Comparison — 0.16.0-rc1 vs prior rounds & the goals ladder

0.16.0-rc1 targeted fix-validation · flight 0.16.0-rc1@sha256:79c7907d (multi-arch INDEX, src f4f13a7cb) · connector 0.16.0-rc1 · Trino 481 · Cassandra 5.0.8 RF=3 · 3× i4i.xlarge · ~1.70M partitions/node, 2 SSTable gens · 2026-07-21

Throughput / latency, round over round

Round	Threads	qps	p50	p99	errors	Notes
R11b / R12	8	~33–34	~242ms	~363ms	0	8-thread warm baselines
soak-0.15	32	~39	798ms	1366ms	0	90-min steady block
0.16.0-rc1	8	~35	226ms	328ms	0	comparability block (parity+ vs R12)
0.16.0-rc1	32	~32	990ms	1255ms	0	3.5h main soak (flat, 0 err)

The 8-thread block reproduces the R12 baseline closely (parity+). The 32-thread numbers sit in the same band as soak-0.15; the ~32–39 qps ceiling is unchanged because #2679 (which would have relieved keyed reads) is inert (#2806).

Lane-B goals ladder

Goal	Target	0.16.0-rc1	Status
B1 warm point-read latency	≤300ms	226ms @ 8-thr (floor-adj); ~990ms @ 32-thr	partial (thread-dependent)
B2 throughput @ 32 threads	≥100 qps	~32 qps	not met (known ceiling)
Zero client errors under load	0	0 / 402,932 queries	met
Integrity (no silent loss)	0 delta	1,703,038 == 1,703,038	met
Memory bounded (no leak)	flat	RSS flat 3.5h, 0 OOM	met
Genuine-error SLI	low	0.026% (internal)	met (down from 0.89%)

Targeted-fix scorecard (this round's purpose)

Fix	Expected	Observed	Verdict
#2679 split pruning	keyed read → 1 split / 1 pod	13 splits / 3 pods (inert)	regressed → #2806
#2349 UDT both paths	cold==warm==sstabledump	DDL parse fails first	blocked → #2807
#2372 BTI over Flight	rows returned, not errored	no da corpus	could-not-observe
#2681 abort taxonomy	reasons bucketed	client_cancel/admission_shed/internal correct	verified