| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | #include "rrdengine.h" |
| 3 | |
| 4 | #ifdef OS_WINDOWS |
| 5 | void sync_uv_file_data(uv_file file) |
| 6 | { |
| 7 | uv_fs_t req; |
| 8 | (void) uv_fs_fsync(NULL, &req, file, NULL); |
| 9 | uv_fs_req_cleanup(&req); |
| 10 | } |
| 11 | #endif |
| 12 | |
| 13 | void datafile_list_insert(struct rrdengine_instance *ctx, struct rrdengine_datafile *datafile) |
| 14 | { |
| 15 | netdata_rwlock_wrlock(&ctx->datafiles.rwlock); |
| 16 | Pvoid_t *Pvalue = JudyLIns(&ctx->datafiles.JudyL, (Word_t ) datafile->fileno, PJE0); |
| 17 | if(!Pvalue || Pvalue == PJERR) |
| 18 | fatal("DBENGINE: cannot insert datafile %u of tier %d into the datafiles list", |
| 19 | datafile->fileno, ctx->config.tier); |
| 20 | *Pvalue = datafile; |
| 21 | |
| 22 | #ifdef OS_WINDOWS |
| 23 | sync_uv_file_data(datafile->file); |
| 24 | sync_uv_file_data(datafile->journalfile->file); |
| 25 | datafile->writers.last_sync_time = now_realtime_sec(); |
| 26 | #endif |
| 27 | |
| 28 | netdata_rwlock_wrunlock(&ctx->datafiles.rwlock); |
| 29 | } |
| 30 | |
| 31 | void datafile_list_delete_unsafe(struct rrdengine_instance *ctx, struct rrdengine_datafile *datafile) |
| 32 | { |
| 33 | (void) JudyLDel(&ctx->datafiles.JudyL, (Word_t)datafile->fileno, PJE0); |
| 34 | } |
| 35 | |
| 36 | |
| 37 | static struct rrdengine_datafile *datafile_alloc_and_init(struct rrdengine_instance *ctx, unsigned tier, unsigned fileno) |
| 38 | { |
| 39 | fatal_assert(tier == 1); |
| 40 | |
| 41 | struct rrdengine_datafile *datafile = callocz(1, sizeof(struct rrdengine_datafile)); |
| 42 | |
| 43 | datafile->tier = tier; |
| 44 | datafile->fileno = fileno; |
| 45 | fatal_assert(0 == netdata_rwlock_init(&datafile->extent_rwlock)); |
| 46 | datafile->ctx = ctx; |
| 47 | datafile->magic1 = datafile->magic2 = DATAFILE_MAGIC; |
| 48 | |
| 49 | datafile->users.available = true; |
| 50 | datafile->users.pending_deletion = false; |
| 51 | |
| 52 | spinlock_init(&datafile->users.spinlock); |
| 53 | spinlock_init(&datafile->writers.spinlock); |
| 54 | rw_spinlock_init(&datafile->extent_epdl.spinlock); |
| 55 | |
| 56 | return datafile; |
| 57 | } |
| 58 | |
| 59 | ALWAYS_INLINE bool datafile_acquire(struct rrdengine_datafile *df, DATAFILE_ACQUIRE_REASONS reason) |
| 60 | { |
| 61 | bool ret = false; |
| 62 | |
| 63 | spinlock_lock(&df->users.spinlock); |
| 64 | |
| 65 | if(df->users.available && reason < DATAFILE_ACQUIRE_MAX) { |
| 66 | bool allow = true; |
| 67 | |
| 68 | if(df->users.pending_deletion) { |
| 69 | if(reason == DATAFILE_ACQUIRE_OPEN_CACHE) { |
| 70 | // Hold writers.spinlock to read the writer counters, ensuring proper |
| 71 | // memory ordering on weakly-ordered architectures (ARM) - consistent |
| 72 | // with all other read sites in the codebase. |
| 73 | spinlock_lock(&df->writers.spinlock); |
| 74 | size_t writers_running = df->writers.running; |
| 75 | size_t flushed_to_open_running = df->writers.flushed_to_open_running; |
| 76 | spinlock_unlock(&df->writers.spinlock); |
| 77 | allow = (writers_running || flushed_to_open_running); |
| 78 | } |
| 79 | else |
| 80 | allow = false; |
| 81 | } |
| 82 | |
| 83 | if(allow) { |
| 84 | ret = true; |
| 85 | df->users.lockers++; |
| 86 | df->users.lockers_by_reason[reason]++; |
| 87 | } |
| 88 | } |
| 89 | |
| 90 | spinlock_unlock(&df->users.spinlock); |
| 91 | |
| 92 | return ret; |
| 93 | } |
| 94 | |
| 95 | void datafile_release_with_trace(struct rrdengine_datafile *df, DATAFILE_ACQUIRE_REASONS reason, const char *func) { |
| 96 | spinlock_lock(&df->users.spinlock); |
| 97 | if(!df->users.lockers) |
| 98 | fatal("DBENGINE DATAFILE: cannot release datafile %u of tier %u - it is not acquired, called from %s() with reason %u", |
| 99 | df->fileno, df->tier, func, reason); |
| 100 | |
| 101 | df->users.lockers--; |
| 102 | df->users.lockers_by_reason[reason]--; |
| 103 | spinlock_unlock(&df->users.spinlock); |
| 104 | } |
| 105 | |
| 106 | bool datafile_acquire_for_deletion(struct rrdengine_datafile *df, bool is_shutdown) |
| 107 | { |
| 108 | bool can_be_deleted = false; |
| 109 | bool marked_pending = false; |
| 110 | bool should_evict_open_pages = false; |
| 111 | |
| 112 | spinlock_lock(&df->users.spinlock); |
| 113 | |
| 114 | if(!df->users.pending_deletion) { |
| 115 | df->users.pending_deletion = true; |
| 116 | marked_pending = true; |
| 117 | } |
| 118 | |
| 119 | // Hold writers.spinlock to read the writer counters, ensuring proper memory ordering |
| 120 | // on weakly-ordered architectures (ARM). Without this, stale reads could cause premature |
| 121 | // deletion while a writer is still active - writers use these counters (not lockers) for |
| 122 | // their lifecycle, so lockers alone cannot protect against this race. |
| 123 | spinlock_lock(&df->writers.spinlock); |
| 124 | size_t writers_running = df->writers.running; |
| 125 | size_t flushed_to_open_running = df->writers.flushed_to_open_running; |
| 126 | spinlock_unlock(&df->writers.spinlock); |
| 127 | |
| 128 | if(!writers_running && !flushed_to_open_running && !df->users.lockers) { |
| 129 | can_be_deleted = true; |
| 130 | df->users.available = false; |
| 131 | } |
| 132 | else if(df->users.lockers) |
| 133 | should_evict_open_pages = true; |
| 134 | spinlock_unlock(&df->users.spinlock); |
| 135 | |
| 136 | if(marked_pending) |
| 137 | netdata_log_info("DBENGINE: tier %d: " DATAFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL " is pending deletion (%s)", |
| 138 | datafile_ctx(df)->config.tier, df->tier, df->fileno, is_shutdown ? "shutdown" : "runtime"); |
| 139 | |
| 140 | if(can_be_deleted) |
| 141 | return true; |
| 142 | |
| 143 | if(should_evict_open_pages) |
| 144 | pgc_open_evict_clean_pages_of_datafile(open_cache, df); |
| 145 | |
| 146 | usec_t time_to_scan_ut = now_monotonic_usec(); |
| 147 | size_t clean_pages_in_open_cache = pgc_count_clean_pages_having_data_ptr(open_cache, (Word_t)datafile_ctx(df), df); |
| 148 | size_t hot_pages_in_open_cache = pgc_count_hot_pages_having_data_ptr(open_cache, (Word_t)datafile_ctx(df), df); |
| 149 | time_to_scan_ut = now_monotonic_usec() - time_to_scan_ut; |
| 150 | |
| 151 | spinlock_lock(&df->users.spinlock); |
| 152 | |
| 153 | spinlock_lock(&df->writers.spinlock); |
| 154 | writers_running = df->writers.running; |
| 155 | flushed_to_open_running = df->writers.flushed_to_open_running; |
| 156 | spinlock_unlock(&df->writers.spinlock); |
| 157 | |
| 158 | if(!writers_running && !flushed_to_open_running) { |
| 159 | if(df->users.available) { |
| 160 | df->users.available = false; |
| 161 | netdata_log_info("DBENGINE: tier %d: " DATAFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL " entered deletion phase-2 (new users blocked)", |
| 162 | datafile_ctx(df)->config.tier, df->tier, df->fileno); |
| 163 | } |
| 164 | |
| 165 | if(!df->users.lockers) |
| 166 | can_be_deleted = true; |
| 167 | |
| 168 | else if(!clean_pages_in_open_cache && !hot_pages_in_open_cache) { |
| 169 | time_t now_s = now_monotonic_sec(); |
| 170 | |
| 171 | if(!df->users.time_to_evict) { |
| 172 | df->users.time_to_evict = now_s + (is_shutdown ? DATAFILE_DELETE_TIMEOUT_SHORT : DATAFILE_DELETE_TIMEOUT_LONG); |
| 173 | internal_error(true, "DBENGINE: datafile %u of tier %d pending deletion has %u lockers " |
| 174 | "(oc:%u, pd:%u, rt:%u, ix:%u), writers %zu/%zu, open-cache clean/hot %zu/%zu " |
| 175 | "- will force-delete shortly (scanned in %"PRIu64" usecs)", |
| 176 | df->fileno, datafile_ctx(df)->config.tier, |
| 177 | df->users.lockers, |
| 178 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_OPEN_CACHE], |
| 179 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_PAGE_DETAILS], |
| 180 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_RETENTION], |
| 181 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_INDEXING], |
| 182 | writers_running, |
| 183 | flushed_to_open_running, |
| 184 | clean_pages_in_open_cache, |
| 185 | hot_pages_in_open_cache, |
| 186 | time_to_scan_ut); |
| 187 | } |
| 188 | else if(now_s > df->users.time_to_evict) { |
| 189 | can_be_deleted = true; |
| 190 | internal_error(true, "DBENGINE: datafile %u of tier %d pending deletion has %u lockers " |
| 191 | "(oc:%u, pd:%u, rt:%u, ix:%u), writers %zu/%zu, open-cache clean/hot %zu/%zu " |
| 192 | "- forcing delete now (scanned in %"PRIu64" usecs)", |
| 193 | df->fileno, datafile_ctx(df)->config.tier, |
| 194 | df->users.lockers, |
| 195 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_OPEN_CACHE], |
| 196 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_PAGE_DETAILS], |
| 197 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_RETENTION], |
| 198 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_INDEXING], |
| 199 | writers_running, |
| 200 | flushed_to_open_running, |
| 201 | clean_pages_in_open_cache, |
| 202 | hot_pages_in_open_cache, |
| 203 | time_to_scan_ut); |
| 204 | } |
| 205 | } |
| 206 | } |
| 207 | |
| 208 | if(!can_be_deleted) |
| 209 | internal_error(true, "DBENGINE: datafile %u of tier %d pending deletion has %u lockers " |
| 210 | "(oc:%u, pd:%u, rt:%u, ix:%u), writers %zu/%zu, open-cache clean/hot %zu/%zu " |
| 211 | "(scanned in %"PRIu64" usecs)", |
| 212 | df->fileno, datafile_ctx(df)->config.tier, |
| 213 | df->users.lockers, |
| 214 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_OPEN_CACHE], |
| 215 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_PAGE_DETAILS], |
| 216 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_RETENTION], |
| 217 | df->users.lockers_by_reason[DATAFILE_ACQUIRE_INDEXING], |
| 218 | writers_running, |
| 219 | flushed_to_open_running, |
| 220 | clean_pages_in_open_cache, |
| 221 | hot_pages_in_open_cache, |
| 222 | time_to_scan_ut); |
| 223 | |
| 224 | spinlock_unlock(&df->users.spinlock); |
| 225 | |
| 226 | return can_be_deleted; |
| 227 | } |
| 228 | |
| 229 | void generate_datafilepath(struct rrdengine_datafile *datafile, char *str, size_t maxlen) |
| 230 | { |
| 231 | (void) snprintfz(str, maxlen - 1, "%s/" DATAFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL DATAFILE_EXTENSION, |
| 232 | datafile_ctx(datafile)->config.dbfiles_path, datafile->tier, datafile->fileno); |
| 233 | } |
| 234 | |
| 235 | int close_data_file(struct rrdengine_datafile *datafile) |
| 236 | { |
| 237 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 238 | int ret; |
| 239 | char path[RRDENG_PATH_MAX]; |
| 240 | generate_datafilepath(datafile, path, sizeof(path)); |
| 241 | CLOSE_FILE(ctx, path, datafile->file, ret); |
| 242 | return ret; |
| 243 | } |
| 244 | |
| 245 | int unlink_data_file(struct rrdengine_datafile *datafile) |
| 246 | { |
| 247 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 248 | int ret; |
| 249 | char path[RRDENG_PATH_MAX]; |
| 250 | |
| 251 | generate_datafilepath(datafile, path, sizeof(path)); |
| 252 | |
| 253 | UNLINK_FILE(ctx, path, ret); |
| 254 | if (ret == 0) |
| 255 | __atomic_add_fetch(&ctx->stats.datafile_deletions, 1, __ATOMIC_RELAXED); |
| 256 | |
| 257 | return ret; |
| 258 | } |
| 259 | |
| 260 | int destroy_data_file_unsafe(struct rrdengine_datafile *datafile) |
| 261 | { |
| 262 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 263 | int ret; |
| 264 | char path[RRDENG_PATH_MAX]; |
| 265 | |
| 266 | generate_datafilepath(datafile, path, sizeof(path)); |
| 267 | |
| 268 | CLOSE_FILE(ctx, path, datafile->file, ret); |
| 269 | ret = unlink_data_file(datafile); |
| 270 | |
| 271 | return ret; |
| 272 | } |
| 273 | |
| 274 | int create_data_file(struct rrdengine_datafile *datafile) |
| 275 | { |
| 276 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 277 | uv_fs_t req; |
| 278 | uv_file file; |
| 279 | int ret, fd; |
| 280 | struct rrdeng_df_sb *superblock = NULL; |
| 281 | uv_buf_t iov; |
| 282 | char path[RRDENG_PATH_MAX]; |
| 283 | |
| 284 | generate_datafilepath(datafile, path, sizeof(path)); |
| 285 | fd = open_file_for_io(path, O_CREAT | O_RDWR | O_TRUNC, &file, dbengine_use_direct_io); |
| 286 | if (fd < 0) { |
| 287 | ctx_fs_error(ctx); |
| 288 | return fd; |
| 289 | } |
| 290 | datafile->file = file; |
| 291 | |
| 292 | (void)posix_memalignz((void *)&superblock, RRDFILE_ALIGNMENT, sizeof(*superblock)); |
| 293 | memset(superblock, 0, sizeof(*superblock)); |
| 294 | (void) strncpy(superblock->magic_number, RRDENG_DF_MAGIC, RRDENG_MAGIC_SZ); |
| 295 | (void) strncpy(superblock->version, RRDENG_DF_VER, RRDENG_VER_SZ); |
| 296 | superblock->tier = 1; |
| 297 | |
| 298 | iov = uv_buf_init((void *)superblock, sizeof(*superblock)); |
| 299 | |
| 300 | int retries = 10; |
| 301 | ret = -1; |
| 302 | while (ret < 0 && --retries) { |
| 303 | ret = uv_fs_write(NULL, &req, file, &iov, 1, 0, NULL); |
| 304 | uv_fs_req_cleanup(&req); |
| 305 | if (ret < 0) { |
| 306 | if (ret == -ENOSPC || ret == -EBADF || ret == -EACCES || ret == -EROFS || ret == -EINVAL) |
| 307 | break; |
| 308 | sleep_usec(300 * USEC_PER_MS); |
| 309 | } |
| 310 | } |
| 311 | |
| 312 | posix_memalign_freez(superblock); |
| 313 | if (ret < 0) { |
| 314 | (void) destroy_data_file_unsafe(datafile); |
| 315 | ctx_io_error(ctx); |
| 316 | nd_log_limit_static_global_var(dbengine_erl, 10, 0); |
| 317 | nd_log_limit(&dbengine_erl, NDLS_DAEMON, NDLP_ERR, "DBENGINE: Failed to create datafile %s", path); |
| 318 | return ret; |
| 319 | } |
| 320 | |
| 321 | __atomic_add_fetch(&ctx->stats.datafile_creations, 1, __ATOMIC_RELAXED); |
| 322 | datafile->pos = sizeof(*superblock); |
| 323 | ctx_io_write_op_bytes(ctx, sizeof(*superblock)); |
| 324 | |
| 325 | return 0; |
| 326 | } |
| 327 | |
| 328 | static int check_data_file_superblock(uv_file file) |
| 329 | { |
| 330 | int ret; |
| 331 | struct rrdeng_df_sb *superblock = NULL; |
| 332 | uv_buf_t iov; |
| 333 | uv_fs_t req; |
| 334 | |
| 335 | (void)posix_memalignz((void *)&superblock, RRDFILE_ALIGNMENT, sizeof(*superblock)); |
| 336 | iov = uv_buf_init((void *)superblock, sizeof(*superblock)); |
| 337 | |
| 338 | ret = uv_fs_read(NULL, &req, file, &iov, 1, 0, NULL); |
| 339 | if (ret < 0) { |
| 340 | netdata_log_error("DBENGINE: uv_fs_read: %s", uv_strerror(ret)); |
| 341 | uv_fs_req_cleanup(&req); |
| 342 | goto error; |
| 343 | } |
| 344 | fatal_assert(req.result >= 0); |
| 345 | uv_fs_req_cleanup(&req); |
| 346 | |
| 347 | if (strncmp(superblock->magic_number, RRDENG_DF_MAGIC, RRDENG_MAGIC_SZ) || |
| 348 | strncmp(superblock->version, RRDENG_DF_VER, RRDENG_VER_SZ) || |
| 349 | superblock->tier != 1) { |
| 350 | netdata_log_error("DBENGINE: file has invalid superblock."); |
| 351 | ret = UV_EINVAL; |
| 352 | } else { |
| 353 | ret = 0; |
| 354 | } |
| 355 | error: |
| 356 | posix_memalign_freez(superblock); |
| 357 | return ret; |
| 358 | } |
| 359 | |
| 360 | static int load_data_file(struct rrdengine_datafile *datafile) |
| 361 | { |
| 362 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 363 | uv_file file; |
| 364 | int ret, fd, error; |
| 365 | uint64_t file_size; |
| 366 | char path[RRDENG_PATH_MAX]; |
| 367 | |
| 368 | generate_datafilepath(datafile, path, sizeof(path)); |
| 369 | fd = open_file_for_io(path, O_RDWR, &file, dbengine_use_direct_io); |
| 370 | if (fd < 0) { |
| 371 | ctx_fs_error(ctx); |
| 372 | return fd; |
| 373 | } |
| 374 | |
| 375 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: initializing data file \"%s\".", path); |
| 376 | |
| 377 | ret = check_file_properties(file, &file_size, sizeof(struct rrdeng_df_sb)); |
| 378 | if (ret) |
| 379 | goto err_exit; |
| 380 | file_size = ALIGN_BYTES_CEILING(file_size); |
| 381 | |
| 382 | ret = check_data_file_superblock(file); |
| 383 | if (ret) |
| 384 | goto err_exit; |
| 385 | |
| 386 | ctx_io_read_op_bytes(ctx, sizeof(struct rrdeng_df_sb)); |
| 387 | |
| 388 | datafile->file = file; |
| 389 | datafile->pos = file_size; |
| 390 | |
| 391 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: data file \"%s\" initialized (size:%" PRIu64 ").", path, file_size); |
| 392 | |
| 393 | return 0; |
| 394 | |
| 395 | err_exit: |
| 396 | error = ret; |
| 397 | CLOSE_FILE(ctx, path, file, ret); |
| 398 | return error; |
| 399 | } |
| 400 | |
| 401 | static int scan_data_files_cmp(const void *a, const void *b) |
| 402 | { |
| 403 | struct rrdengine_datafile *file1, *file2; |
| 404 | char path1[RRDENG_PATH_MAX], path2[RRDENG_PATH_MAX]; |
| 405 | |
| 406 | file1 = *(struct rrdengine_datafile **)a; |
| 407 | file2 = *(struct rrdengine_datafile **)b; |
| 408 | generate_datafilepath(file1, path1, sizeof(path1)); |
| 409 | generate_datafilepath(file2, path2, sizeof(path2)); |
| 410 | return strcmp(path1, path2); |
| 411 | } |
| 412 | |
| 413 | /* Returns number of datafiles that were loaded or < 0 on error */ |
| 414 | static int scan_data_files(struct rrdengine_instance *ctx) |
| 415 | { |
| 416 | int ret, matched_files, failed_to_load, i; |
| 417 | unsigned tier, fileno; |
| 418 | uv_fs_t req; |
| 419 | uv_dirent_t dent; |
| 420 | struct rrdengine_datafile **datafiles, *datafile; |
| 421 | struct rrdengine_journalfile *journalfile; |
| 422 | |
| 423 | ret = uv_fs_scandir(NULL, &req, ctx->config.dbfiles_path, 0, NULL); |
| 424 | if (ret < 0) { |
| 425 | fatal_assert(req.result < 0); |
| 426 | uv_fs_req_cleanup(&req); |
| 427 | netdata_log_error("DBENGINE: uv_fs_scandir(%s): %s", ctx->config.dbfiles_path, uv_strerror(ret)); |
| 428 | ctx_fs_error(ctx); |
| 429 | return ret; |
| 430 | } |
| 431 | netdata_log_info("DBENGINE: tier %d: found %d files in path %s", ctx->config.tier, ret, ctx->config.dbfiles_path); |
| 432 | |
| 433 | Pvoid_t datafiles_JudyL = NULL; |
| 434 | Pvoid_t journafile_JudyL = NULL; |
| 435 | datafiles = callocz(MIN(ret, MAX_DATAFILES), sizeof(*datafiles)); |
| 436 | bool validate_files = true; |
| 437 | for (matched_files = 0 ; UV_EOF != uv_fs_scandir_next(&req, &dent) && matched_files < MAX_DATAFILES ; ) { |
| 438 | ret = sscanf(dent.name, DATAFILE_PREFIX RRDENG_FILE_NUMBER_SCAN_TMPL DATAFILE_EXTENSION, &tier, &fileno); |
| 439 | |
| 440 | // This is a datafile |
| 441 | if (2 == ret) { |
| 442 | datafile = datafile_alloc_and_init(ctx, tier, fileno); |
| 443 | datafiles[matched_files++] = datafile; |
| 444 | Pvoid_t *Pvalue = JudyLIns(&datafiles_JudyL, (Word_t)fileno, PJE0); |
| 445 | if (!Pvalue || Pvalue == PJERR) |
| 446 | validate_files = false; |
| 447 | continue; |
| 448 | } |
| 449 | |
| 450 | // Check for journal v1 or v2 |
| 451 | char expected_name[RRDENG_PATH_MAX]; |
| 452 | ret = sscanf(dent.name, WALFILE_PREFIX RRDENG_FILE_NUMBER_SCAN_TMPL WALFILE_EXTENSION, &tier, &fileno); |
| 453 | bool unknown_file = true; |
| 454 | if (2 == ret) { |
| 455 | (void) snprintfz(expected_name, sizeof(expected_name), WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION, |
| 456 | 1U, fileno); |
| 457 | |
| 458 | unknown_file = (strcmp(dent.name, expected_name) != 0); |
| 459 | if (unknown_file) { |
| 460 | (void) snprintfz(expected_name, sizeof(expected_name), WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION_V2, |
| 461 | 1U, fileno); |
| 462 | unknown_file = (strcmp(dent.name, expected_name) != 0); |
| 463 | } |
| 464 | |
| 465 | if (!unknown_file) |
| 466 | (void) JudyLIns(&journafile_JudyL, (Word_t)fileno, PJE0); |
| 467 | } |
| 468 | |
| 469 | if (unknown_file) |
| 470 | nd_log_daemon(NDLP_WARNING, "Unknown file detected : \"%s/%s\"", ctx->config.dbfiles_path, dent.name); |
| 471 | } |
| 472 | uv_fs_req_cleanup(&req); |
| 473 | |
| 474 | if (0 == matched_files) { |
| 475 | freez(datafiles); |
| 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | if (matched_files == MAX_DATAFILES) |
| 480 | netdata_log_error("DBENGINE: warning: hit maximum database engine file limit of %d files", MAX_DATAFILES); |
| 481 | |
| 482 | qsort(datafiles, matched_files, sizeof(*datafiles), scan_data_files_cmp); |
| 483 | |
| 484 | ctx->atomic.last_fileno = datafiles[matched_files - 1]->fileno; |
| 485 | |
| 486 | // Remove journal files that do not have a matching data file |
| 487 | // by scanning the judy array of the journal files |
| 488 | if (validate_files) { |
| 489 | bool first_then_next = true; |
| 490 | Word_t idx = 0; |
| 491 | Pvoid_t *PValue; |
| 492 | size_t deleted_journals = 0; |
| 493 | while ((PValue = JudyLFirstThenNext(journafile_JudyL, &idx, &first_then_next))) { |
| 494 | char path[RRDENG_PATH_MAX]; |
| 495 | if (unlikely(!JudyLGet(datafiles_JudyL, (Word_t)idx, PJE0))) { |
| 496 | (void)snprintfz( |
| 497 | path, |
| 498 | sizeof(path), |
| 499 | "%s/" WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION, |
| 500 | datafile_ctx(datafile)->config.dbfiles_path, |
| 501 | 1U, |
| 502 | (unsigned)idx); |
| 503 | |
| 504 | UNLINK_FILE(ctx, path, ret); |
| 505 | if (ret == 0) { |
| 506 | netdata_log_info("DBENGINE: deleting journal file without matching data file: %s", path); |
| 507 | __atomic_add_fetch(&ctx->stats.journalfile_deletions, 1, __ATOMIC_RELAXED); |
| 508 | deleted_journals++; |
| 509 | } |
| 510 | |
| 511 | (void)snprintfz( |
| 512 | path, |
| 513 | sizeof(path), |
| 514 | "%s/" WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION_V2, |
| 515 | datafile_ctx(datafile)->config.dbfiles_path, |
| 516 | 1U, |
| 517 | (unsigned)idx); |
| 518 | |
| 519 | UNLINK_FILE(ctx, path, ret); |
| 520 | if (ret == 0) { |
| 521 | netdata_log_info("DBENGINE: deleting journal file without matching data file: %s", path); |
| 522 | __atomic_add_fetch(&ctx->stats.journalfile_deletions, 1, __ATOMIC_RELAXED); |
| 523 | deleted_journals++; |
| 524 | } |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | if (deleted_journals) |
| 529 | netdata_log_info("DBENGINE: deleted %zu journal files without matching data files", deleted_journals); |
| 530 | } |
| 531 | |
| 532 | (void) JudyLFreeArray(&journafile_JudyL, NULL); |
| 533 | (void) JudyLFreeArray(&datafiles_JudyL, NULL); |
| 534 | |
| 535 | |
| 536 | netdata_log_info("DBENGINE: tier %d: loading %d data/journal files...", ctx->config.tier, matched_files); |
| 537 | for (failed_to_load = 0, i = 0 ; i < matched_files ; ++i) { |
| 538 | uint8_t must_delete_pair = 0; |
| 539 | |
| 540 | datafile = datafiles[i]; |
| 541 | ret = load_data_file(datafile); |
| 542 | if (0 != ret) |
| 543 | must_delete_pair = 1; |
| 544 | |
| 545 | journalfile = journalfile_alloc_and_init(datafile); |
| 546 | ret = journalfile_load(ctx, journalfile, datafile); |
| 547 | if (0 != ret) { |
| 548 | if (!must_delete_pair) /* If datafile is still open close it */ |
| 549 | close_data_file(datafile); |
| 550 | must_delete_pair = 1; |
| 551 | } |
| 552 | |
| 553 | if (must_delete_pair) { |
| 554 | char path[RRDENG_PATH_MAX]; |
| 555 | |
| 556 | netdata_log_error("DBENGINE: deleting invalid data and journal file pair."); |
| 557 | ret = journalfile_unlink(journalfile); |
| 558 | if (!ret) { |
| 559 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 560 | netdata_log_info("DBENGINE: deleted journal file \"%s\".", path); |
| 561 | } |
| 562 | ret = unlink_data_file(datafile); |
| 563 | if (!ret) { |
| 564 | generate_datafilepath(datafile, path, sizeof(path)); |
| 565 | netdata_log_info("DBENGINE: deleted data file \"%s\".", path); |
| 566 | } |
| 567 | freez(journalfile); |
| 568 | freez(datafile); |
| 569 | ++failed_to_load; |
| 570 | continue; |
| 571 | } |
| 572 | |
| 573 | ctx_current_disk_space_increase(ctx, datafile->pos + journalfile->unsafe.pos); |
| 574 | datafile_list_insert(ctx, datafile); |
| 575 | } |
| 576 | |
| 577 | matched_files -= failed_to_load; |
| 578 | freez(datafiles); |
| 579 | |
| 580 | return matched_files; |
| 581 | } |
| 582 | |
| 583 | /* Creates a datafile and a journalfile pair */ |
| 584 | int create_new_datafile_pair(struct rrdengine_instance *ctx) |
| 585 | { |
| 586 | __atomic_add_fetch(&rrdeng_cache_efficiency_stats.datafile_creation_started, 1, __ATOMIC_RELAXED); |
| 587 | |
| 588 | struct rrdengine_datafile *datafile; |
| 589 | struct rrdengine_journalfile *journalfile; |
| 590 | unsigned fileno = ctx_last_fileno_get(ctx) + 1; |
| 591 | int ret; |
| 592 | |
| 593 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 594 | "DBENGINE: creating new data and journal files in path \"%s\"", |
| 595 | ctx->config.dbfiles_path); |
| 596 | |
| 597 | datafile = datafile_alloc_and_init(ctx, 1, fileno); |
| 598 | ret = create_data_file(datafile); |
| 599 | if(ret) |
| 600 | goto error_after_datafile; |
| 601 | |
| 602 | journalfile = journalfile_alloc_and_init(datafile); |
| 603 | ret = journalfile_create(journalfile, datafile); |
| 604 | if (ret) |
| 605 | goto error_after_journalfile; |
| 606 | |
| 607 | nd_log(NDLS_DAEMON, NDLP_INFO, |
| 608 | "DBENGINE: tier %d: created " DATAFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL " (.ndf, .njf).", |
| 609 | ctx->config.tier, datafile->tier, datafile->fileno); |
| 610 | |
| 611 | ctx_current_disk_space_increase(ctx, datafile->pos + journalfile->unsafe.pos); |
| 612 | datafile_list_insert(ctx, datafile); |
| 613 | ctx_last_fileno_increment(ctx); |
| 614 | |
| 615 | return 0; |
| 616 | |
| 617 | error_after_journalfile: |
| 618 | (void) destroy_data_file_unsafe(datafile); |
| 619 | freez(journalfile); |
| 620 | |
| 621 | error_after_datafile: |
| 622 | freez(datafile); |
| 623 | return ret; |
| 624 | } |
| 625 | |
| 626 | /* Page cache must already be initialized. |
| 627 | * Return 0 on success. |
| 628 | */ |
| 629 | int init_data_files(struct rrdengine_instance *ctx) |
| 630 | { |
| 631 | int ret; |
| 632 | |
| 633 | ret = scan_data_files(ctx); |
| 634 | if (ret < 0) { |
| 635 | netdata_log_error("DBENGINE: failed to scan path \"%s\".", ctx->config.dbfiles_path); |
| 636 | return ret; |
| 637 | } else if (0 == ret) { |
| 638 | netdata_log_info("DBENGINE: data files not found, creating in path \"%s\".", ctx->config.dbfiles_path); |
| 639 | ctx->atomic.last_fileno = 0; |
| 640 | ret = create_new_datafile_pair(ctx); |
| 641 | if (ret) { |
| 642 | netdata_log_error("DBENGINE: failed to create data and journal files in path \"%s\".", ctx->config.dbfiles_path); |
| 643 | return ret; |
| 644 | } |
| 645 | } |
| 646 | else { |
| 647 | if (ctx->loading.create_new_datafile_pair) |
| 648 | create_new_datafile_pair(ctx); |
| 649 | } |
| 650 | |
| 651 | pgc_reset_hot_max(open_cache); |
| 652 | ctx->loading.create_new_datafile_pair = false; |
| 653 | return 0; |
| 654 | } |
| 655 | |
| 656 | void cleanup_datafile_epdl_structures(struct rrdengine_datafile *datafile) |
| 657 | { |
| 658 | rw_spinlock_write_lock(&datafile->extent_epdl.spinlock); |
| 659 | bool first = true; |
| 660 | Word_t idx = 0; |
| 661 | Pvoid_t *PValue; |
| 662 | while ((PValue = JudyLFirstThenNext(datafile->extent_epdl.epdl_per_extent, &idx, &first))) { |
| 663 | EPDL_EXTENT *e = *PValue; |
| 664 | internal_error(e->base, "DBENGINE: unexpected active EPDLs during datafile cleanup"); |
| 665 | epdl_extent_release(e); |
| 666 | *PValue = NULL; |
| 667 | } |
| 668 | JudyLFreeArray(&datafile->extent_epdl.epdl_per_extent, PJE0); |
| 669 | rw_spinlock_write_unlock(&datafile->extent_epdl.spinlock); |
| 670 | } |
| 671 | |
| 672 | void finalize_data_files(struct rrdengine_instance *ctx) |
| 673 | { |
| 674 | bool logged = false; |
| 675 | |
| 676 | if (!ctx->datafiles.JudyL) |
| 677 | return; |
| 678 | |
| 679 | while(__atomic_load_n(&ctx->atomic.extents_currently_being_flushed, __ATOMIC_RELAXED)) { |
| 680 | if(!logged) { |
| 681 | netdata_log_info("Waiting for inflight flush to finish on tier %d...", ctx->config.tier); |
| 682 | logged = true; |
| 683 | } |
| 684 | sleep_usec(100 * USEC_PER_MS); |
| 685 | } |
| 686 | |
| 687 | bool first_then_next = true; |
| 688 | Pvoid_t *PValue; |
| 689 | Word_t Index = 0; |
| 690 | |
| 691 | while ((PValue = JudyLFirstThenNext(ctx->datafiles.JudyL, &Index, &first_then_next))) { |
| 692 | struct rrdengine_datafile *datafile = *PValue; |
| 693 | struct rrdengine_journalfile *journalfile = datafile->journalfile; |
| 694 | |
| 695 | logged = false; |
| 696 | size_t iterations = 10; |
| 697 | while(!datafile_acquire_for_deletion(datafile, true) && --iterations > 0) { |
| 698 | if(!logged) { |
| 699 | netdata_log_info("Waiting to acquire data file %u of tier %d to close it...", datafile->fileno, ctx->config.tier); |
| 700 | logged = true; |
| 701 | } |
| 702 | sleep_usec(100 * USEC_PER_MS); |
| 703 | } |
| 704 | |
| 705 | logged = false; |
| 706 | bool available = false; |
| 707 | do { |
| 708 | netdata_rwlock_wrlock(&ctx->datafiles.rwlock); |
| 709 | spinlock_lock(&datafile->writers.spinlock); |
| 710 | available = (datafile->writers.running || datafile->writers.flushed_to_open_running) ? false : true; |
| 711 | |
| 712 | if(!available) { |
| 713 | spinlock_unlock(&datafile->writers.spinlock); |
| 714 | netdata_rwlock_wrunlock(&ctx->datafiles.rwlock); |
| 715 | if(!logged) { |
| 716 | netdata_log_info("Waiting for writers to data file %u of tier %d to finish...", datafile->fileno, ctx->config.tier); |
| 717 | logged = true; |
| 718 | } |
| 719 | sleep_usec(100 * USEC_PER_MS); |
| 720 | } |
| 721 | } while(!available); |
| 722 | |
| 723 | journalfile_close(journalfile, datafile); |
| 724 | close_data_file(datafile); |
| 725 | datafile_list_delete_unsafe(ctx, datafile); |
| 726 | spinlock_unlock(&datafile->writers.spinlock); |
| 727 | netdata_rwlock_wrunlock(&ctx->datafiles.rwlock); |
| 728 | |
| 729 | // Clean up EPDL_EXTENT structures |
| 730 | cleanup_datafile_epdl_structures(datafile); |
| 731 | |
| 732 | memset(journalfile, 0, sizeof(*journalfile)); |
| 733 | memset(datafile, 0, sizeof(*datafile)); |
| 734 | |
| 735 | freez(journalfile); |
| 736 | freez(datafile); |
| 737 | } |
| 738 | } |