示例

Check a resource out of a NimblePool and survive the failure paths: what the checkout function must return, which failures the pool recovers from, and which one silently leaks a worker

sha256:b9a321085b5992dc0c641045e8b697e67dbb23c7f9958ea2d4e0b145d1a2f486

案例

目标

Check a resource out of a NimblePool and survive the failure paths: what the checkout function must return, which failures the pool recovers from, and which one silently leaks a worker HOW

包

nimble_pool 1.1.0

环境

elixir 1

创建时间

2026-08-14T20:12:13Z

契约

• assert :pool_size defaults to 10 rather than System.schedulers_online(), by counting exactly ten init_worker callbacks at start_link and no eleventh
• assert lazy: true starts no workers at all and builds exactly one on the first checkout
• assert the checkout! function body runs in the CLIENT process, and that the from it receives is {pool_pid, ref} carrying the POOL's pid, unlike GenServer.handle_call/3 where from carries the caller's
• assert checkout! returns the first element of the two-tuple and only handle_checkin/4 ever sees the second
• assert the asynchronous check-in does not race a re-checkout: twenty-five sequential checkouts against a one-worker pool all succeed
• assert raising inside the checkout! function is the recoverable failure: handle_cancelled(:checked_out) then terminate_worker(:error, ...) then a replacement init_worker, with handle_checkin skipped
• assert returning a value that is not a two-tuple raises {:try_clause, value} and silently leaks the worker: no handle_cancelled, no terminate_worker, no replacement, and every later checkout on that pool exits :timeout
• assert the leaked worker is reclaimed only when the calling process itself dies, which is when the pool's monitor fires handle_cancelled(:checked_out) and terminate_worker(:DOWN, ...)
• assert {:skip, exception, pool_state} from handle_checkout/4 raises that exception in the client while leaving the worker checked in and serving the next call, terminating nothing
• assert {:remove, reason, pool_state} from handle_checkout/4 retries the same checkout against another worker, so a remove keyed on the command destroys and rebuilds more workers than the pool size and the caller only ever sees its own checkout timeout
• assert a caller that times out while still queued cancels as :queued and tears down nothing, while a caller killed holding a worker cancels as :checked_out with terminate_worker reason :DOWN
• assert NimblePool.update/2 is called from inside the checkout! function with from, routes to handle_update/3, and replaces the worker_state that handle_checkin/4 later receives while the client_state passes through unchanged

文件

• csx.json
• lib/csx_nimble_pool/churn_worker.ex
• lib/csx_nimble_pool/instrumented_worker.ex
• mix.exs
• mix.lock
• test/contract_test.exs
• test/test_helper.exs

下载已验证的构件 (tar.gz) — 契约实际运行的那些字节

原始种子者

anonymous

验证回执

• elixir 1 · CONTAINER_RUN · compile:PASS · contract:PASS · load:PASS · resolve:PASS · hex@1 · 2026-08-14 · ed25519:a2ec939a4c60e243