bunqueue v2.1.0

Name: bunqueue
Author: egeominotti

Fixes 4 bugs and ships 3 performance improvements. Hello command, protocol version 2, semaphore utility, fair benchmark comparison and more.

bun add bunqueue@2.1.0

npm install bunqueue@2.1.0

bun add bunqueue@latest

Hello command

For protocol version negotiation (cmd: 'Hello')

With pipelining capability support

For server-side concurrency limiting (src/shared/semaphore.ts)

(bench/comparison/run.ts):

(bench/comprehensive.ts):

Added pingInterval, commandTimeout, pipelining, maxInFlight to public API

Comprehensive pipelining test suites:
- test/protocol-reqid.test.ts - 7 tests for reqId handling
- test/client-pipelining.test.ts - 7 tests for client pipelining
- test/server-pipelining.test.ts - 7 tests for server parallel processing
- test/backward-compat.test.ts - 10 tests for backward compatibility

TCP Pipelining - Major throughput improvement for TCP client operations:
- Client-side: Multiple commands in flight per connection (up to 100 by default)
- Server-side: Parallel command processing with Promise.all()
- reqId-based response matching for correct command-response pairing
- 125,000 ops/sec in pipelining benchmarks (vs ~20,000 before)
- Configurable via pipelining: boolean and maxInFlight: number options
SQLite indexes for high-throughput operations - Added 4 new indexes for 30-50% faster queries:
- idx_jobs_state_started: Stall detection now O(log n) instead of O(n) table scan
- idx_jobs_group_id: Fast lookup for group operations
- idx_jobs_pending_priority: Compound index for priority-ordered job retrieval
- idx_dlq_entered_at: DLQ expiration cleanup now O(log n)
Date.now() caching in pull loop - Reduced syscalls by caching timestamp per iteration (+3-5% throughput)

Schema version bumped to 5 (auto-migrates existing databases)
TCP client now includes reqId in all commands for response matching
Server processes multiple frames in parallel (max 50 concurrent per connection)
Documentation: Rewrote comparison page with real benchmark data and methodology explanation

SQLITE_BUSY under high concurrency - Added PRAGMA busy_timeout = 5000 to wait for locks instead of failing immediately
“Database has closed” errors during shutdown - Added stopped flag to WriteBuffer to prevent flush attempts after stop()
Critical: Worker pendingJobs race condition - Concurrent tryProcess() calls could overwrite each other’s job buffers, causing ~30% job loss under high concurrency. Now preserves existing buffered jobs when pulling new batches.
Connection options not passed through - Worker, Queue, and FlowProducer now correctly pass pingInterval, commandTimeout, pipelining, and maxInFlight options to the TCP connection pool.