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
benchmarks: D9 calls the strategy arms "the same call with a different row"; two do more store work #1139
Found by round 5's code lens of the benchmark rig (#1131, docs/rewrite/benchmarks.md), recorded as "should be fixed next, as a clarifying word". It blocks nothing: the sentence is true of the rig, and the rig calls no strategy.
What
docs/rewrite/benchmarks.md:283-288 says the arms that re-enter create — delete, gift wrap, DVM job request, group, report, timestamp and vanish, beside the default arm — "are the same call with a different row, so a corpus using them measures what the create shape measures".
That is true of the create call inside each arm (delete.rs:105, gift_wrap.rs:38, dvm_job_request.rs:30, group.rs:37, report.rs:35, timestamp.rs:40, vanish.rs:37, default.rs:36) and true of the rig, which times EventRepository::create alone (crates/nostream-storage/examples/bench.rs:1005) and runs no strategy. It is not true of the arms as wholes: delete calls delete_by_addresses before the append (delete.rs:96-101), and vanish calls delete_by_pubkey_except_kinds before it and users.set_vanished after (vanish.rs:30-43). A reader planning a corpus change from this sentence would price a delete or vanish event at the create row's cost.
Acceptance
The sentence is scoped to what it can support — the create call, not the arm — and names the two arms that do more store work in the relay (delete, vanish) as work the rig does not time.
The claim stays true if D9 is later extended to time an arm end to end, or says which arms such an extension would measure differently.
Activity
feat(storage): the benchmark rig, and the first LMDB recording (#1141)