KalamDB Benchmark Report

Performance analysis — v0.3.0-alpha2

http://localhost:8080 2026-02-24T18:42:01.364879+00:00 100 iters 5 warmup 10 concurrency

Total Benchmarks

Passed

Failed

719.24s

Total Duration

Latency by Operation (µs)

Throughput (ops/sec)

Avg Latency by Category (µs)

Detailed Results

Status	Benchmark	Category	Description	Iters	Mean	P50	P95	P99	Min	Max	Ops/sec	Total	Verdict	vs Prev
PASS	create_table	DDL	CREATE TABLE with 3 columns	100	51.0ms	50.1ms	77.6ms	80.4ms	24.1ms	80.9ms	20	5.10s	🟢 Excellent	↓36% slower
PASS	drop_table	DDL	DROP TABLE on a previously created table	100	49.9ms	49.8ms	55.4ms	56.2ms	43.2ms	57.0ms	20	4.99s	🟢 Excellent	↓18% slower
PASS	single_insert	Insert	INSERT a single row into a table	100	280µs	268µs	387µs	426µs	231µs	475µs	3.6K	28.0ms	🟢 Excellent	↓16% slower
PASS	bulk_insert	Insert	INSERT 50 rows in a single statement	100	649µs	613µs	832µs	954µs	560µs	956µs	1.5K	64.9ms	🟢 Excellent	↓9% slower
PASS	select_all	Select	SELECT * from a 200-row table	100	1.1ms	1.1ms	1.4ms	1.6ms	1.0ms	1.8ms	872	114.7ms	🟢 Excellent	↓8% slower
PASS	select_by_filter	Select	SELECT with WHERE clause on a 200-row table	100	466µs	434µs	639µs	686µs	403µs	718µs	2.1K	46.6ms	🟢 Excellent	↓7% slower
PASS	select_count	Select	SELECT COUNT(*) on a 200-row table	100	867µs	823µs	1.1ms	1.2ms	791µs	1.2ms	1.2K	86.7ms	🟢 Excellent	↓6% slower
PASS	select_order_by_limit	Select	SELECT with ORDER BY + LIMIT 10 on a 200-row table	100	1.1ms	1.1ms	1.4ms	1.6ms	1.0ms	1.7ms	882	113.4ms	🟢 Excellent	↓8% slower
PASS	single_update	Update	UPDATE a single row by filter condition	100	545µs	492µs	814µs	932µs	453µs	947µs	1.8K	54.5ms	🟢 Excellent	↓13% slower
PASS	single_delete	Delete	DELETE a single row by filter condition	100	441µs	411µs	565µs	753µs	380µs	842µs	2.3K	44.1ms	🟢 Excellent	~ 452µs prior
PASS	concurrent_insert	Concurrent	N concurrent INSERT operations in parallel (N = concurrency setting)	100	615µs	596µs	660µs	881µs	524µs	1.9ms	1.6K	61.5ms	🟢 Excellent	~ 615µs prior
PASS	concurrent_select	Concurrent	N concurrent SELECT operations in parallel (N = concurrency setting)	100	923µs	916µs	1.0ms	1.1ms	793µs	1.4ms	1.1K	92.3ms	🟢 Excellent	↑5% faster
PASS	point_lookup	Select	SELECT by primary key from a 10K-row table (single row lookup)	100	436µs	414µs	553µs	576µs	371µs	596µs	2.3K	43.6ms	🟢 Excellent	↑9% faster
PASS	aggregate_query	Select	GROUP BY + SUM/AVG/COUNT on a 10K-row table (analytical query performance)	100	37.8ms	37.3ms	40.0ms	54.2ms	34.8ms	70.0ms	26	3.78s	🔴 Slow	~ 36665µs prior
PASS	multi_table_join	Select	SELECT with subquery across two tables (200 customers, 1000 orders)	100	4.0ms	3.9ms	4.6ms	4.8ms	3.7ms	4.9ms	251	399.0ms	🟢 Excellent	↓6% slower
PASS	large_payload_insert	Insert	INSERT rows with ~4KB TEXT payloads (serialization + storage throughput)	100	893µs	868µs	1.1ms	1.2ms	810µs	1.2ms	1.1K	89.3ms	🟢 Excellent	↓6% slower
PASS	wide_column_insert	Insert	INSERT into a 20-column table (wide schema overhead)	100	639µs	609µs	803µs	848µs	566µs	978µs	1.6K	63.9ms	🟢 Excellent	↓8% slower
PASS	bulk_delete	Delete	DELETE 100 rows at once with a range filter (bulk deletion)	100	176.3ms	173.8ms	193.3ms	194.2ms	151.7ms	194.9ms	6	17.63s	🟡 Acceptable	↓4% slower
PASS	sequential_crud	DML	INSERT → UPDATE → SELECT → DELETE full DML lifecycle per iteration	100	1.9ms	1.8ms	2.4ms	2.8ms	1.6ms	3.0ms	518	193.1ms	🟢 Excellent	~ 1907µs prior
PASS	alter_table	DDL	ALTER TABLE ADD COLUMN + DROP COLUMN (schema evolution latency)	100	2.0ms	1.8ms	2.3ms	2.8ms	1.5ms	14.5ms	510	196.1ms	🟢 Excellent	↑3% faster
PASS	concurrent_update	Concurrent	N concurrent UPDATE operations on the same table (write contention test)	100	2.1ms	2.0ms	2.8ms	3.3ms	1.1ms	5.7ms	487	205.4ms	🟢 Excellent	~ 2003µs prior
PASS	concurrent_mixed_dml	Concurrent	Concurrent INSERT + UPDATE + DELETE on the same table (multi-op contention)	100	3.8ms	3.8ms	4.7ms	4.8ms	2.7ms	4.9ms	262	381.3ms	🟢 Excellent	↓9% slower
PASS	namespace_isolation	Concurrent	Concurrent queries across 5 different namespaces (isolation test)	100	2.1ms	2.1ms	2.3ms	2.7ms	1.7ms	3.0ms	482	207.4ms	🟢 Excellent	↓6% slower
PASS	subscribe_initial_load	Subscribe	Subscribe to a 1000-row user table and receive the full initial data batch	100	10.8ms	10.5ms	14.8ms	15.6ms	9.4ms	15.7ms	93	1.08s	🟢 Excellent	↓10% slower
PASS	subscribe_change_latency	Subscribe	Latency from INSERT to subscriber receiving the change notification	100	5.02s	5.02s	5.03s	5.04s	5.01s	5.05s	0	501.81s	🔴 Slow	~ 5014269µs prior
PASS	reconnect_subscribe	Subscribe	Disconnect and re-subscribe to a user table (reconnection overhead)	100	16.4ms	15.9ms	20.5ms	22.4ms	13.6ms	22.6ms	61	1.64s	🟢 Excellent	↓23% slower
FAIL	flushed_parquet_query Warmup failed: Expected >= 200000 rows, got 0	Storage	SELECT from a shared table with 20 flushed Parquet files (200K rows)	0	0µs	0µs	0µs	0µs	0µs	0µs	0	0µs	⛔ Failed	—
PASS	concurrent_subscribers	Load	N WebSocket live-query subscribers receiving changes from concurrent writes	100	1.52s	1.52s	1.53s	1.54s	1.50s	1.54s	1	152.16s	🟢 Excellent	~ 1510244µs prior
PASS	concurrent_publishers	Load	N concurrent INSERTs into a topic-sourced table (measures publish overhead)	100	759µs	756µs	863µs	929µs	588µs	961µs	1.3K	75.9ms	🟢 Excellent	↑4% faster
PASS	concurrent_consumers	Load	N concurrent topic CONSUME calls pulling messages in parallel	100	628µs	626µs	689µs	723µs	552µs	760µs	1.6K	62.8ms	🟢 Excellent	↓7% slower
PASS	sql_1k_concurrent	Load	1000 concurrent SQL SELECT queries at once (RPS degradation test)	100	261.7ms	249.1ms	325.4ms	372.6ms	226.5ms	474.8ms	4	26.17s	🟢 Excellent	↓24% slower
PASS	create_user	Load	CREATE USER (auth subsystem stress test)	100	1.7ms	1.7ms	2.0ms	2.2ms	1.5ms	2.3ms	585	170.9ms	🟢 Excellent	↓6% slower
PASS	drop_user	Load	DROP USER (auth subsystem teardown stress test)	100	523µs	500µs	753µs	908µs	311µs	1.3ms	1.9K	52.3ms	🟢 Excellent	↓14% slower
PASS	connection_storm	Load	N simultaneous login + SQL + cycles (connection setup overhead)	100	4.8ms	4.6ms	5.4ms	11.3ms	3.6ms	19.8ms	209	478.8ms	🟢 Excellent	↓7% slower
PASS	mixed_read_write	Load	50/50 concurrent reads + writes on same table (contention test)	100	3.0ms	3.0ms	4.0ms	4.5ms	1.7ms	5.7ms	330	302.8ms	🟢 Excellent	↓5% slower
PASS	wide_fanout_query	Load	N concurrent large-result-set SELECTs (serialization + memory pressure)	100	12.6ms	12.2ms	14.5ms	15.8ms	10.3ms	28.1ms	79	1.26s	🟢 Excellent	↓34% slower
FAIL	subscriber_scale Iteration 0 failed: Single WS target (ws://localhost:8080/v1/ws) is likely capped by local ephemeral ports near ~32K. Use --urls with multiple endpoints (example: --urls http://127.0.0.1:8080,http://127.0.0.2:8080,http://127.0.0.3:8080,http://127.0.0.4:8080), or set KALAMDB_ALLOW_SINGLE_WS_TARGET=1 to force this run.	Scale	Progressive subscriber scale test (up to --max-subscribers, default 100K)	0	0µs	0µs	0µs	0µs	0µs	0µs	0	0µs	⛔ Failed	—