@cryptotaxi247 / netdata-1 / commits / 936832d57

Handle journal_v2 file creation failure due to OOM (#19965)

Stelios Fragkakis committed Apr 3, 2025 at 11:08 UTC 936832d5777e4b8c24c24d2987493929b88918aa

4 files changed +18 -12

src/database/engine/cache.c

+7 -5

@@ -2552,7 +2552,7 @@ void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_

2552 pgc_queue_unlock(cache, &cache->hot);

2553

2554 // callback

2555 - cb(section, datafile_fileno, type, JudyL_metrics, JudyL_extents_pos, count_of_unique_extents, count_of_unique_metrics, count_of_unique_pages, data);

2555 + bool success = cb(section, datafile_fileno, type, JudyL_metrics, JudyL_extents_pos, count_of_unique_extents, count_of_unique_metrics, count_of_unique_pages, data);

2556

2557 {

2558 Pvoid_t *PValue1;

@@ -2567,12 +2567,14 @@ void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_

2567 while ((PValue2 = JudyLFirstThenNext(mi->JudyL_pages_by_start_time, &start_time, &start_time_first))) {

2568 struct jv2_page_info *pi = *PValue2;

2569

2570 - // balance-parents: transition from hot to clean directly

2570 yield_the_processor(); // do not lock too aggressively

2572 - page_set_clean(cache, pi->page, true, false, PGC_QUEUE_LOCK_PRIO_LOW);

2573 - page_transition_unlock(cache, pi->page);

2574 - page_release(cache, pi->page, true);

2571 + if (likely(success))

2572 + page_set_clean(cache, pi->page, true, false, PGC_QUEUE_LOCK_PRIO_LOW);

2573 + else

2574 + page_flag_clear(pi->page, PGC_PAGE_IS_BEING_MIGRATED_TO_V2);

2575

2576 + page_transition_unlock(cache, pi->page);

2577 + page_release(cache, pi->page, success);

2578 // before balance-parents:

2579 // page_transition_unlock(cache, pi->page);

2580 // pgc_page_hot_to_dirty_and_release(cache, pi->page, true);

src/database/engine/cache.h

+1 -1

@@ -227,7 +227,7 @@ int64_t pgc_get_wanted_cache_size(PGC *cache);

227 void pgc_page_hot_set_end_time_s(PGC *cache, PGC_PAGE *page, time_t end_time_s, size_t additional_bytes);

228 bool pgc_page_to_clean_evict_or_release(PGC *cache, PGC_PAGE *page);

229

230 -typedef void (*migrate_to_v2_callback)(Word_t section, unsigned datafile_fileno, uint8_t type, Pvoid_t JudyL_metrics, Pvoid_t JudyL_extents_pos, size_t count_of_unique_extents, size_t count_of_unique_metrics, size_t count_of_unique_pages, void *data);

230 +typedef bool (*migrate_to_v2_callback)(Word_t section, unsigned datafile_fileno, uint8_t type, Pvoid_t JudyL_metrics, Pvoid_t JudyL_extents_pos, size_t count_of_unique_extents, size_t count_of_unique_metrics, size_t count_of_unique_pages, void *data);

231 void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_fileno, uint8_t type, migrate_to_v2_callback cb, void *data);

232 void pgc_open_evict_clean_pages_of_datafile(PGC *cache, struct rrdengine_datafile *datafile);

233 size_t pgc_count_clean_pages_having_data_ptr(PGC *cache, Word_t section, void *ptr);

src/database/engine/journalfile.c

+9 -5

@@ -1293,7 +1293,7 @@ static void *journalfile_v2_write_descriptors(struct journal_v2_header *j2_heade

1293 // startup : if the migration is done during agent startup

1294 // this will allow us to optimize certain things

1295

1296 -void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno __maybe_unused, uint8_t type __maybe_unused,

1296 +bool journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno __maybe_unused, uint8_t type __maybe_unused,

1297 Pvoid_t JudyL_metrics, Pvoid_t JudyL_extents_pos,

1298 size_t number_of_extents, size_t number_of_metrics, size_t number_of_pages, void *user_data)

1299 {

@@ -1346,8 +1346,10 @@ void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno

1346

1347 int fd_v2;

1348 uint8_t *data_start = nd_mmap_advanced(path, total_file_size, MAP_SHARED, 0, false, true, &fd_v2);

1349 - if(!data_start)

1350 - out_of_memory(__FUNCTION__, total_file_size, path);

1349 + if(!data_start) {

1350 + nd_log_daemon(NDLP_WARNING, "DBENGINE: Failed to allocate %"PRIu64" bytes of memory for journal file '%s'. Will retry later", total_file_size, path);

1351 + return false;

1352 + }

1353

1354 memset(data_start, 0, extent_offset);

1355

@@ -1492,7 +1494,7 @@ void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno

1494 internal_error(true, "DBENGINE: ACTIVATING NEW INDEX JNL %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS);

1495 ctx_current_disk_space_increase(ctx, total_file_size);

1496 freez(uuid_list);

1495 - return;

1497 + return true;

1498 }

1499 else {

1500 netdata_log_info("DBENGINE: failed to build index '%s', file will be skipped", path);

@@ -1506,7 +1508,7 @@ void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno

1508 freez(uuid_list);

1509

1510 if (likely(resize_file_to == total_file_size))

1509 - return;

1511 + return true;

1512

1513 int ret = truncate(path, (long) resize_file_to);

1514 if (ret < 0) {

@@ -1516,6 +1518,8 @@ void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno

1518 }

1519 else

1520 ctx_current_disk_space_increase(ctx, resize_file_to);

1521 +

1522 + return true;

1523 }

1524

1525 int journalfile_load(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile,

src/database/engine/journalfile.h

+1 -1

@@ -264,7 +264,7 @@ int journalfile_load(struct rrdengine_instance *ctx, struct rrdengine_journalfil

264 struct rrdengine_datafile *datafile);

265 void journalfile_v2_populate_retention_to_mrg(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile);

266

267 -void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno __maybe_unused, uint8_t type __maybe_unused,

267 +bool journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno __maybe_unused, uint8_t type __maybe_unused,

268 Pvoid_t JudyL_metrics, Pvoid_t JudyL_extents_pos,

269 size_t number_of_extents, size_t number_of_metrics, size_t number_of_pages, void *user_data);

270