We're encountering occasional deadlocks in a mirrored IRIS deployment. How can we track down which global or object write caused the lock cycle, and how does IRIS mirror lock propagation internally?

We use Apache Flink for complex event processing. Is there a way to integrate IRIS (as a source/sink) with Flink’s streaming API, possibly using the IRIS Native API or JDBC?

For geographically distributed nodes using async mirroring or ECP, how can I detect and resolve data conflicts manually (custom logic) while maintaining eventual consistency?

I’m using recursive CTEs for hierarchical data, but the planner seems to produce inefficient plans. Can I influence or extend the query optimizer behavior in IRIS?

We require automatic injection of security predicates at runtime, depending on the user or API token. Is there a supported or hackable mechanism to manipulate SQL parsing/compilation before execution?

I have large joins involving millions of rows. How can I profile and tune the SQL engine’s parallel execution? Are there EXPLAIN plan features to inspect threading and task distribution?

This anthropic article made me think of several InterSystems presentations and articles on the topic of data quality for AI applications. InterSystems is right that data quality is crucial for AI, but I imagined there would be room for small errors, but this study suggests otherwise. That small errors can lead to big hallucinations. What do you think of this? And how can InterSystems technology help?

https://www.anthropic.com/research/small-samples-poison

We are using IRIS with a sharded architecture. Complex SQL queries (with joins, aggregates, and subqueries) are performing slowly. How can I design queries or indexes to optimize distributed execution across shards?

Can the IRIS SQL engine be extended with custom optimization rules (e.g., prioritizing certain indices or join orders)? If not, is there a supported way to influence cost models?