| 1 | #include "git-compat-util.h" |
| 2 | #include "abspath.h" |
| 3 | #include "blob.h" |
| 4 | #include "chdir-notify.h" |
| 5 | #include "config.h" |
| 6 | #include "gettext.h" |
| 7 | #include "lockfile.h" |
| 8 | #include "object-file.h" |
| 9 | #include "odb.h" |
| 10 | #include "odb/source.h" |
| 11 | #include "odb/source-files.h" |
| 12 | #include "odb/source-loose.h" |
| 13 | #include "pack-objects.h" |
| 14 | #include "packfile.h" |
| 15 | #include "path.h" |
| 16 | #include "promisor-remote.h" |
| 17 | #include "repack.h" |
| 18 | #include "run-command.h" |
| 19 | #include "strbuf.h" |
| 20 | #include "string-list.h" |
| 21 | #include "strvec.h" |
| 22 | #include "tree.h" |
| 23 | #include "write-or-die.h" |
| 24 | |
| 25 | static void odb_source_files_reparent(const char *name UNUSED, |
| 26 | const char *old_cwd, |
| 27 | const char *new_cwd, |
| 28 | void *cb_data) |
| 29 | { |
| 30 | struct odb_source_files *files = cb_data; |
| 31 | char *path = reparent_relative_path(old_cwd, new_cwd, |
| 32 | files->base.path); |
| 33 | free(files->base.path); |
| 34 | files->base.path = path; |
| 35 | } |
| 36 | |
| 37 | static void odb_source_files_free(struct odb_source *source) |
| 38 | { |
| 39 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 40 | chdir_notify_unregister(NULL, odb_source_files_reparent, files); |
| 41 | odb_source_free(&files->loose->base); |
| 42 | odb_source_free(&files->packed->base); |
| 43 | odb_source_release(&files->base); |
| 44 | free(files); |
| 45 | } |
| 46 | |
| 47 | static void odb_source_files_close(struct odb_source *source) |
| 48 | { |
| 49 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 50 | odb_source_close(&files->loose->base); |
| 51 | odb_source_close(&files->packed->base); |
| 52 | } |
| 53 | |
| 54 | static int odb_source_files_create_on_disk(struct odb_source *source) |
| 55 | { |
| 56 | struct strbuf path = STRBUF_INIT; |
| 57 | |
| 58 | safe_create_dir(source->odb->repo, source->path, 1); |
| 59 | |
| 60 | strbuf_addf(&path, "%s/pack", source->path); |
| 61 | safe_create_dir(source->odb->repo, path.buf, 1); |
| 62 | |
| 63 | strbuf_reset(&path); |
| 64 | strbuf_addf(&path, "%s/info", source->path); |
| 65 | safe_create_dir(source->odb->repo, path.buf, 1); |
| 66 | |
| 67 | strbuf_release(&path); |
| 68 | return 0; |
| 69 | } |
| 70 | |
| 71 | static void odb_source_files_prepare(struct odb_source *source, |
| 72 | enum odb_prepare_flags flags) |
| 73 | { |
| 74 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 75 | odb_source_prepare(&files->loose->base, flags); |
| 76 | odb_source_prepare(&files->packed->base, flags); |
| 77 | } |
| 78 | |
| 79 | static int odb_source_files_read_object_info(struct odb_source *source, |
| 80 | const struct object_id *oid, |
| 81 | struct object_info *oi, |
| 82 | enum object_info_flags flags) |
| 83 | { |
| 84 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 85 | |
| 86 | if (!odb_source_read_object_info(&files->packed->base, oid, oi, flags) || |
| 87 | !odb_source_read_object_info(&files->loose->base, oid, oi, flags)) |
| 88 | return 0; |
| 89 | |
| 90 | return -1; |
| 91 | } |
| 92 | |
| 93 | static int odb_source_files_read_object_stream(struct odb_read_stream **out, |
| 94 | struct odb_source *source, |
| 95 | const struct object_id *oid) |
| 96 | { |
| 97 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 98 | if (!odb_source_read_object_stream(out, &files->packed->base, oid) || |
| 99 | !odb_source_read_object_stream(out, &files->loose->base, oid)) |
| 100 | return 0; |
| 101 | return -1; |
| 102 | } |
| 103 | |
| 104 | static int odb_source_files_for_each_object(struct odb_source *source, |
| 105 | const struct object_info *request, |
| 106 | odb_for_each_object_cb cb, |
| 107 | void *cb_data, |
| 108 | const struct odb_for_each_object_options *opts) |
| 109 | { |
| 110 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 111 | int ret; |
| 112 | |
| 113 | if (!(opts->flags & ODB_FOR_EACH_OBJECT_PROMISOR_ONLY)) { |
| 114 | ret = odb_source_for_each_object(&files->loose->base, request, cb, cb_data, opts); |
| 115 | if (ret) |
| 116 | return ret; |
| 117 | } |
| 118 | |
| 119 | ret = odb_source_for_each_object(&files->packed->base, request, cb, cb_data, opts); |
| 120 | if (ret) |
| 121 | return ret; |
| 122 | |
| 123 | return 0; |
| 124 | } |
| 125 | |
| 126 | static int odb_source_files_count_objects(struct odb_source *source, |
| 127 | enum odb_count_objects_flags flags, |
| 128 | unsigned long *out) |
| 129 | { |
| 130 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 131 | unsigned long count; |
| 132 | int ret; |
| 133 | |
| 134 | ret = odb_source_count_objects(&files->packed->base, flags, &count); |
| 135 | if (ret < 0) |
| 136 | goto out; |
| 137 | |
| 138 | if (!(flags & ODB_COUNT_OBJECTS_APPROXIMATE)) { |
| 139 | unsigned long loose_count; |
| 140 | |
| 141 | ret = odb_source_count_objects(&files->loose->base, flags, &loose_count); |
| 142 | if (ret < 0) |
| 143 | goto out; |
| 144 | |
| 145 | count += loose_count; |
| 146 | } |
| 147 | |
| 148 | *out = count; |
| 149 | ret = 0; |
| 150 | |
| 151 | out: |
| 152 | return ret; |
| 153 | } |
| 154 | |
| 155 | static int odb_source_files_find_abbrev_len(struct odb_source *source, |
| 156 | const struct object_id *oid, |
| 157 | unsigned min_len, |
| 158 | unsigned *out) |
| 159 | { |
| 160 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 161 | unsigned len = min_len; |
| 162 | int ret; |
| 163 | |
| 164 | ret = odb_source_find_abbrev_len(&files->packed->base, oid, len, &len); |
| 165 | if (ret < 0) |
| 166 | goto out; |
| 167 | |
| 168 | ret = odb_source_find_abbrev_len(&files->loose->base, oid, len, &len); |
| 169 | if (ret < 0) |
| 170 | goto out; |
| 171 | |
| 172 | *out = len; |
| 173 | ret = 0; |
| 174 | |
| 175 | out: |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | static int odb_source_files_freshen_object(struct odb_source *source, |
| 180 | const struct object_id *oid, |
| 181 | const time_t *mtime) |
| 182 | { |
| 183 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 184 | if (odb_source_freshen_object(&files->packed->base, oid, mtime) || |
| 185 | odb_source_freshen_object(&files->loose->base, oid, mtime)) |
| 186 | return 1; |
| 187 | return 0; |
| 188 | } |
| 189 | |
| 190 | static int odb_source_files_write_object(struct odb_source *source, |
| 191 | const void *buf, size_t len, |
| 192 | enum object_type type, |
| 193 | const struct object_id *oid, |
| 194 | const struct object_id *compat_oid, |
| 195 | const time_t *mtime, |
| 196 | enum odb_write_object_flags flags) |
| 197 | { |
| 198 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 199 | return odb_source_write_object(&files->loose->base, buf, len, type, |
| 200 | oid, compat_oid, mtime, flags); |
| 201 | } |
| 202 | |
| 203 | static int odb_source_files_write_object_stream(struct odb_source *source, |
| 204 | struct odb_write_stream *stream, |
| 205 | size_t len, |
| 206 | struct object_id *oid) |
| 207 | { |
| 208 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 209 | return odb_source_write_object_stream(&files->loose->base, stream, len, oid); |
| 210 | } |
| 211 | |
| 212 | static int odb_source_files_begin_transaction(struct odb_source *source, |
| 213 | struct odb_transaction **out, |
| 214 | enum odb_transaction_flags flags) |
| 215 | { |
| 216 | return odb_transaction_files_begin(source, out, flags); |
| 217 | } |
| 218 | |
| 219 | static int odb_source_files_read_alternates(struct odb_source *source, |
| 220 | struct strvec *out) |
| 221 | { |
| 222 | struct strbuf buf = STRBUF_INIT; |
| 223 | char *path; |
| 224 | |
| 225 | path = xstrfmt("%s/info/alternates", source->path); |
| 226 | if (strbuf_read_file(&buf, path, 1024) < 0) { |
| 227 | warn_on_fopen_errors(path); |
| 228 | free(path); |
| 229 | return 0; |
| 230 | } |
| 231 | parse_alternates(buf.buf, '\n', source->path, out); |
| 232 | |
| 233 | strbuf_release(&buf); |
| 234 | free(path); |
| 235 | return 0; |
| 236 | } |
| 237 | |
| 238 | static int odb_source_files_write_alternate(struct odb_source *source, |
| 239 | const char *alternate) |
| 240 | { |
| 241 | struct lock_file lock = LOCK_INIT; |
| 242 | char *path = xstrfmt("%s/%s", source->path, "info/alternates"); |
| 243 | FILE *in, *out; |
| 244 | int found = 0; |
| 245 | int ret; |
| 246 | |
| 247 | repo_hold_lock_file_for_update(source->odb->repo, &lock, path, |
| 248 | LOCK_DIE_ON_ERROR); |
| 249 | out = fdopen_lock_file(&lock, "w"); |
| 250 | if (!out) { |
| 251 | ret = error_errno(_("unable to fdopen alternates lockfile")); |
| 252 | goto out; |
| 253 | } |
| 254 | |
| 255 | in = fopen(path, "r"); |
| 256 | if (in) { |
| 257 | struct strbuf line = STRBUF_INIT; |
| 258 | |
| 259 | while (strbuf_getline(&line, in) != EOF) { |
| 260 | if (!strcmp(alternate, line.buf)) { |
| 261 | found = 1; |
| 262 | break; |
| 263 | } |
| 264 | fprintf_or_die(out, "%s\n", line.buf); |
| 265 | } |
| 266 | |
| 267 | strbuf_release(&line); |
| 268 | fclose(in); |
| 269 | } else if (errno != ENOENT) { |
| 270 | ret = error_errno(_("unable to read alternates file")); |
| 271 | goto out; |
| 272 | } |
| 273 | |
| 274 | if (found) { |
| 275 | rollback_lock_file(&lock); |
| 276 | } else { |
| 277 | fprintf_or_die(out, "%s\n", alternate); |
| 278 | if (commit_lock_file(&lock)) { |
| 279 | ret = error_errno(_("unable to move new alternates file into place")); |
| 280 | goto out; |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | ret = 0; |
| 285 | |
| 286 | out: |
| 287 | free(path); |
| 288 | return ret; |
| 289 | } |
| 290 | |
| 291 | static int too_many_loose_objects(struct odb_source_files *files, int limit) |
| 292 | { |
| 293 | unsigned long loose_count; |
| 294 | |
| 295 | if (limit <= 0) |
| 296 | return 0; |
| 297 | |
| 298 | if (odb_source_count_objects(&files->loose->base, ODB_COUNT_OBJECTS_APPROXIMATE, |
| 299 | &loose_count) < 0) |
| 300 | return 0; |
| 301 | |
| 302 | /* |
| 303 | * This is weird, but stems from legacy behaviour: the GC auto |
| 304 | * threshold was always essentially interpreted as if it was rounded up |
| 305 | * to the next multiple 256 of, so we retain this behaviour for now. |
| 306 | */ |
| 307 | return loose_count > (DIV_ROUND_UP(((unsigned long) limit), 256) * 256); |
| 308 | } |
| 309 | |
| 310 | static struct packed_git *find_base_packs(struct odb_source_files *files, |
| 311 | struct string_list *packs, |
| 312 | unsigned long limit) |
| 313 | { |
| 314 | struct packfile_list_entry *e; |
| 315 | struct packed_git *base = NULL; |
| 316 | |
| 317 | for (e = packfile_store_get_packs(files->packed); e; e = e->next) { |
| 318 | if (e->pack->is_cruft) |
| 319 | continue; |
| 320 | if (limit) { |
| 321 | if ((uintmax_t) e->pack->pack_size >= limit) |
| 322 | string_list_append(packs, e->pack->pack_name); |
| 323 | } else if (!base || base->pack_size < e->pack->pack_size) { |
| 324 | base = e->pack; |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | if (base) |
| 329 | string_list_append(packs, base->pack_name); |
| 330 | |
| 331 | return base; |
| 332 | } |
| 333 | |
| 334 | static int too_many_packs(struct odb_source_files *files, int gc_auto_pack_limit) |
| 335 | { |
| 336 | struct packfile_list_entry *e; |
| 337 | int cnt = 0; |
| 338 | |
| 339 | if (gc_auto_pack_limit <= 0) |
| 340 | return 0; |
| 341 | |
| 342 | for (e = packfile_store_get_packs(files->packed); e; e = e->next) { |
| 343 | if (e->pack->pack_keep) |
| 344 | continue; |
| 345 | /* |
| 346 | * Perhaps check the size of the pack and count only |
| 347 | * very small ones here? |
| 348 | */ |
| 349 | cnt++; |
| 350 | } |
| 351 | return gc_auto_pack_limit < cnt; |
| 352 | } |
| 353 | |
| 354 | static uint64_t total_ram(void) |
| 355 | { |
| 356 | #if defined(HAVE_SYSINFO) |
| 357 | struct sysinfo si; |
| 358 | |
| 359 | if (!sysinfo(&si)) { |
| 360 | uint64_t total = si.totalram; |
| 361 | |
| 362 | if (si.mem_unit > 1) |
| 363 | total *= (uint64_t)si.mem_unit; |
| 364 | return total; |
| 365 | } |
| 366 | #elif defined(HAVE_BSD_SYSCTL) && (defined(HW_MEMSIZE) || defined(HW_PHYSMEM) || defined(HW_PHYSMEM64)) |
| 367 | uint64_t physical_memory; |
| 368 | int mib[2]; |
| 369 | size_t length; |
| 370 | |
| 371 | mib[0] = CTL_HW; |
| 372 | # if defined(HW_MEMSIZE) |
| 373 | mib[1] = HW_MEMSIZE; |
| 374 | # elif defined(HW_PHYSMEM64) |
| 375 | mib[1] = HW_PHYSMEM64; |
| 376 | # else |
| 377 | mib[1] = HW_PHYSMEM; |
| 378 | # endif |
| 379 | length = sizeof(physical_memory); |
| 380 | if (!sysctl(mib, 2, &physical_memory, &length, NULL, 0)) { |
| 381 | if (length == 4) { |
| 382 | uint32_t mem; |
| 383 | |
| 384 | if (!sysctl(mib, 2, &mem, &length, NULL, 0)) |
| 385 | physical_memory = mem; |
| 386 | } |
| 387 | return physical_memory; |
| 388 | } |
| 389 | #elif defined(GIT_WINDOWS_NATIVE) |
| 390 | MEMORYSTATUSEX memInfo; |
| 391 | |
| 392 | memInfo.dwLength = sizeof(MEMORYSTATUSEX); |
| 393 | if (GlobalMemoryStatusEx(&memInfo)) |
| 394 | return memInfo.ullTotalPhys; |
| 395 | #endif |
| 396 | return 0; |
| 397 | } |
| 398 | |
| 399 | static uint64_t estimate_repack_memory(struct odb_source_files *files, |
| 400 | struct packed_git *pack) |
| 401 | { |
| 402 | unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; |
| 403 | unsigned long delta_base_cache_limit = DEFAULT_DELTA_BASE_CACHE_LIMIT; |
| 404 | unsigned long nr_objects; |
| 405 | size_t os_cache, heap; |
| 406 | |
| 407 | if (odb_source_count_objects(&files->base, ODB_COUNT_OBJECTS_APPROXIMATE, |
| 408 | &nr_objects) < 0) |
| 409 | return 0; |
| 410 | |
| 411 | if (!pack || !nr_objects) |
| 412 | return 0; |
| 413 | |
| 414 | repo_config_get_ulong(files->base.odb->repo, "pack.deltacachesize", |
| 415 | &max_delta_cache_size); |
| 416 | repo_config_get_ulong(files->base.odb->repo, "core.deltabasecachelimit", |
| 417 | &delta_base_cache_limit); |
| 418 | |
| 419 | /* |
| 420 | * First we have to scan through at least one pack. |
| 421 | * Assume enough room in OS file cache to keep the entire pack |
| 422 | * or we may accidentally evict data of other processes from |
| 423 | * the cache. |
| 424 | */ |
| 425 | os_cache = pack->pack_size + pack->index_size; |
| 426 | /* then pack-objects needs lots more for book keeping */ |
| 427 | heap = sizeof(struct object_entry) * nr_objects; |
| 428 | /* |
| 429 | * internal rev-list --all --objects takes up some memory too, |
| 430 | * let's say half of it is for blobs |
| 431 | */ |
| 432 | heap += sizeof(struct blob) * nr_objects / 2; |
| 433 | /* |
| 434 | * and the other half is for trees (commits and tags are |
| 435 | * usually insignificant) |
| 436 | */ |
| 437 | heap += sizeof(struct tree) * nr_objects / 2; |
| 438 | /* and then obj_hash[], underestimated in fact */ |
| 439 | heap += sizeof(struct object *) * nr_objects; |
| 440 | /* revindex is used also */ |
| 441 | heap += (sizeof(off_t) + sizeof(uint32_t)) * nr_objects; |
| 442 | /* |
| 443 | * read_sha1_file() (either at delta calculation phase, or |
| 444 | * writing phase) also fills up the delta base cache |
| 445 | */ |
| 446 | heap += delta_base_cache_limit; |
| 447 | /* and of course pack-objects has its own delta cache */ |
| 448 | heap += max_delta_cache_size; |
| 449 | |
| 450 | return os_cache + heap; |
| 451 | } |
| 452 | |
| 453 | static int keep_one_pack(struct string_list_item *item, void *data) |
| 454 | { |
| 455 | struct strvec *args = data; |
| 456 | strvec_pushf(args, "--keep-pack=%s", basename(item->string)); |
| 457 | return 0; |
| 458 | } |
| 459 | |
| 460 | static void add_repack_all_option(struct repository *repo, |
| 461 | const struct odb_optimize_options *opts, |
| 462 | struct string_list *keep_pack, |
| 463 | struct strvec *args) |
| 464 | { |
| 465 | char *repack_filter = NULL; |
| 466 | char *repack_filter_to = NULL; |
| 467 | |
| 468 | repo_config_get_string(repo, "gc.repackfilter", &repack_filter); |
| 469 | repo_config_get_string(repo, "gc.repackfilterto", &repack_filter_to); |
| 470 | |
| 471 | if (opts->prune_expire && !strcmp(opts->prune_expire, "now") && |
| 472 | !(opts->cruft_packs && opts->expire_to)) |
| 473 | strvec_push(args, "-a"); |
| 474 | else if (opts->cruft_packs) { |
| 475 | strvec_push(args, "--cruft"); |
| 476 | if (opts->prune_expire) |
| 477 | strvec_pushf(args, "--cruft-expiration=%s", opts->prune_expire); |
| 478 | if (opts->max_cruft_size) |
| 479 | strvec_pushf(args, "--max-cruft-size=%lu", |
| 480 | opts->max_cruft_size); |
| 481 | if (opts->expire_to) |
| 482 | strvec_pushf(args, "--expire-to=%s", opts->expire_to); |
| 483 | } else { |
| 484 | strvec_push(args, "-A"); |
| 485 | if (opts->prune_expire) |
| 486 | strvec_pushf(args, "--unpack-unreachable=%s", opts->prune_expire); |
| 487 | } |
| 488 | |
| 489 | if (keep_pack) |
| 490 | for_each_string_list(keep_pack, keep_one_pack, args); |
| 491 | |
| 492 | if (repack_filter && *repack_filter) |
| 493 | strvec_pushf(args, "--filter=%s", repack_filter); |
| 494 | if (repack_filter_to && *repack_filter_to) |
| 495 | strvec_pushf(args, "--filter-to=%s", repack_filter_to); |
| 496 | |
| 497 | free(repack_filter); |
| 498 | free(repack_filter_to); |
| 499 | } |
| 500 | |
| 501 | static void add_repack_incremental_option(struct strvec *args) |
| 502 | { |
| 503 | strvec_push(args, "--no-write-bitmap-index"); |
| 504 | } |
| 505 | |
| 506 | bool odb_source_files_optimize_required(struct odb_source *source, |
| 507 | const struct odb_optimize_options *opts) |
| 508 | { |
| 509 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 510 | struct repository *repo = source->odb->repo; |
| 511 | |
| 512 | switch (opts->strategy) { |
| 513 | case ODB_OPTIMIZE_INCREMENTAL: { |
| 514 | int gc_auto_threshold = 6700; |
| 515 | int gc_auto_pack_limit = 50; |
| 516 | |
| 517 | repo_config_get_int(repo, "gc.auto", &gc_auto_threshold); |
| 518 | repo_config_get_int(repo, "gc.autopacklimit", &gc_auto_pack_limit); |
| 519 | |
| 520 | /* |
| 521 | * Setting gc.auto to 0 or negative can disable the |
| 522 | * automatic gc. |
| 523 | */ |
| 524 | if (gc_auto_threshold <= 0) |
| 525 | return false; |
| 526 | if (!too_many_packs(files, gc_auto_pack_limit) && |
| 527 | !too_many_loose_objects(files, gc_auto_threshold)) |
| 528 | return false; |
| 529 | |
| 530 | return true; |
| 531 | } |
| 532 | case ODB_OPTIMIZE_GEOMETRIC: { |
| 533 | struct pack_geometry geometry = { |
| 534 | .split_factor = 2, |
| 535 | }; |
| 536 | struct pack_objects_args po_args = { |
| 537 | .local = 1, |
| 538 | }; |
| 539 | struct existing_packs existing_packs = EXISTING_PACKS_INIT; |
| 540 | struct string_list kept_packs = STRING_LIST_INIT_DUP; |
| 541 | int auto_value = 100; |
| 542 | bool ret; |
| 543 | |
| 544 | repo_config_get_int(repo, "maintenance.geometric-repack.auto", |
| 545 | &auto_value); |
| 546 | if (!auto_value) |
| 547 | return false; |
| 548 | if (auto_value < 0) |
| 549 | return true; |
| 550 | |
| 551 | repo_config_get_int(repo, "maintenance.geometric-repack.splitFactor", |
| 552 | &geometry.split_factor); |
| 553 | |
| 554 | existing_packs.repo = repo; |
| 555 | existing_packs_collect(&existing_packs, &kept_packs); |
| 556 | pack_geometry_init(&geometry, &existing_packs, &po_args); |
| 557 | pack_geometry_split(&geometry); |
| 558 | |
| 559 | /* |
| 560 | * When we'd merge at least two packs with one another we always |
| 561 | * perform the repack. |
| 562 | */ |
| 563 | if (geometry.split) { |
| 564 | ret = true; |
| 565 | goto out; |
| 566 | } |
| 567 | |
| 568 | /* |
| 569 | * Otherwise, we estimate the number of loose objects to determine |
| 570 | * whether we want to create a new packfile or not. |
| 571 | */ |
| 572 | if (too_many_loose_objects(files, auto_value)) { |
| 573 | ret = true; |
| 574 | goto out; |
| 575 | } |
| 576 | |
| 577 | ret = false; |
| 578 | |
| 579 | out: |
| 580 | existing_packs_release(&existing_packs); |
| 581 | pack_geometry_release(&geometry); |
| 582 | return ret; |
| 583 | } |
| 584 | default: |
| 585 | BUG("unknown maintenance strategy '%d'", opts->strategy); |
| 586 | } |
| 587 | } |
| 588 | |
| 589 | int odb_source_files_optimize(struct odb_source *source, |
| 590 | const struct odb_optimize_options *opts) |
| 591 | { |
| 592 | struct odb_source_files *files = odb_source_files_downcast(source); |
| 593 | struct repository *repo = source->odb->repo; |
| 594 | struct child_process repack_cmd = CHILD_PROCESS_INIT; |
| 595 | unsigned long big_pack_threshold = 0; |
| 596 | int gc_auto_threshold = 6700; |
| 597 | int gc_auto_pack_limit = 50; |
| 598 | int ret; |
| 599 | |
| 600 | repo_config_get_int(repo, "gc.auto", &gc_auto_threshold); |
| 601 | repo_config_get_int(repo, "gc.autopacklimit", &gc_auto_pack_limit); |
| 602 | repo_config_get_ulong(repo, "gc.bigpackthreshold", &big_pack_threshold); |
| 603 | |
| 604 | if (repo->repository_format_precious_objects) |
| 605 | return 0; |
| 606 | |
| 607 | repack_cmd.git_cmd = 1; |
| 608 | repack_cmd.odb_to_close = repo->objects; |
| 609 | |
| 610 | strvec_pushl(&repack_cmd.args, "repack", "-d", "-l", NULL); |
| 611 | if (opts->flags & ODB_OPTIMIZE_NO_REUSE_DELTAS) |
| 612 | strvec_push(&repack_cmd.args, "-f"); |
| 613 | if (opts->depth > 0) |
| 614 | strvec_pushf(&repack_cmd.args, "--depth=%d", opts->depth); |
| 615 | if (opts->window > 0) |
| 616 | strvec_pushf(&repack_cmd.args, "--window=%d", opts->window); |
| 617 | if (!(opts->flags & ODB_OPTIMIZE_VERBOSE)) |
| 618 | strvec_push(&repack_cmd.args, "-q"); |
| 619 | |
| 620 | /* |
| 621 | * There's three cases we need to consider: |
| 622 | * |
| 623 | * - If we're invoked without `--auto` we'll need to perform a full |
| 624 | * repack. |
| 625 | * |
| 626 | * - If we're invoked with `--auto` and there's too many packs, then |
| 627 | * we perform a full repack, as well. |
| 628 | * |
| 629 | * - Otherwise we perform an incremental repack. |
| 630 | */ |
| 631 | switch (opts->strategy) { |
| 632 | case ODB_OPTIMIZE_INCREMENTAL: |
| 633 | if (!(opts->flags & ODB_OPTIMIZE_AUTO)) { |
| 634 | struct string_list keep_pack = STRING_LIST_INIT_NODUP; |
| 635 | |
| 636 | if (opts->keep_largest_pack != -1) { |
| 637 | if (opts->keep_largest_pack) |
| 638 | find_base_packs(files, &keep_pack, 0); |
| 639 | } else if (big_pack_threshold) { |
| 640 | find_base_packs(files, &keep_pack, big_pack_threshold); |
| 641 | } |
| 642 | |
| 643 | add_repack_all_option(repo, opts, &keep_pack, &repack_cmd.args); |
| 644 | string_list_clear(&keep_pack, 0); |
| 645 | } else { |
| 646 | if (too_many_packs(files, gc_auto_pack_limit)) { |
| 647 | struct string_list keep_pack = STRING_LIST_INIT_NODUP; |
| 648 | |
| 649 | if (big_pack_threshold) { |
| 650 | find_base_packs(files, &keep_pack, big_pack_threshold); |
| 651 | if (keep_pack.nr >= (unsigned long) gc_auto_pack_limit) { |
| 652 | string_list_clear(&keep_pack, 0); |
| 653 | find_base_packs(files, &keep_pack, 0); |
| 654 | } |
| 655 | } else { |
| 656 | struct packed_git *p = find_base_packs(files, &keep_pack, 0); |
| 657 | uint64_t mem_have, mem_want; |
| 658 | |
| 659 | mem_have = total_ram(); |
| 660 | mem_want = estimate_repack_memory(files, p); |
| 661 | |
| 662 | /* |
| 663 | * Only allow 1/2 of memory for pack-objects, leave |
| 664 | * the rest for the OS and other processes in the |
| 665 | * system. |
| 666 | */ |
| 667 | if (!mem_have || mem_want < mem_have / 2) |
| 668 | string_list_clear(&keep_pack, 0); |
| 669 | } |
| 670 | |
| 671 | add_repack_all_option(repo, opts, &keep_pack, &repack_cmd.args); |
| 672 | string_list_clear(&keep_pack, 0); |
| 673 | } else { |
| 674 | add_repack_incremental_option(&repack_cmd.args); |
| 675 | } |
| 676 | } |
| 677 | |
| 678 | break; |
| 679 | case ODB_OPTIMIZE_GEOMETRIC: { |
| 680 | struct pack_geometry geometry = { |
| 681 | .split_factor = 2, |
| 682 | }; |
| 683 | struct pack_objects_args po_args = { |
| 684 | .local = 1, |
| 685 | }; |
| 686 | struct existing_packs existing_packs = EXISTING_PACKS_INIT; |
| 687 | struct string_list kept_packs = STRING_LIST_INIT_DUP; |
| 688 | |
| 689 | repo_config_get_int(repo, "maintenance.geometric-repack.splitFactor", |
| 690 | &geometry.split_factor); |
| 691 | |
| 692 | existing_packs.repo = repo; |
| 693 | existing_packs_collect(&existing_packs, &kept_packs); |
| 694 | pack_geometry_init(&geometry, &existing_packs, &po_args); |
| 695 | pack_geometry_split(&geometry); |
| 696 | |
| 697 | if (geometry.split < geometry.pack_nr) { |
| 698 | strvec_pushf(&repack_cmd.args, "--geometric=%d", |
| 699 | geometry.split_factor); |
| 700 | } else { |
| 701 | add_repack_all_option(repo, opts, NULL, &repack_cmd.args); |
| 702 | } |
| 703 | if (repo->settings.core_multi_pack_index) |
| 704 | strvec_push(&repack_cmd.args, "--write-midx"); |
| 705 | |
| 706 | existing_packs_release(&existing_packs); |
| 707 | pack_geometry_release(&geometry); |
| 708 | break; |
| 709 | } |
| 710 | default: |
| 711 | die("unknown maintenance strategy '%d'", opts->strategy); |
| 712 | } |
| 713 | |
| 714 | if (run_command(&repack_cmd)) { |
| 715 | ret = error("failed to run %s", repack_cmd.args.v[0]); |
| 716 | goto out; |
| 717 | } |
| 718 | |
| 719 | /* Geometric repacking uses cruft packs, so we don't have to prune separately. */ |
| 720 | if (opts->strategy != ODB_OPTIMIZE_GEOMETRIC && opts->prune_expire) { |
| 721 | struct child_process prune_cmd = CHILD_PROCESS_INIT; |
| 722 | |
| 723 | strvec_pushl(&prune_cmd.args, "prune", "--expire", NULL); |
| 724 | /* run `git prune` even if using cruft packs */ |
| 725 | strvec_push(&prune_cmd.args, opts->prune_expire); |
| 726 | if (!(opts->flags & ODB_OPTIMIZE_VERBOSE)) |
| 727 | strvec_push(&prune_cmd.args, "--no-progress"); |
| 728 | if (repo_has_promisor_remote(repo)) |
| 729 | strvec_push(&prune_cmd.args, |
| 730 | "--exclude-promisor-objects"); |
| 731 | prune_cmd.git_cmd = 1; |
| 732 | |
| 733 | if (run_command(&prune_cmd)) { |
| 734 | ret = error("failed to run %s", prune_cmd.args.v[0]); |
| 735 | goto out; |
| 736 | } |
| 737 | } |
| 738 | |
| 739 | if (opts->flags & ODB_OPTIMIZE_AUTO && too_many_loose_objects(files, gc_auto_threshold)) |
| 740 | warning(_("There are too many unreachable loose objects; " |
| 741 | "run 'git prune' to remove them.")); |
| 742 | |
| 743 | ret = 0; |
| 744 | |
| 745 | out: |
| 746 | return ret; |
| 747 | } |
| 748 | |
| 749 | struct odb_source_files *odb_source_files_new(struct object_database *odb, |
| 750 | const char *path, |
| 751 | bool local) |
| 752 | { |
| 753 | struct odb_source_files *files; |
| 754 | |
| 755 | CALLOC_ARRAY(files, 1); |
| 756 | odb_source_init(&files->base, odb, ODB_SOURCE_FILES, path, local); |
| 757 | files->loose = odb_source_loose_new(odb, path, local); |
| 758 | files->packed = odb_source_packed_new(odb, path, local); |
| 759 | |
| 760 | files->base.free = odb_source_files_free; |
| 761 | files->base.close = odb_source_files_close; |
| 762 | files->base.create_on_disk = odb_source_files_create_on_disk; |
| 763 | files->base.prepare = odb_source_files_prepare; |
| 764 | files->base.read_object_info = odb_source_files_read_object_info; |
| 765 | files->base.read_object_stream = odb_source_files_read_object_stream; |
| 766 | files->base.for_each_object = odb_source_files_for_each_object; |
| 767 | files->base.count_objects = odb_source_files_count_objects; |
| 768 | files->base.find_abbrev_len = odb_source_files_find_abbrev_len; |
| 769 | files->base.freshen_object = odb_source_files_freshen_object; |
| 770 | files->base.write_object = odb_source_files_write_object; |
| 771 | files->base.write_object_stream = odb_source_files_write_object_stream; |
| 772 | files->base.begin_transaction = odb_source_files_begin_transaction; |
| 773 | files->base.read_alternates = odb_source_files_read_alternates; |
| 774 | files->base.write_alternate = odb_source_files_write_alternate; |
| 775 | files->base.optimize = odb_source_files_optimize; |
| 776 | files->base.optimize_required = odb_source_files_optimize_required; |
| 777 | |
| 778 | /* |
| 779 | * Ideally, we would only ever store absolute paths in the source. This |
| 780 | * is not (yet) possible though because we access and assume relative |
| 781 | * paths in the primary ODB source in some user-facing functionality. |
| 782 | */ |
| 783 | if (!is_absolute_path(path)) |
| 784 | chdir_notify_register(NULL, odb_source_files_reparent, files); |
| 785 | |
| 786 | return files; |
| 787 | } |