vole
Clone
Read only, no credentials:
https://rodent.software/vole.git
Write over SSH with a registered key:
git clone ssh://git@rodent.software:2222/vole.git
A free and open-source forge that lives in your Git repo.
Phase 0: measure metadata writes, git transport, and tree serving
c5626f1b87084870524ea54987ef47a65ea4740b · Keenan Nemetz · 18d ago · parent 17311880
Phase 0: measure metadata writes, git transport, and tree serving Three throwaway spikes under docs/spikes, each with its own harness, README, and reproduce command. Findings are folded into the roadmap and its decision log. - ref-cas: sharded + batched refs sustain 25,637 entity writes/s at 20k entities with zero lost writes; a single unsharded ref manages 30/s with 22 CAS retries per write and decays quadratically as it fills. Metadata refs must be sharded. Unsharded is now off the table. - smart-http: driving git http-backend as a subprocess is correct (hook stderr reaches the client, aborts leave no ref and no orphaned children) and pushes 150 MiB/s on compressible data, but a 1 GiB push peaks at 1.13 GB of git-child RSS. Phase 1 needs a push size cap and a concurrency semaphore, plus chunked-upload spooling. - tree-serving: sites can be served straight from git trees with no export step - 1.2 ms p50 cold, 0.075 ms warm, 170k req/s behind an 18.5 MiB cache. Harness bugs caught during the work are recorded in the spike READMEs, including one that silently dropped 83% of writes while reporting plausible throughput.
docs/spikes/ref-cas/main.go +585
| @@ -0,0 +1,585 @@ | |||
| 1 | // Spike A: measure the throughput of writing forge metadata as commits on | ||
| 2 | // sidecar refs, using git compare-and-swap ref updates. | ||
| 3 | // | ||
| 4 | // Every mode stores entity files in a tree, one commit per write (or per | ||
| 5 | // batch), and moves the ref with `git update-ref <ref> <new> <old>` so a | ||
| 6 | // lost race is detected rather than clobbering. | ||
| 7 | package main | ||
| 8 | |||
| 9 | import ( | ||
| 10 | "bytes" | ||
| 11 | "encoding/json" | ||
| 12 | "flag" | ||
| 13 | "fmt" | ||
| 14 | "math/rand" | ||
| 15 | "os" | ||
| 16 | "os/exec" | ||
| 17 | "path/filepath" | ||
| 18 | "sort" | ||
| 19 | "strings" | ||
| 20 | "sync" | ||
| 21 | "sync/atomic" | ||
| 22 | "time" | ||
| 23 | ) | ||
| 24 | |||
| 25 | const zeroSHA = "0000000000000000000000000000000000000000" | ||
| 26 | |||
| 27 | type repo struct { | ||
| 28 | dir string | ||
| 29 | env []string | ||
| 30 | } | ||
| 31 | |||
| 32 | func newRepo(dir string) (*repo, error) { | ||
| 33 | if err := os.RemoveAll(dir); err != nil { | ||
| 34 | return nil, err | ||
| 35 | } | ||
| 36 | cmd := exec.Command("git", "init", "--bare", "-q", dir) | ||
| 37 | if out, err := cmd.CombinedOutput(); err != nil { | ||
| 38 | return nil, fmt.Errorf("git init: %w: %s", err, out) | ||
| 39 | } | ||
| 40 | return &repo{dir: dir, env: append(os.Environ(), | ||
| 41 | "GIT_DIR="+dir, | ||
| 42 | "GIT_AUTHOR_NAME=spike", "GIT_AUTHOR_EMAIL=spike@example.com", | ||
| 43 | "GIT_COMMITTER_NAME=spike", "GIT_COMMITTER_EMAIL=spike@example.com", | ||
| 44 | )}, nil | ||
| 45 | } | ||
| 46 | |||
| 47 | func (r *repo) run(stdin []byte, args ...string) ([]byte, error) { | ||
| 48 | cmd := exec.Command("git", args...) | ||
| 49 | cmd.Env = r.env | ||
| 50 | if stdin != nil { | ||
| 51 | cmd.Stdin = bytes.NewReader(stdin) | ||
| 52 | } | ||
| 53 | var out, errb bytes.Buffer | ||
| 54 | cmd.Stdout = &out | ||
| 55 | cmd.Stderr = &errb | ||
| 56 | if err := cmd.Run(); err != nil { | ||
| 57 | return out.Bytes(), fmt.Errorf("%w: %s", err, strings.TrimSpace(errb.String())) | ||
| 58 | } | ||
| 59 | return out.Bytes(), nil | ||
| 60 | } | ||
| 61 | |||
| 62 | // revParse returns the commit the ref points at, or "" if it does not exist. | ||
| 63 | func (r *repo) revParse(ref string) (string, error) { | ||
| 64 | out, err := r.run(nil, "rev-parse", "--verify", "--quiet", ref+"^{commit}") | ||
| 65 | if err != nil { | ||
| 66 | return "", nil // missing ref is not an error for us | ||
| 67 | } | ||
| 68 | return strings.TrimSpace(string(out)), nil | ||
| 69 | } | ||
| 70 | |||
| 71 | type entry struct { | ||
| 72 | mode, typ, sha, name string | ||
| 73 | } | ||
| 74 | |||
| 75 | func (r *repo) treeEntries(commit string) ([]entry, error) { | ||
| 76 | if commit == "" { | ||
| 77 | return nil, nil | ||
| 78 | } | ||
| 79 | out, err := r.run(nil, "ls-tree", "-z", commit) | ||
| 80 | if err != nil { | ||
| 81 | return nil, err | ||
| 82 | } | ||
| 83 | var es []entry | ||
| 84 | for _, rec := range bytes.Split(out, []byte{0}) { | ||
| 85 | if len(rec) == 0 { | ||
| 86 | continue | ||
| 87 | } | ||
| 88 | // <mode> SP <type> SP <sha> TAB <name> | ||
| 89 | tab := bytes.IndexByte(rec, '\t') | ||
| 90 | if tab < 0 { | ||
| 91 | return nil, fmt.Errorf("bad ls-tree record %q", rec) | ||
| 92 | } | ||
| 93 | f := strings.Fields(string(rec[:tab])) | ||
| 94 | if len(f) != 3 { | ||
| 95 | return nil, fmt.Errorf("bad ls-tree fields %q", rec) | ||
| 96 | } | ||
| 97 | es = append(es, entry{f[0], f[1], f[2], string(rec[tab+1:])}) | ||
| 98 | } | ||
| 99 | return es, nil | ||
| 100 | } | ||
| 101 | |||
| 102 | func (r *repo) hashBlob(payload []byte) (string, error) { | ||
| 103 | out, err := r.run(payload, "hash-object", "-w", "--stdin") | ||
| 104 | if err != nil { | ||
| 105 | return "", err | ||
| 106 | } | ||
| 107 | return strings.TrimSpace(string(out)), nil | ||
| 108 | } | ||
| 109 | |||
| 110 | func (r *repo) hashBlobs(paths []string) ([]string, error) { | ||
| 111 | in := []byte(strings.Join(paths, "\n") + "\n") | ||
| 112 | out, err := r.run(in, "hash-object", "-w", "--stdin-paths") | ||
| 113 | if err != nil { | ||
| 114 | return nil, err | ||
| 115 | } | ||
| 116 | return strings.Fields(string(out)), nil | ||
| 117 | } | ||
| 118 | |||
| 119 | func (r *repo) mkTree(es []entry) (string, error) { | ||
| 120 | sort.Slice(es, func(i, j int) bool { return es[i].name < es[j].name }) | ||
| 121 | var buf bytes.Buffer | ||
| 122 | for _, e := range es { | ||
| 123 | fmt.Fprintf(&buf, "%s %s %s\t%s\x00", e.mode, e.typ, e.sha, e.name) | ||
| 124 | } | ||
| 125 | out, err := r.run(buf.Bytes(), "mktree", "-z") | ||
| 126 | if err != nil { | ||
| 127 | return "", err | ||
| 128 | } | ||
| 129 | return strings.TrimSpace(string(out)), nil | ||
| 130 | } | ||
| 131 | |||
| 132 | func (r *repo) commitTree(tree, parent, msg string) (string, error) { | ||
| 133 | args := []string{"commit-tree", tree} | ||
| 134 | if parent != "" { | ||
| 135 | args = append(args, "-p", parent) | ||
| 136 | } | ||
| 137 | out, err := r.run([]byte(msg), args...) | ||
| 138 | if err != nil { | ||
| 139 | return "", err | ||
| 140 | } | ||
| 141 | return strings.TrimSpace(string(out)), nil | ||
| 142 | } | ||
| 143 | |||
| 144 | // casRef moves ref from old to new. Returns false if another writer won. | ||
| 145 | func (r *repo) casRef(ref, new, old string) (bool, error) { | ||
| 146 | if old == "" { | ||
| 147 | old = zeroSHA | ||
| 148 | } | ||
| 149 | _, err := r.run(nil, "update-ref", ref, new, old) | ||
| 150 | if err == nil { | ||
| 151 | return true, nil | ||
| 152 | } | ||
| 153 | if strings.Contains(err.Error(), "cannot lock ref") || | ||
| 154 | strings.Contains(err.Error(), "is at") || | ||
| 155 | strings.Contains(err.Error(), "unable to resolve") { | ||
| 156 | return false, nil | ||
| 157 | } | ||
| 158 | return false, err | ||
| 159 | } | ||
| 160 | |||
| 161 | type stats struct { | ||
| 162 | mode string | ||
| 163 | commits int64 | ||
| 164 | entities int64 | ||
| 165 | elapsed time.Duration | ||
| 166 | retries int64 | ||
| 167 | failures int64 | ||
| 168 | latencyNanos []int64 | ||
| 169 | } | ||
| 170 | |||
| 171 | func (s *stats) report() map[string]any { | ||
| 172 | sort.Slice(s.latencyNanos, func(i, j int) bool { return s.latencyNanos[i] < s.latencyNanos[j] }) | ||
| 173 | p := func(q float64) float64 { | ||
| 174 | if len(s.latencyNanos) == 0 { | ||
| 175 | return 0 | ||
| 176 | } | ||
| 177 | i := int(q * float64(len(s.latencyNanos)-1)) | ||
| 178 | return float64(s.latencyNanos[i]) / 1e6 | ||
| 179 | } | ||
| 180 | return map[string]any{ | ||
| 181 | "mode": s.mode, | ||
| 182 | "entities": s.entities, | ||
| 183 | "commits": s.commits, | ||
| 184 | "elapsed_s": s.elapsed.Seconds(), | ||
| 185 | "entities_per_sec": float64(s.entities) / s.elapsed.Seconds(), | ||
| 186 | "commits_per_sec": float64(s.commits) / s.elapsed.Seconds(), | ||
| 187 | "cas_retries": s.retries, | ||
| 188 | "failures": s.failures, | ||
| 189 | "retries_per_entity": float64(s.retries) / float64(max64(s.entities, 1)), | ||
| 190 | "write_p50_ms": p(0.50), | ||
| 191 | "write_p95_ms": p(0.95), | ||
| 192 | "write_p99_ms": p(0.99), | ||
| 193 | } | ||
| 194 | } | ||
| 195 | |||
| 196 | func max64(a, b int64) int64 { | ||
| 197 | if a > b { | ||
| 198 | return a | ||
| 199 | } | ||
| 200 | return b | ||
| 201 | } | ||
| 202 | |||
| 203 | type entity struct { | ||
| 204 | id string | ||
| 205 | payload []byte | ||
| 206 | shard int | ||
| 207 | } | ||
| 208 | |||
| 209 | func refFor(shards int, shard int) string { | ||
| 210 | if shards <= 1 { | ||
| 211 | return "refs/vole/issues/0" | ||
| 212 | } | ||
| 213 | return fmt.Sprintf("refs/vole/issues/%03d", shard) | ||
| 214 | } | ||
| 215 | |||
| 216 | func shardOf(id string, shards int) int { | ||
| 217 | if shards <= 1 { | ||
| 218 | return 0 | ||
| 219 | } | ||
| 220 | // FNV-1a, matching the sharding scheme the roadmap proposes. | ||
| 221 | var h uint32 = 2166136261 | ||
| 222 | for i := 0; i < len(id); i++ { | ||
| 223 | h ^= uint32(id[i]) | ||
| 224 | h *= 16777619 | ||
| 225 | } | ||
| 226 | return int(h % uint32(shards)) | ||
| 227 | } | ||
| 228 | |||
| 229 | // writeOne is the honest per-entity path: build (or update) the shard tree, | ||
| 230 | // commit, and CAS the ref. Retries on a lost race. | ||
| 231 | func (r *repo) writeOne(e entity, shards int, maxRetries int, retries *int64, failures *int64) (time.Duration, bool) { | ||
| 232 | start := time.Now() | ||
| 233 | ref := refFor(shards, e.shard) | ||
| 234 | for attempt := 0; attempt <= maxRetries; attempt++ { | ||
| 235 | if attempt > 0 { | ||
| 236 | atomic.AddInt64(retries, 1) | ||
| 237 | time.Sleep(time.Duration(rand.Intn(200)+50) * time.Microsecond) | ||
| 238 | } | ||
| 239 | old, err := r.revParse(ref) | ||
| 240 | if err != nil { | ||
| 241 | atomic.AddInt64(failures, 1) | ||
| 242 | return time.Since(start), false | ||
| 243 | } | ||
| 244 | es, err := r.treeEntries(old) | ||
| 245 | if err != nil { | ||
| 246 | atomic.AddInt64(failures, 1) | ||
| 247 | return time.Since(start), false | ||
| 248 | } | ||
| 249 | blob, err := r.hashBlob(e.payload) | ||
| 250 | if err != nil { | ||
| 251 | atomic.AddInt64(failures, 1) | ||
| 252 | return time.Since(start), false | ||
| 253 | } | ||
| 254 | es = upsert(es, entry{"100644", "blob", blob, e.id + ".json"}) | ||
| 255 | tree, err := r.mkTree(es) | ||
| 256 | if err != nil { | ||
| 257 | atomic.AddInt64(failures, 1) | ||
| 258 | return time.Since(start), false | ||
| 259 | } | ||
| 260 | commit, err := r.commitTree(tree, old, "issue "+e.id) | ||
| 261 | if err != nil { | ||
| 262 | atomic.AddInt64(failures, 1) | ||
| 263 | return time.Since(start), false | ||
| 264 | } | ||
| 265 | ok, err := r.casRef(ref, commit, old) | ||
| 266 | if err != nil { | ||
| 267 | atomic.AddInt64(failures, 1) | ||
| 268 | return time.Since(start), false | ||
| 269 | } | ||
| 270 | if ok { | ||
| 271 | return time.Since(start), true | ||
| 272 | } | ||
| 273 | } | ||
| 274 | atomic.AddInt64(failures, 1) | ||
| 275 | return time.Since(start), false | ||
| 276 | } | ||
| 277 | |||
| 278 | func upsert(es []entry, add entry) []entry { | ||
| 279 | for i := range es { | ||
| 280 | if es[i].name == add.name { | ||
| 281 | es[i] = add | ||
| 282 | return es | ||
| 283 | } | ||
| 284 | } | ||
| 285 | return append(es, add) | ||
| 286 | } | ||
| 287 | |||
| 288 | // runNaive: N writers, no batching, one commit per entity. | ||
| 289 | func runNaive(r *repo, name string, shards, writers, n, maxRetries int, tmp string) *stats { | ||
| 290 | st := &stats{mode: name} | ||
| 291 | jobs := make(chan entity, 256) | ||
| 292 | var wg sync.WaitGroup | ||
| 293 | var mu sync.Mutex | ||
| 294 | start := time.Now() | ||
| 295 | for w := 0; w < writers; w++ { | ||
| 296 | wg.Add(1) | ||
| 297 | go func() { | ||
| 298 | defer wg.Done() | ||
| 299 | for e := range jobs { | ||
| 300 | d, ok := r.writeOne(e, shards, maxRetries, &st.retries, &st.failures) | ||
| 301 | mu.Lock() | ||
| 302 | st.latencyNanos = append(st.latencyNanos, d.Nanoseconds()) | ||
| 303 | mu.Unlock() | ||
| 304 | if ok { | ||
| 305 | atomic.AddInt64(&st.entities, 1) | ||
| 306 | atomic.AddInt64(&st.commits, 1) | ||
| 307 | } | ||
| 308 | } | ||
| 309 | }() | ||
| 310 | } | ||
| 311 | for i := 0; i < n; i++ { | ||
| 312 | id := fmt.Sprintf("%06d", i) | ||
| 313 | jobs <- entity{id: id, payload: payloadFor(id), shard: shardOf(id, shards)} | ||
| 314 | } | ||
| 315 | close(jobs) | ||
| 316 | wg.Wait() | ||
| 317 | st.elapsed = time.Since(start) | ||
| 318 | return st | ||
| 319 | } | ||
| 320 | |||
| 321 | // batcher commits a group of entities per shard in one commit. | ||
| 322 | func runBatched(r *repo, name string, shards, batchers, n, batchSize int, maxRetries int) *stats { | ||
| 323 | st := &stats{mode: name} | ||
| 324 | jobs := make(chan entity, 1024) | ||
| 325 | var wg sync.WaitGroup | ||
| 326 | var mu sync.Mutex | ||
| 327 | start := time.Now() | ||
| 328 | |||
| 329 | // A batcher must only ever consume work it owns. Routing happens once, in | ||
| 330 | // the dispatcher, so no batcher can swallow an entity belonging to a peer | ||
| 331 | // that has already exited. | ||
| 332 | owned := make([]chan entity, batchers) | ||
| 333 | for i := range owned { | ||
| 334 | owned[i] = make(chan entity, 1024) | ||
| 335 | } | ||
| 336 | |||
| 337 | for b := 0; b < batchers; b++ { | ||
| 338 | wg.Add(1) | ||
| 339 | go func(b int) { | ||
| 340 | defer wg.Done() | ||
| 341 | pending := map[int][]entity{} | ||
| 342 | flush := func(shard int, final bool) { | ||
| 343 | batch := pending[shard] | ||
| 344 | delete(pending, shard) | ||
| 345 | if len(batch) == 0 { | ||
| 346 | return | ||
| 347 | } | ||
| 348 | t0 := time.Now() | ||
| 349 | if ok := commitBatch(r, shards, shard, batch, maxRetries, &st.retries); ok { | ||
| 350 | d := time.Since(t0) | ||
| 351 | atomic.AddInt64(&st.commits, 1) | ||
| 352 | atomic.AddInt64(&st.entities, int64(len(batch))) | ||
| 353 | mu.Lock() | ||
| 354 | st.latencyNanos = append(st.latencyNanos, d.Nanoseconds()) | ||
| 355 | mu.Unlock() | ||
| 356 | } else { | ||
| 357 | atomic.AddInt64(&st.failures, int64(len(batch))) | ||
| 358 | } | ||
| 359 | } | ||
| 360 | tick := time.NewTicker(100 * time.Millisecond) | ||
| 361 | defer tick.Stop() | ||
| 362 | for { | ||
| 363 | select { | ||
| 364 | case e, ok := <-owned[b]: | ||
| 365 | if !ok { | ||
| 366 | for s := range pending { | ||
| 367 | flush(s, true) | ||
| 368 | } | ||
| 369 | return | ||
| 370 | } | ||
| 371 | pending[e.shard] = append(pending[e.shard], e) | ||
| 372 | if len(pending[e.shard]) >= batchSize { | ||
| 373 | flush(e.shard, false) | ||
| 374 | } | ||
| 375 | case <-tick.C: | ||
| 376 | for s := range pending { | ||
| 377 | flush(s, false) | ||
| 378 | } | ||
| 379 | } | ||
| 380 | } | ||
| 381 | }(b) | ||
| 382 | } | ||
| 383 | |||
| 384 | go func() { | ||
| 385 | for i := 0; i < n; i++ { | ||
| 386 | id := fmt.Sprintf("%06d", i) | ||
| 387 | shard := shardOf(id, shards) | ||
| 388 | jobs <- entity{id: id, payload: payloadFor(id), shard: shard} | ||
| 389 | } | ||
| 390 | close(jobs) | ||
| 391 | }() | ||
| 392 | |||
| 393 | // Dispatcher: single reader, deterministic ownership. | ||
| 394 | dispatched := int64(0) | ||
| 395 | for e := range jobs { | ||
| 396 | owner := e.shard % batchers | ||
| 397 | if shards <= 1 { | ||
| 398 | owner = 0 | ||
| 399 | } | ||
| 400 | owned[owner] <- e | ||
| 401 | dispatched++ | ||
| 402 | } | ||
| 403 | for i := range owned { | ||
| 404 | close(owned[i]) | ||
| 405 | } | ||
| 406 | wg.Wait() | ||
| 407 | st.elapsed = time.Since(start) | ||
| 408 | if dispatched != int64(n) { | ||
| 409 | st.failures += int64(n) - dispatched | ||
| 410 | } | ||
| 411 | return st | ||
| 412 | } | ||
| 413 | |||
| 414 | func ownsShard(shard, b, batchers, shards int) bool { | ||
| 415 | if shards <= 1 { | ||
| 416 | return b == 0 | ||
| 417 | } | ||
| 418 | return shard%batchers == b | ||
| 419 | } | ||
| 420 | |||
| 421 | // commitBatch writes K entities into one tree update + one commit + one CAS. | ||
| 422 | func commitBatch(r *repo, shards, shard int, batch []entity, maxRetries int, retries *int64) bool { | ||
| 423 | ref := refFor(shards, shard) | ||
| 424 | tmp := filepath.Join(os.TempDir(), fmt.Sprintf("vole-spike-a-%d-%d", os.Getpid(), rand.Int63())) | ||
| 425 | if err := os.MkdirAll(tmp, 0o755); err != nil { | ||
| 426 | return false | ||
| 427 | } | ||
| 428 | defer os.RemoveAll(tmp) | ||
| 429 | |||
| 430 | paths := make([]string, len(batch)) | ||
| 431 | for i, e := range batch { | ||
| 432 | p := filepath.Join(tmp, e.id+".json") | ||
| 433 | if err := os.WriteFile(p, e.payload, 0o644); err != nil { | ||
| 434 | return false | ||
| 435 | } | ||
| 436 | paths[i] = p | ||
| 437 | } | ||
| 438 | blobs, err := r.hashBlobs(paths) | ||
| 439 | if err != nil || len(blobs) != len(batch) { | ||
| 440 | return false | ||
| 441 | } | ||
| 442 | for attempt := 0; attempt <= maxRetries; attempt++ { | ||
| 443 | if attempt > 0 { | ||
| 444 | atomic.AddInt64(retries, 1) | ||
| 445 | time.Sleep(time.Duration(rand.Intn(200)+50) * time.Microsecond) | ||
| 446 | } | ||
| 447 | old, _ := r.revParse(ref) | ||
| 448 | es, err := r.treeEntries(old) | ||
| 449 | if err != nil { | ||
| 450 | return false | ||
| 451 | } | ||
| 452 | for i, e := range batch { | ||
| 453 | es = upsert(es, entry{"100644", "blob", blobs[i], e.id + ".json"}) | ||
| 454 | } | ||
| 455 | tree, err := r.mkTree(es) | ||
| 456 | if err != nil { | ||
| 457 | return false | ||
| 458 | } | ||
| 459 | commit, err := r.commitTree(tree, old, fmt.Sprintf("batch of %d", len(batch))) | ||
| 460 | if err != nil { | ||
| 461 | return false | ||
| 462 | } | ||
| 463 | ok, err := r.casRef(ref, commit, old) | ||
| 464 | if err != nil { | ||
| 465 | return false | ||
| 466 | } | ||
| 467 | if ok { | ||
| 468 | return true | ||
| 469 | } | ||
| 470 | } | ||
| 471 | return false | ||
| 472 | } | ||
| 473 | |||
| 474 | // countEntities walks every metadata ref and counts tree entries, so the | ||
| 475 | // harness can prove no write was silently dropped. | ||
| 476 | func countEntities(r *repo, shards int) (int64, error) { | ||
| 477 | var refs []string | ||
| 478 | if shards <= 1 { | ||
| 479 | refs = []string{refFor(1, 0)} | ||
| 480 | } else { | ||
| 481 | for i := 0; i < shards; i++ { | ||
| 482 | refs = append(refs, refFor(shards, i)) | ||
| 483 | } | ||
| 484 | } | ||
| 485 | var total int64 | ||
| 486 | for _, ref := range refs { | ||
| 487 | c, err := r.revParse(ref) | ||
| 488 | if err != nil || c == "" { | ||
| 489 | continue | ||
| 490 | } | ||
| 491 | es, err := r.treeEntries(c) | ||
| 492 | if err != nil { | ||
| 493 | return 0, err | ||
| 494 | } | ||
| 495 | total += int64(len(es)) | ||
| 496 | } | ||
| 497 | return total, nil | ||
| 498 | } | ||
| 499 | |||
| 500 | func payloadFor(id string) []byte { | ||
| 501 | b, _ := json.Marshal(map[string]any{ | ||
| 502 | "id": id, | ||
| 503 | "title": "issue " + id, | ||
| 504 | "body": strings.Repeat("x", 400), | ||
| 505 | "state": "open", | ||
| 506 | "created": "2026-10-22T00:00:00Z", | ||
| 507 | }) | ||
| 508 | return b | ||
| 509 | } | ||
| 510 | |||
| 511 | // spawnCost measures the fixed cost of invoking git, which bounds every | ||
| 512 | // per-commit design. | ||
| 513 | func spawnCost(r *repo, n int) float64 { | ||
| 514 | start := time.Now() | ||
| 515 | for i := 0; i < n; i++ { | ||
| 516 | if _, err := r.run(nil, "rev-parse", "--verify", "--quiet", "HEAD^{commit}"); err != nil { | ||
| 517 | // HEAD does not exist in a fresh bare repo; rev-parse still exits fast. | ||
| 518 | _ = err | ||
| 519 | } | ||
| 520 | } | ||
| 521 | return time.Since(start).Seconds() / float64(n) * 1000 // ms | ||
| 522 | } | ||
| 523 | |||
| 524 | func main() { | ||
| 525 | n := flag.Int("n", 500, "entities per mode") | ||
| 526 | writers := flag.Int("writers", 32, "concurrent writers") | ||
| 527 | maxRetries := flag.Int("retries", 200, "max CAS retries per write") | ||
| 528 | batchSize := flag.Int("batch", 64, "entities per commit in batched modes") | ||
| 529 | modes := flag.String("modes", "shared-naive,sharded-naive,shared-batched,sharded-batched", "comma separated") | ||
| 530 | flag.Parse() | ||
| 531 | |||
| 532 | root, err := os.MkdirTemp("", "vole-spike-a-") | ||
| 533 | if err != nil { | ||
| 534 | panic(err) | ||
| 535 | } | ||
| 536 | defer os.RemoveAll(root) | ||
| 537 | |||
| 538 | results := []map[string]any{} | ||
| 539 | |||
| 540 | r1, err := newRepo(filepath.Join(root, "repo")) | ||
| 541 | if err != nil { | ||
| 542 | panic(err) | ||
| 543 | } | ||
| 544 | results = append(results, map[string]any{ | ||
| 545 | "mode": "git-spawn-cost", | ||
| 546 | "ms_per_spawn": spawnCost(r1, 200), | ||
| 547 | "note": "fixed cost of one git invocation; every per-commit design pays 5-6 of these", | ||
| 548 | }) | ||
| 549 | |||
| 550 | for _, m := range strings.Split(*modes, ",") { | ||
| 551 | r, err := newRepo(filepath.Join(root, strings.ReplaceAll(m, "/", "-"))) | ||
| 552 | if err != nil { | ||
| 553 | panic(err) | ||
| 554 | } | ||
| 555 | var st *stats | ||
| 556 | shardCount := 1 | ||
| 557 | switch m { | ||
| 558 | case "shared-naive": | ||
| 559 | st = runNaive(r, m, 1, *writers, *n, *maxRetries, root) | ||
| 560 | case "sharded-naive": | ||
| 561 | shardCount = 256 | ||
| 562 | st = runNaive(r, m, 256, *writers, *n, *maxRetries, root) | ||
| 563 | case "shared-batched": | ||
| 564 | st = runBatched(r, m, 1, 1, *n, *batchSize, *maxRetries) | ||
| 565 | case "sharded-batched": | ||
| 566 | shardCount = 256 | ||
| 567 | st = runBatched(r, m, 256, 8, *n, *batchSize, *maxRetries) | ||
| 568 | default: | ||
| 569 | panic("unknown mode " + m) | ||
| 570 | } | ||
| 571 | rep := st.report() | ||
| 572 | got, err := countEntities(r, shardCount) | ||
| 573 | if err != nil { | ||
| 574 | panic(err) | ||
| 575 | } | ||
| 576 | rep["shards"] = shardCount | ||
| 577 | rep["entities_in_refs"] = got | ||
| 578 | rep["lost_writes"] = int64(*n) - got | ||
| 579 | results = append(results, rep) | ||
| 580 | } | ||
| 581 | |||
| 582 | out, _ := json.MarshalIndent(results, "", " ") | ||
| 583 | fmt.Println(string(out)) | ||
| 584 | os.WriteFile(filepath.Join(root, "results.json"), out, 0o644) | ||
| 585 | } | ||