Hi Cloudberry Team, I wanted to share findings from testing PostGIS 3.3.2 with Cloudberry and see if we can learn from this experience together.
Issue Summary: While attempting to validate the PostGIS 3.3.2 integration by running its embedded regression test suite, I encountered critical memory corruption in distributed queries with spatial predicates. I've filed a detailed bug report with reproduction steps: GitHub Issue: [Bug] PostGIS Memory Corruption in Distributed Querieshttps:// github.com/apache/cloudberry/issues/1395 The issue manifests as consistent crashes (mcxt.c:933 assertion failures) when executing cross-segment joins with geometry operations like ST_Contains and ST_Intersection. Single-segment queries work fine. Testing Approach: My goal was straightforward: run PostGIS's own regression test suite against Cloudberry to validate the integration. The embedded tests are designed to exercise PostGIS functionality comprehensively, and they revealed this distributed architecture incompatibility immediately. Questions for the Community: I'm curious about the testing methodology used during the PostGIS 3.3.2 upgrade work: 1. What testing was performed to validate the PostGIS 3.3.2 integration? 2. Were the PostGIS embedded regression tests run against Cloudberry's distributed architecture? 3. If testing was done, were distributed queries with geometry joins tested, or primarily single-segment scenarios? I'm asking not to point fingers, but to understand where we can improve our extension testing practices. This seems like a gap we should address systematically. Opportunity for Improvement: Running upstream extension test suites against Cloudberry's distributed architecture seems like valuable validation that could catch integration issues early. Would the community be interested in: - Establishing testing guidelines for extension upgrades? - Sharing test frameworks/approaches for distributed query validation? - Creating a checklist for extension integration testing? The full technical analysis, core dumps, and test reproduction suite are available in the GitHub issue. Happy to discuss findings or testing approaches. Best regards, -=e -- Ed Espino Apache Cloudberry (Incubating) & MADlib
