Jepsen: Amazon RDS for PostgreSQL 17.4 (jepsen.io)

Article summary

Amazon RDS for PostgreSQL multi-AZ clusters have been found to violate Snapshot Isolation, the strongest consistency model supported across all endpoints. This means that healthy clusters can occasionally allow Long Fork and other G-nonadjacent cycles, which can lead to inconsistencies in the database. The issue has been observed in every version of PostgreSQL tested, from 13.15 to 17.4. Amazon RDS for PostgreSQL may instead provide Parallel Snapshot Isolation, a slightly weaker consistency model.

Main themes

• Distributed database consistency
• Amazon RDS for PostgreSQL
• Jepsen testing
• MongoDB reliability
• Technical writing and communication
• Distributed system complexity

What commenters say

• MongoDB is still widely used in production despite its known issues with data consistency.
• The documentation for Amazon RDS for PostgreSQL does not adequately address isolation and serializability.
• Some databases, such as FoundationDB, have been reported to pass Jepsen testing without data loss bugs.
• The complexity of distributed systems makes it difficult for users to understand and work with them.
• The industry has become increasingly unforgiving, making it harder for new developers to start from scratch.
• Academic-style writing can be a barrier to understanding technical topics, but practice and patience can help.
• The use of specific terminology and hypertext can make technical writing more accessible to readers of different knowledge levels.