Michel de Graaf
5b4ef30f08
The client had never been driven at scale. A stress harness that mixes searches, downloads, uploads and browses against a local soulfind found four ceilings, three of which lose work silently rather than merely being slow. Peers were capped at one per core. ActorSystem ran each actor's whole lifetime as one job on a fixed-size ThreadPool, and actor loops block until stopped, so the pool acted as a semaphore: past the cap an actor never started, never connected and never errored, and uploads stopped working entirely. The pool had exactly one consumer and every job it received was non-terminating, so it is gone; actors get their own thread. TransferResponse raced the bookkeeping it depends on. It was sent straight back to the peer while UpdateDownloadTokens was still queued for the client loop — but the response is what invites the peer's file connection, and that connection is matched by token. Under load the client loop lost, the file connection matched nothing, and the download sat on Queued forever. The response now goes out after the token is recorded. Download tokens were md5(filename) truncated to 20 bits, and the store deletes every entry matching a token, so two downloads sharing one destroyed each other — which fetching the same filename from two peers does every time. They now come from a counter. Port 0 from GetPeerAddress means the server does not know where a user listens. It was cached and dialled, and the upload waiting on it was dropped. It is no longer cached or served against; the connect attempt stays, because its failure is what makes an unreachable peer fall back to server brokering. Mixed profile at 1024 peers: 512/512 downloads, 256/256 uploads, 400-850 MiB/s in ~7s, against 71/96 downloads at 7.9 MiB/s and a 60s timeout before.
Concurrency benchmark
Answers one question: how much simultaneous Soulseek activity can the client absorb before it starts losing work?
scripts/stress.sh # run the default profile
scripts/stress.sh --save # ...and store it as the new baseline if it wins
A run takes about 7 seconds. It spawns a local soulfind, surrounds one real
Client with a swarm of mock peers, and drives searches, downloads, uploads and
browses at the same time — the mixing is the point, since every ceiling found
so far only appeared under a mixed load.
The score
One number out of 100, weighted across five dimensions:
| Dimension | Weight | Measures |
|---|---|---|
| downloads | 0.30 | completed / started |
| uploads | 0.20 | served / requested |
| searches | 0.20 | queries that got at least one result |
| browses | 0.15 | browse requests answered |
| throughput | 0.15 | aggregate MiB/s against TARGET_MBPS (500) |
The four functional dimensions should stay at 100%: anything less means work was silently dropped, which is the failure mode that matters. Throughput is the term with headroom, so the score still moves when transfer handling regresses.
--save only overwrites stress-baseline.json when the run beats it, so the
baseline ratchets upward. Without --save a run just prints the delta.
Tuning the load
Every knob is an env var; the defaults are in Config::from_env.
STRESS_PEERS=2048 STRESS_FILE_KB=8192 scripts/stress.sh
STRESS_PEERS STRESS_SEARCHES STRESS_DOWNLOADS STRESS_LEECHERS
STRESS_BROWSES STRESS_FILE_KB STRESS_TIMEOUT
Reading a failure
downloads completed 480/512 is the headline, but the two lines under it say
where to look:
download states— a straggler onQueuednever got its transfer negotiated; one onInProgressis merely slow. These are different bugs.mocks never online— mock peers the harness could not bring up. That is a harness/server limit, not the client, and it is reported separately so it cannot quietly shrink a denominator and flatter the score.
peak threads counts the whole process — client and mock swarm — so treat it
as a trend, not a client measurement.
LOG_LEVEL=debug scripts/stress.sh correlates a stuck peer's name against the
client's own trace, which is how each ceiling below was actually found.
Ceilings found and removed
| Symptom | Cause |
|---|---|
| Hard stop at ~15 peers; later peers never start | Every PeerActor held a ThreadPool worker for life, so the pool capped peers at one per core |
Downloads stuck on Queued under load |
TransferResponse was sent before the client recorded the peer's token, so the file connection raced its own bookkeeping |
| One download vanishing per few hundred | Token was md5(filename), and the store deletes every entry with a token — same filename from two peers destroyed both |
| Uploads dropped after the peer accepted | Server answered GetPeerAddress with port 0; the client cached it and dialled it |
Requirements
A soulfind binary (SOULFIND_BIN, defaulted in scripts/stress.sh), or
SOULSEEK_TEST_SERVER=host:port to point at a running server. The script raises
the descriptor limit — a swarm holds a lot of sockets open.