Known Issues
Known Issues
Section titled “Known Issues”This page tracks active bugs and sharp edges in the current release (v0.11.0).
It is deliberately short and honest: if something here bites you, you are not doing
it wrong.
- For things CQLite does not do by design or yet, see Limitations.
- For planned work and milestones, see Roadmap.
- Hit something not listed here? Open an issue — that is the single most useful thing you can do for the project.
Last reviewed: 2026-06-17 (v0.11.0).
Python bindings
Section titled “Python bindings”SET<FROZEN<UDT>> fails to deserialize (#804)
Section titled “SET<FROZEN<UDT>> fails to deserialize (#804)”Querying a table with a set<frozen<UDT>> column through the Python bindings raises
a TypeError: UDT values are decoded as dicts and the binding tries to place them in
a frozenset, but dicts are unhashable. The CLI and Rust reads of the same table are
unaffected. Tracked in #804.
Concurrent queries can race (#805)
Section titled “Concurrent queries can race (#805)”Multiple threads issuing queries against the same Database object can
intermittently fail with Column not found: id. A warm-up query reduces but does not
fully eliminate the race. Workaround: open one Database per thread, or serialize
queries through a single thread. Tracked in
#805.
Performance
Section titled “Performance”Wide partitions scan linearly (#751, #752)
Section titled “Wide partitions scan linearly (#751, #752)”SSTables written by CQLite emit promoted_index_length = 0, so within-partition
seeks fall back to a linear scan.
- Narrow partitions (< 100 rows): no measurable impact.
- Wide partitions (10 000+ rows): O(n) scan within the partition.
Reading SSTables written by Cassandra is unaffected — this only concerns the CQLite write path. The promoted-index writer and BTI-based O(log n) seeks are on the roadmap (epic #751).
Format coverage
Section titled “Format coverage”BTI (da) SSTables are rejected, not read
Section titled “BTI (da) SSTables are rejected, not read”BTI/trie-index SSTables (da-*-bti-*, opt-in in Cassandra 5.0) are detected and
rejected with a clear error rather than misread. This is by design until the
dedicated BTI read path lands — see
Limitations and roadmap item
#660.
Reporting something new
Section titled “Reporting something new”If you hit a bug — wrong output, a panic, a parsing error, a performance cliff — please report it. Good reports include:
- The Cassandra version that wrote the SSTable and the file names (e.g.
nb-1-big-Data.db). - The CQL schema for the table.
- The exact command or code you ran and the output (or a hex dump for parsing issues).