profiler_load

Overview ¶

profiler_load is a high-load stress test and heap generator for bubblepprof. It spawns a configurable pool of goroutines — each stamped with standard runtime/pprof labels — and keeps a configurable amount of heap alive so /debug/memusage has real attributed bytes to report per bubble.

Worker types (distributed round-robin across -workers goroutines):

• role=cpu-hash, pool=compute — SHA256-hashes 32 KiB buffers in a tight loop
• role=sorter, pool=compute — sorts a 4096-element int slice each iteration
• role=allocator, pool=memory — allocates 8–63 KiB chunks, retains the last four
• role=heap-scan, pool=memory — strides through the pinned resident heap array
• role=channel-producer, pool=pipeline — pushes 128-byte jobs into a buffered channel
• role=mutex-and-channel-consumer, pool=pipeline — consumes jobs and contends on a shared map mutex

Each worker also carries a shard label (worker_id % 32). A single reporter goroutine (role=reporter, pool=monitor) logs goroutine count, heap stats, and ops/sec every two seconds.

Querying bubbles while the load test runs:

# All compute workers (CPU-hash + sorter)
curl -X POST http://127.0.0.1:6060/debug/memusage \
  -H 'Content-Type: application/json' \
  -d '{"labels":{"pool":"compute"}}'

# Allocator workers only (dynamic heap churn)
curl -X POST http://127.0.0.1:6060/debug/memusage \
  -H 'Content-Type: application/json' \
  -d '{"labels":{"role":"allocator"}}'

# Heap-scan workers (large resident heap visible from matched goroutines)
curl -X POST http://127.0.0.1:6060/debug/memusage \
  -H 'Content-Type: application/json' \
  -d '{"labels":{"role":"heap-scan"}}'

Usage:

go run ./examples/profiler_load
go run ./examples/profiler_load -mem-mb 500 -workers 1024 -duration 2m

Flags:

-addr      HTTP listen address for /debug/memusage (default 127.0.0.1:6060)
-workers   number of labeled worker goroutines (default 768)
-mem-mb    resident heap to keep alive in MiB (default 160; use 500+ for heavier loads)
-duration  how long to run; 0 means until Ctrl+C/SIGTERM
-queue     channel buffer size between producers and consumers (default 8192)