fix: Stop circuit breaker spawning a thread per call - #1
Open
Ashish-CodeJourney wants to merge 9 commits into
Open
fix: Stop circuit breaker spawning a thread per call#1Ashish-CodeJourney wants to merge 9 commits into
Ashish-CodeJourney wants to merge 9 commits into
Conversation
CircuitBreaker creates a new single-thread executor on every call, so each invocation spawns a fresh thread. The test drives 50 sequential calls through one breaker and records the worker thread used by each. Currently fails: expected 1 thread, was 50.
The previous assertion (exactly one thread) over-specified: a pooled executor hands work off through a SynchronousQueue, so a returning worker may not be parked yet when the next call is submitted, and a correct implementation can legitimately use a second thread. Assert the behaviour that matters instead: thread use stays within a configured concurrency limit no matter how many calls are made. Adds the maxConcurrentCalls parameter the test needs; it is not yet wired up, so the assertion rather than the compiler reports the defect. Currently fails: used 50 threads for 50 calls.
withCircuitBreaker created a new single-thread executor on every invocation, so every request through ProjectClient spawned a fresh thread. The executor was never shut down; on JDK 21+ a Cleaner reclaims it once unreachable, but only after a GC, so threads pile up under load. On the timeout path the abandoned pool was left holding a task that cancel(true) only interrupts. Use one bounded pool per breaker instead. Idle workers are reused and time out after 60s, and maxConcurrentCalls caps thread growth when a downstream stalls: once the pool is saturated, submit is rejected, the existing catch treats it as a failure and the fallback runs. 5 tests passing.
Pool threads inherit daemon status from whichever thread happens to submit first, so a worker can outlive the work it was created for and hold the JVM open. Each server also starts a discovery heartbeat scheduler that is never shut down, so a clean exit already depends on no other non-daemon threads lingering. Currently fails: worker thread is not a daemon.
Pool workers no longer inherit daemon status from the submitting thread, so a lingering worker cannot hold the JVM open. Naming them circuit-breaker-N also makes them identifiable in a thread dump. 6 tests passing.
Mutation testing surfaced a false negative: the thread-bound test counted distinct thread names, so a mutant that gave every worker a constant name masked a mutant that restored the original pool-per-call bug. Both applied together left the suite green. Collect Thread instances instead. The pair is now caught.
Mutation testing showed the pool's maximum size was unverified: widening it to Int.MAX_VALUE left the suite green, because the existing test issues calls sequentially and never saturates the pool. Drive eight overlapping calls through a breaker limited to two and assert only two workers are used, so the bulkhead that protects a server from a stalled downstream is actually covered.
Mutation testing left cancel(true) -> cancel(false) alive. That matters more now the pool is shared and bounded: an uninterrupted task keeps its worker busy after the caller has given up, so repeated timeouts would consume the concurrency limit and reject healthy calls. Assert the abandoned work is interrupted once the breaker falls back.
Split the constructor across lines, name the idle worker timeout, and drop the property qualifier from maxConcurrentCalls now that only the pool it configures reads it. Refactor only, no behaviour change.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
withCircuitBreakercalledExecutors.newSingleThreadExecutor()on every invocation, soevery request through
ProjectClient(allocations, backlog, timesheets) spawned a thread.Nothing shut the executor down. JDK 21+ reclaims it via a Cleaner once unreachable, but
only after a GC, so threads pile up under load.
Now one bounded pool per breaker. Workers are reused, idle out after 60s, and run as
daemon threads so a straggler cannot hold the JVM open. New
maxConcurrentCallsparam(default 10, last position, so existing call sites are unaffected) caps the pool: when it
saturates,
submitis rejected, the existing catch counts it as a failure and thefallback runs.
Tests: 8 passing in
components:circuit-breaker-support, ran 6 times since the two newconcurrency tests are timing sensitive. Other modules need local MySQL and Redis, so only
this one was run.
Mutation testing by hand (no pitest here) caught a false negative worth flagging: the
first test counted thread names, so a constant-name mutant hid a mutant restoring the
original bug. It counts
Threadinstances now.Follow up, not in this PR:
testMaxFailurescannot fail, becausefail()runs on theworker thread and its
AssertionErrorgets swallowed by the breaker'scatch (e: Exception). That leaves the failure counting boundaries effectively untested.