Qdrant 1.12.2 release notes, 8 Nov 2024

Change log

Improvements

https://github.com/qdrant/qdrant/pull/5313, https://github.com/qdrant/qdrant/pull/5314, https://github.com/qdrant/qdrant/pull/5303, https://github.com/qdrant/qdrant/pull/5317 - Improve performance of matrix API
https://github.com/qdrant/qdrant/pull/5219, https://github.com/qdrant/qdrant/pull/5251, https://github.com/qdrant/qdrant/pull/5254, https://github.com/qdrant/qdrant/pull/5305 - Improve performance by not capturing backtrace if not necessary
https://github.com/qdrant/qdrant/pull/5336, https://github.com/qdrant/qdrant/pull/5365, https://github.com/qdrant/qdrant/pull/5393 - Improve HNSW search performance by tweaking visited list
https://github.com/qdrant/qdrant/pull/5301, https://github.com/qdrant/qdrant/pull/5309 - Improve performance by reusing RocksDB options, do not recreate them all the time
https://github.com/qdrant/qdrant/pull/5265 - Add support for reinitializing consensus with new peer URLs
https://github.com/qdrant/qdrant/pull/5373 - Put quantized data in RAM by default to prevent competing with memory mapped vectors
https://github.com/qdrant/qdrant/pull/5252 - Use streaming creation of snapshot in shard snapshot transfer
https://github.com/qdrant/qdrant/pull/5343 - Prevent killing replicas eagerly if node is out of sync, making replicas more robust
https://github.com/qdrant/qdrant/pull/5266 - Report memory usage in metrics and telemetry
https://github.com/qdrant/qdrant/pull/5307 - Enable Jemalloc in RocksDB
https://github.com/qdrant/qdrant/pull/5312 - Enable Jemalloc background thread to gradually release unused memory
https://github.com/qdrant/qdrant/pull/5292 - Report storage bytes estimate for each segment in telemetry
https://github.com/qdrant/qdrant/pull/5380 - Add log message when shard transfer is aborted
https://github.com/qdrant/qdrant/pull/5391 - Prevent unused repeated calls to other peers when waiting on /readyz
https://github.com/qdrant/qdrant/pull/5180 - Report CPU endianness in telemetry output
https://github.com/qdrant/qdrant/pull/5228 - Minimize point version lookups during search
https://github.com/qdrant/qdrant/pull/5399 - Improve error when using unknown vector name

Bug fixes

https://github.com/qdrant/qdrant/pull/5346 - Fix mishandling point deletions with payload indices
https://github.com/qdrant/qdrant/pull/5211 - Fix panic in query API if vector name is missing
https://github.com/qdrant/qdrant/pull/5232 - Fix potential panic if Qdrant crashed during sparse vector deletion
https://github.com/qdrant/qdrant/pull/5271, https://github.com/qdrant/qdrant/pull/5286, https://github.com/qdrant/qdrant/pull/5302 - Fix potentially missing an update on WAL delta transfer with empty delta
https://github.com/qdrant/qdrant/pull/5141, https://github.com/qdrant/qdrant/pull/5147 - Fix consensus errors on new peers if first peer is removed, now remember first ever peer
https://github.com/qdrant/qdrant/pull/5367 - Properly update all collection configuration properties when Raft snapshot is applied
https://github.com/qdrant/qdrant/pull/5350 - Fix consensus synchronization waiting on the wrong value, now speeding up synchronization
https://github.com/qdrant/qdrant/pull/5383 - In /readyz wait for consensus operations to be applied, instead of just being received
https://github.com/qdrant/qdrant/pull/5227, https://github.com/qdrant/qdrant/pull/5296 - Fix shard snapshot operations when using S3
https://github.com/qdrant/qdrant/pull/5237 - Properly clean up temporary snapshot files when using S3
https://github.com/qdrant/qdrant/pull/5253 - Properly clean up temporary segment files if optimization is cancelled
https://github.com/qdrant/qdrant/pull/5386 - Fix flushing of point ID mappings potentially causing inconsistencies