Pull Request Explorer

Summary

Fix negative LIMIT on covering index queries

Corrects silently wrong results for SELECT with LIMIT -N over covering index with filter, ensuring last N rows are returned instead of first N.

Health Assessment

• • The PR had multiple review rounds and commits, indicating some complexity and potential for rework. However, the fix is well-structured and includes comprehensive tests.

Summary

Fix crash when deduplicating STRING keys over 2GB

This fix prevents database crashes caused by large STRING dedup keys, ensuring reliable data ingestion. It eliminates boot-time crash loops for affected tables.

Health Assessment

• • Quick turnaround with minimal rework; the PR resolved a critical crash with a single commit and was merged within 13 hours.

Summary

Split Windows/Mac CI into separate pipeline

Reduces CI cost by running Windows/macOS tests only on demand, improving build speed and resource usage.

Health Assessment

• • Fast review and merge with minimal changes to CI config, indicating low risk and high efficiency.

Summary

Add fluent query assertion builder and fix bugs

Introduces a fluent builder for query tests, simplifying test code and exposing nine production bugs, improving reliability and maintainability.

Health Assessment

• • The PR had a moderate cycle time with a single review round, but the large code churn and production bug fixes indicate significant risk.

Summary

Migrate array_agg to batch-descriptor merge API

This change restores the build and aligns array_agg with the efficient batch‑descriptor merge mechanism, preserving result correctness while improving memory usage for high‑cardinality GROUP BY queries.

Health Assessment

• • Merged within 1.5 hours with no conflicts, indicating a straightforward refactor with minimal risk to the codebase.

Summary

Improve performance of twap() and sparkline() aggregates

Enhances query performance for time‑series analytics, reducing latency for large datasets and tightening correctness by enforcing timestamp constraints.

Health Assessment

• • Long review cycle (10 days) and significant rework indicate high friction; the PR introduces large code changes that improve performance but also alter behavior, raising risk of regressions.

Summary

Add byte caps for CachedWindow and ORDER BY

Introduces configurable memory caps to prevent out‑of‑memory crashes in window and ORDER BY operators, enhancing system stability for large queries.

Health Assessment

• • Multiple iterations and a large code change set indicate significant rework; long cycle time and delayed reviews suggest bottlenecks and elevated risk for downstream impact.

Summary

Fix parquet sort metadata consistency in O3 merges

Ensures replica restores from backups succeed by preventing inconsistent sort metadata, improving reliability for enterprise replication.

Health Assessment

• • Long review time and multiple rework commits indicate complex changes; high risk due to critical replica crash fix.

Summary

Fix Parquet read of added columns as NULL

Ensures queries selecting columns added after Parquet conversion return NULL instead of crashing or stale data, improving stability for analytics workloads.

Health Assessment

• • The PR resolved a critical crash scenario with minimal review friction, but involved a large code change across two files, indicating significant refactoring of Parquet handling logic.

Summary

Validate months-based TTL granularity on materialized views

Ensures materialized views with derived partitions reject invalid months-based TTL values, preventing misconfigurations and potential data integrity issues. This tightens validation, improving reliability for users creating materialized views.

Health Assessment

• • Quick fix with minimal changes, merged within 15 hours, indicating efficient review and low risk.

Summary

Add storage policy TTL floor-based check

Introduces a helper to compute storage policy TTL based on partition floor, aligning internal logic and adding unit tests.

Health Assessment

• • PR merged after ~45 hours with no review comments, indicating a low‑complexity refactor with minimal friction.

Summary

Single-pass Parquet Decode for Covering Index

Reduces disk I/O and decoding passes, improving query performance and lowering resource usage.

Health Assessment

• • Performance optimization required a single review cycle but involved a large code change, indicating moderate review friction and high risk due to scope.

Summary

Fix storage policy for tables with posting indexes

Corrects a bug that prevented tables with posting-indexed symbol columns from switching to parquet storage, ensuring proper hard‑linking of index files and preventing partition switch failures.

Health Assessment

• • The PR resolved a critical storage policy issue with a concise change set and a quick review cycle, demonstrating efficient problem resolution.

Summary

Health Assessment

Summary

Fix rare JIT compiler crash in SQL

Addresses a JIT compiler crash caused by allocator boundary overflow, improving stability for SQL query execution.

Health Assessment

• • Quick fix with minimal changes, resolved in under 20 hours, indicating efficient review and low complexity.

Summary

Fix statistical aggregates returning NULL on sparse-NULL data

Corrects aggregate functions (corr, stddev, var, covar) that returned NULL on sparse-NULL data in parallel GROUP BY, ensuring accurate results and data integrity.

Health Assessment

• • Single approval after a quick review; tests added to validate the fix; large code changes focused on a specific bug, minimizing overall risk.

Summary

Add git low-speed timeout to fuzz CI checkouts

Improves CI reliability by terminating slow Git checkouts, reducing build times and preventing long stalls on Hetzner agents.

Health Assessment

• • The PR was reviewed quickly but required several iterations to correctly gate the low-speed timeout across different agent types, indicating moderate complexity in CI configuration.

Summary

Exclude dropped datagrams from processed count

Fixes flaky test by preventing double-counting of dropped datagrams, improving reliability of UDP receiver.

Health Assessment

• • Quick resolution of flaky test with minimal changes, indicating efficient review and low complexity.

Summary

Add bloom filter inlining to parquet metadata

Enhances query performance by embedding bloom‑filter bitsets directly into partition metadata, eliminating extra reads for indexed columns and ensuring consistent metadata across migration paths.

Health Assessment

• • The PR required extensive refactoring and new tests but was reviewed in a single round and merged within two days, indicating efficient collaboration. The change improves performance for existing databases while adding a one‑off regeneration cost for older migrations.

Summary

Bump questdb-client version to 1.3.2

Ensures the project uses the latest questdb-client with pooling, lifecycle, and timeout fixes, improving stability and performance.

Health Assessment

• • Quick version bump with minimal changes, merged within 7 hours, indicating efficient review and low risk.

Summary

Add deferred commit for transactional ingestion

Enables clients to split large ingestion batches across multiple messages, reducing memory pressure and decoupling from buffer limits, improving overall ingestion performance.

Health Assessment

• • The PR involved over 1,100 lines of change across core server and client modules, requiring multiple iterations and extensive testing. This large scope and repeated revisions indicate a high-risk, high-impact change that demanded careful review and validation.

Summary

Fix covering indexes on Parquet partitions and bound RSS

Improves database stability and performance by ensuring covering indexes work correctly on Parquet partitions and reducing memory usage during sealing, preventing crashes on large data sets.

Health Assessment

• • Large-scale changes with many iterations and review rounds indicate significant complexity; memory and performance improvements are critical for production stability.

Summary

Fix WebSocket graceful close to prevent TCP RST

Prevents client connection resets when server receives oversized WebSocket frames, ensuring reliable error reporting and better client experience.

Health Assessment

• • Quick fix with minimal changes, resolved in under 2 hours, indicating efficient review and low complexity.

Summary

Add regr_r2 coefficient-of-determination aggregate

Adds the SQL aggregate regr_r2 to compute the coefficient of determination, enhancing analytical capabilities and laying groundwork for additional regression functions.

Health Assessment

• • Long review cycle with many commits indicates a complex change; large code modifications and extended cycle time increase risk of regressions and integration issues.

Summary

Fix storage policy handling in core

Improves stability and performance of storage policy operations by draining async queues, rejecting invalid commands, and updating configuration defaults.

Health Assessment

• • The PR required multiple rounds of rework after review, with a large number of commits and significant code churn, indicating a complex change that impacted core storage logic.

Summary

Remove vector group-by factory, add batch paths

Improves performance by eliminating vectorized group-by for non-keyed queries and adding batch-aware aggregation functions, reducing memory usage and simplifying plan output. This change streamlines query execution and reduces code complexity.

Health Assessment

• • Long cycle time and large code churn indicate significant rework; review friction high; potential risk to stability due to extensive changes.

Summary

Fix WebSocket test timeout for worst-case fragmentation

Increases WebSocket close‑drain timeout to prevent flaky test failures. This improves CI reliability and ensures accurate WebSocket behavior under extreme fragmentation.

Health Assessment

• • Fast review and minimal changes led to quick merge, indicating low complexity and high confidence in the fix.

Summary

Fix NULL lookup bug in Parquet partitions

Ensures accurate query results when searching for NULL values in Parquet partitions, preventing incorrect data retrieval and improving reliability.

Health Assessment

• • Merged within 22 hours with minimal review friction.

Summary

Add window function support for DECIMAL types

Enables advanced analytical queries on decimal data, improving precision and performance for financial and scientific workloads.

Health Assessment

• • PR added extensive decimal window functions with large codebase changes, but review was quick and no rework needed, indicating efficient implementation.

Summary

Improve mat view incremental refresh performance

Optimizes materialized view refresh for out‑of‑order writes, cutting refresh times from seconds to milliseconds and enhancing overall database throughput.

Health Assessment

• • Large performance optimization with extensive code changes, but review process was straightforward with minimal back‑and‑forth, resulting in a slow overall cycle time.

Summary

Stop no-op WAL transactions for materialized views

This fix eliminates excessive write-ahead log traffic caused by materialized view refreshes that performed no work, improving performance and reducing replication lag.

Health Assessment

• • Fast cycle time and single review round indicate a straightforward bug fix with minimal risk.

Summary

Fix WebSocket handshake under partial writes

This fix resolves WebSocket handshake failures during partial socket writes, ensuring reliable real-time data ingestion and preventing connection timeouts for clients.

Health Assessment

Summary

Prevent execution during SQL validation

Ensures the SQL validation endpoint runs without side effects, improving security and reducing unnecessary logging.

Health Assessment

• • The PR had a moderate cycle time with a single review, but the large code changes required extensive testing to ensure no side effects during validation.

Summary

Fix crash when adding covering index to Parquet partition

Prevents table suspension when adding covering indexes to Parquet partitions, ensuring continuous data availability.

Health Assessment

• • Resolved in a single commit with one approval; minimal code churn and no rework required.

Summary

Recover WAL sequencer metadata from committed log

Rebuilds bricked or version-mismatched WAL sequencer metadata from committed log, ensuring data consistency and preventing errors.

Health Assessment

• • The PR has a large scope with 15 files changed and over 1000 lines of code added or removed.
• • The cycle time is long, indicating potential delays or difficulties in the review process.

Summary

Allow larger query results in QWP egress

Improves reliability by preventing server errors for high‑cardinality queries, ensuring larger result sets can be transmitted without failure.

Health Assessment

• • The PR required multiple iterations to resolve test failures and buffer handling logic, indicating complex changes to core egress logic.

Summary

Fix twap and sparkline under concurrent load

Ensures accurate query results for twap() and sparkline() functions under high concurrency, preventing incorrect analytics data.

Health Assessment

• • The PR required extensive rework and multiple merges, indicating a complex concurrency bug fix with significant code changes.

Summary

Speed up Mac and Windows test jobs

Reduce CI runtime on Mac/Windows by trimming heavy tests, gating large tests to Linux, and optimizing build steps, improving CI reliability and reducing cost.

Health Assessment

• • Large refactor of CI and test suite with 5,706 lines added/removed, long review cycle (~34h) and significant runtime improvements, but risk of missing coverage on non‑Linux platforms and potential build regressions.

Summary

Fix flaky ingestion test failures

These fixes stabilize the ingestion test harness, ensuring reliable CI results and preventing false positives that could mask real issues.

Health Assessment

• • Quick fix with minimal changes, resolved flaky tests in under 15 hours, indicating efficient review and low risk.

Summary

Fix flaky rename race in ILP test

Improves stability of ILP test suite, reducing CI failures and ensuring accurate test results.

Health Assessment

• • Quick fix with minimal changes, no production code affected, low risk.

Summary

Refactor egress to keep partitions open

Improves query performance and reduces memory usage for egress operations by refactoring scan mode handling.

Health Assessment

• • Fast cycle time of 1.5h with minimal rework; performance gains demonstrated in latency metrics; low risk due to small scope and no critical feature changes.

Summary

Expand QWP ingest coverage and advertise batch cap

Enables new data types and exposes server limits to clients, reducing ingestion errors and improving throughput.

Health Assessment

• • The PR required multiple iterations to add new data types and adjust server behavior, resulting in a 4.8‑day cycle and extensive test coverage.

Summary

Prevent internal errors in symbol hash joins

Fixes internal errors when joining on SYMBOL with UUID or DECIMAL keys, ensuring queries return results instead of throwing exceptions.

Health Assessment

• • Fast cycle time and minimal rework indicate a straightforward bug fix; AI review helped surface minor nitpicks.

Summary

Health Assessment

Summary

Fix geohash precision and sentinel null handling

Accurate geohash column creation prevents incorrect data types. Proper null handling stops runtime errors during ingestion.

Health Assessment

• • Fast cycle time and minimal review iterations indicate efficient resolution of a critical ingestion bug.

Summary

Fix JSON escaping of CHAR values in API responses

Ensures API responses are valid JSON by escaping CHAR column values, preventing parser errors for clients.

Health Assessment

• • Quick fix with minimal rework; single review comment and one revert commit; low risk to production.

Summary

Expand QWP fuzz tests to cover all wire types

Expands QWP fuzz tests to cover all wire types, ensuring comprehensive validation of the protocol. This improves data integrity and reliability for downstream consumers.

Health Assessment

• • Fast cycle time with minimal review indicates a straightforward test expansion.

Summary

Fix parquet varchar ASCII flag after column add

Corrects Parquet varchar ASCII flag handling during column addition, preventing potential data integrity issues.

Health Assessment

• • Quick fix with minimal changes and fast review indicates low complexity and high confidence in correctness.

Summary

fix parquet query results for non-ascii varchars

Fixes incorrect query results after writing non-ascii varchars to parquet partitions, ensuring accurate UTF-8 handling.

Health Assessment

• • The PR was merged quickly, indicating low review friction.
• • The fix is well-structured and includes thorough testing.

Summary

Fix WINDOW JOIN crash with projected symbol column

Prevents crashes in query execution when joining tables with projected symbol columns, improving stability for analytical workloads.

Health Assessment

• • Fast cycle time and single review round indicate efficient resolution of a critical bug.