Compiles the proofs may refuse for one key before it stops trying.
One unlucky pair of renders can refuse (an H:i timestamp crossing a minute, a queue count, a
message varies and is not a slot), so a single refusal must not latch for the generation:
/admin/content on wrangler dev is 12.6-17.9 req/s unserved against 221-400 req/s served.
It stays bounded, since a page that varies every render would otherwise spend a diff every third
render forever. A spent key tries again after PLAN_TTL_MS: a re-proof drops a plan whose
page ticks (a "Member for" line) and each recompile can refuse, so a permanent latch would
eventually stop a page that holds a plan most of the time.
Compiles the proofs may refuse for one key before it stops trying.
One unlucky pair of renders can refuse (an
H:itimestamp crossing a minute, a queue count, a message varies and is not a slot), so a single refusal must not latch for the generation:/admin/contentonwrangler devis 12.6-17.9 req/s unserved against 221-400 req/s served.It stays bounded, since a page that varies every render would otherwise spend a diff every third render forever. A spent key tries again after PLAN_TTL_MS: a re-proof drops a plan whose page ticks (a "Member for" line) and each recompile can refuse, so a permanent latch would eventually stop a page that holds a plan most of the time.