Later: Not on the v1 list: the deploy_pull pruning landed and works, and proving a 5,000-row batch holds the write lock under 50 ms is a performance guarantee for after v1.0.0.
Left open on #12009. Done looks like: a test seeds 5,000 prunable deploy_pull rows, runs one pruneDeploys batch from src/system/retention_deploys.ts, and asserts the transaction holds the write lock under 50 ms. If it does not, the batch cap is lowered until it does.
Later: Not on the v1 list: the deploy_pull pruning landed and works, and proving a 5,000-row batch holds the write lock under 50 ms is a performance guarantee for after v1.0.0.
Left open on #12009. Done looks like: a test seeds 5,000 prunable deploy_pull rows, runs one
pruneDeploysbatch fromsrc/system/retention_deploys.ts, and asserts the transaction holds the write lock under 50 ms. If it does not, the batch cap is lowered until it does.