scratch.

Looking forward to zheap! (Which should obviate the need for index rebuilding)