You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Recursive reads 9-20% slower since #304 (runaway-query guards): ?hot_reach(X,Y) 248 -> 301 ms at 100k edges #384
Found by the #378 bisect on the benchmark host. Likely the per-step deadline/memory checks inside recursion. Make the check cheaper (check every N steps) without weakening the guard; add the case to the perf gate.
Fixed by https://github.com/inputlayer/inputlayer-symbolic/pull/10 (merged as 19435541): freed memory is returned once large computations go quiet instead of after every one; ?hot_reach(X, Y) at 100K edges 229 ms vs 277 ms on main (-17%), perf gate passes (ratio 0.960), and recursive_query is now a gated perf-gate fixture.
Activity
perf(execution): return freed memory once large computations go quiet…
perf(execution): return freed memory once large computations go quiet…
jsam commentedon Oct 6, 2026
Fixed by https://github.com/inputlayer/inputlayer-symbolic/pull/10 (merged as 19435541): freed memory is returned once large computations go quiet instead of after every one; ?hot_reach(X, Y) at 100K edges 229 ms vs 277 ms on main (-17%), perf gate passes (ratio 0.960), and recursive_query is now a gated perf-gate fixture.