diff options
author | Markos Fountoulakis <44345837+mfundul@users.noreply.github.com> | 2019-09-24 11:59:15 +0300 |
---|---|---|
committer | GitHub <noreply@github.com> | 2019-09-24 11:59:15 +0300 |
commit | 2728be8b0614fd58d4b530a32fee164509e88fd8 (patch) | |
tree | abd39375b249b832327e35880597f9974be81b22 /database/engine | |
parent | 7977137cee3274d2535fbdb49ec2d68941285b82 (diff) |
Detect deadlock in dbengine page cache (#6911)
* Detect deadlock in dbengine page cache when there are too many metrics and print error message
* Resolve dbengine deadlock by dropping metrics when page cache is too small and define relevant alarms
* Changed printing deadlock errors to only happen once per dbengine instance
Diffstat (limited to 'database/engine')
-rw-r--r-- | database/engine/rrdengine.c | 2 | ||||
-rw-r--r-- | database/engine/rrdengine.h | 13 | ||||
-rw-r--r-- | database/engine/rrdengineapi.c | 30 | ||||
-rw-r--r-- | database/engine/rrdengineapi.h | 4 |
4 files changed, 41 insertions, 8 deletions
diff --git a/database/engine/rrdengine.c b/database/engine/rrdengine.c index 36d917541d..42f59b59f3 100644 --- a/database/engine/rrdengine.c +++ b/database/engine/rrdengine.c @@ -5,6 +5,8 @@ rrdeng_stats_t global_io_errors = 0; rrdeng_stats_t global_fs_errors = 0; +rrdeng_stats_t global_pg_cache_warnings = 0; +rrdeng_stats_t global_pg_cache_errors = 0; rrdeng_stats_t rrdeng_reserved_file_descriptors = 0; void sanity_check(void) diff --git a/database/engine/rrdengine.h b/database/engine/rrdengine.h index 6f6a6f8ffd..6447a685bd 100644 --- a/database/engine/rrdengine.h +++ b/database/engine/rrdengine.h @@ -148,12 +148,25 @@ struct rrdengine_statistics { rrdeng_stats_t page_cache_descriptors; rrdeng_stats_t io_errors; rrdeng_stats_t fs_errors; + rrdeng_stats_t pg_cache_warnings; + rrdeng_stats_t pg_cache_errors; }; /* I/O errors global counter */ extern rrdeng_stats_t global_io_errors; /* File-System errors global counter */ extern rrdeng_stats_t global_fs_errors; +/* + * Page cache warnings global counter. + * Some page cache instance is near critical utilization where metrics will fail to be stored. + */ +extern rrdeng_stats_t global_pg_cache_warnings; +/* + * Page cache errors global counter. + * Some page cache instance has hit critical utilization where metrics failed to be stored as a deadlock resolution + * measure. + */ +extern rrdeng_stats_t global_pg_cache_errors; /* number of File-Descriptors that have been reserved by dbengine */ extern rrdeng_stats_t rrdeng_reserved_file_descriptors; diff --git a/database/engine/rrdengineapi.c b/database/engine/rrdengineapi.c index bf373f31c7..26fbe8f7ef 100644 --- a/database/engine/rrdengineapi.c +++ b/database/engine/rrdengineapi.c @@ -95,9 +95,8 @@ void rrdeng_store_metric_flush_current_page(RRDDIM *rd) if (likely(descr->page_length)) { int ret, page_is_empty; -#ifdef NETDATA_INTERNAL_CHECKS rrd_stat_atomic_add(&ctx->stats.metric_API_producers, -1); -#endif + if (handle->prev_descr) { /* unpin old second page */ pg_cache_put(ctx, handle->prev_descr); @@ -192,9 +191,26 @@ void rrdeng_store_metric_next(RRDDIM *rd, usec_t point_in_time, storage_number n if (unlikely(INVALID_TIME == descr->start_time)) { descr->start_time = point_in_time; -#ifdef NETDATA_INTERNAL_CHECKS rrd_stat_atomic_add(&ctx->stats.metric_API_producers, 1); -#endif + + if (unlikely(((unsigned long)ctx->stats.metric_API_producers) >= ctx->max_cache_pages)) { + if (0 == (unsigned long)ctx->stats.pg_cache_errors) { + /* only print the first time */ + error("Deadlock detected in dbengine instance \"%s\", metric data will not be stored in the database" + ", please increase page cache size.", ctx->dbfiles_path); + } + rrd_stat_atomic_add(&ctx->stats.pg_cache_errors, 1); + rrd_stat_atomic_add(&global_pg_cache_errors, 1); + /* Resolve deadlock */ + descr->page_length = 0; /* make sure the page descriptor is deconstructed */ + rrdeng_store_metric_flush_current_page(rd); + rrd_stat_atomic_add(&ctx->stats.metric_API_producers, -1); + return; + } else if (unlikely(((unsigned long)ctx->stats.metric_API_producers) >= ctx->cache_pages_low_watermark)) { + rrd_stat_atomic_add(&ctx->stats.pg_cache_warnings, 1); + rrd_stat_atomic_add(&global_pg_cache_warnings, 1); + } + pg_cache_insert(ctx, handle->page_index, descr); } else { pg_cache_add_new_metric_time(handle->page_index, descr); @@ -672,7 +688,7 @@ void *rrdeng_get_page(struct rrdengine_instance *ctx, uuid_t *id, usec_t point_i * You must not change the indices of the statistics or user code will break. * You must not exceed RRDENG_NR_STATS or it will crash. */ -void rrdeng_get_33_statistics(struct rrdengine_instance *ctx, unsigned long long *array) +void rrdeng_get_35_statistics(struct rrdengine_instance *ctx, unsigned long long *array) { struct page_cache *pg_cache = &ctx->pg_cache; @@ -709,7 +725,9 @@ void rrdeng_get_33_statistics(struct rrdengine_instance *ctx, unsigned long long array[30] = (uint64_t)global_io_errors; array[31] = (uint64_t)global_fs_errors; array[32] = (uint64_t)rrdeng_reserved_file_descriptors; - assert(RRDENG_NR_STATS == 33); + array[33] = (uint64_t)global_pg_cache_warnings; + array[34] = (uint64_t)global_pg_cache_errors; + assert(RRDENG_NR_STATS == 35); } /* Releases reference to page */ diff --git a/database/engine/rrdengineapi.h b/database/engine/rrdengineapi.h index 9b1ab18742..97e358199e 100644 --- a/database/engine/rrdengineapi.h +++ b/database/engine/rrdengineapi.h @@ -8,7 +8,7 @@ #define RRDENG_MIN_PAGE_CACHE_SIZE_MB (32) #define RRDENG_MIN_DISK_SPACE_MB (256) -#define RRDENG_NR_STATS (33) +#define RRDENG_NR_STATS (35) #define RRDENG_FD_BUDGET_PER_INSTANCE (50) @@ -41,7 +41,7 @@ extern int rrdeng_load_metric_is_finished(struct rrddim_query_handle *rrdimm_han extern void rrdeng_load_metric_finalize(struct rrddim_query_handle *rrdimm_handle); extern time_t rrdeng_metric_latest_time(RRDDIM *rd); extern time_t rrdeng_metric_oldest_time(RRDDIM *rd); -extern void rrdeng_get_33_statistics(struct rrdengine_instance *ctx, unsigned long long *array); +extern void rrdeng_get_35_statistics(struct rrdengine_instance *ctx, unsigned long long *array); /* must call once before using anything */ extern int rrdeng_init(struct rrdengine_instance **ctxp, char *dbfiles_path, unsigned page_cache_mb, |