@cryptotaxi247 / netdata-1 / commits / 101a1e673

Fix SIGSEGV in ctx_hosts_load during shutdown (#22419)

Fix SIGSEGV in ctx_hosts_load during shutdown The libuv worker running ctx_hosts_load only checked for shutdown between hosts, so a host with many dimensions kept the SQL row loops busy past the metadata thread's 15s wait. The daemon then ran sqlite3_shutdown() while the worker was still alive, and the worker's later sqlite3_close_v2() crashed in pcache1EnforceMaxPage on freed pcache1 globals. - Check exit_initiated_get() inside ctx_get_chart_list / ctx_get_dimension_list / ctx_get_context_list row loops, between the three list calls in rrdhost_load_rrdcontext_data, inside its outer trigger loop, and at the top of restore_host_context. - Route every per-thread sqlite3_close_v2 in the worker through a new sql_close_thread_db_safe() that takes sqlite_spinlock and skips the close once sqlite_library_initialized is false, serializing with sqlite_library_shutdown().

Stelios Fragkakis committed May 7, 2026 at 17:07 UTC 101a1e673c8a309c188e315ec6dade329d25b191

5 files changed +51 -19

src/database/contexts/rrdcontext-loading.c

+11

@@ -156,8 +156,16 @@ void rrdhost_load_rrdcontext_data(RRDHOST *host) {

156 th_ignored_metrics = th_ignored_instances = th_zero_retention_metrics = 0;

157

158 ctx_get_context_list(&host->host_id.uuid, rrdcontext_load_context_callback, host);

159 + if (unlikely(exit_initiated_get()))

160 + return;

161 +

162 ctx_get_chart_list(&host->host_id.uuid, rrdinstance_load_instance_callback, host);

163 + if (unlikely(exit_initiated_get()))

164 + return;

165 +

166 ctx_get_dimension_list(&host->host_id.uuid, rrdinstance_load_dimension_callback, host);

167 + if (unlikely(exit_initiated_get()))

168 + return;

169

170 size_t ignored_metrics = th_ignored_metrics, ignored_instances = th_ignored_instances, zero_retention_metrics = th_zero_retention_metrics;

171 size_t loaded_metrics = 0, loaded_instances = 0, loaded_contexts = 0;

@@ -165,6 +173,9 @@ void rrdhost_load_rrdcontext_data(RRDHOST *host) {

173

174 RRDCONTEXT *rc;

175 dfe_start_read(host->rrdctx.contexts, rc) {

176 + if (unlikely(exit_initiated_get()))

177 + break;

178 +

179 size_t instances = 0;

180

181 RRDINSTANCE *ri;

src/database/sqlite/sqlite_context.c

@@ -104,6 +104,9 @@ void ctx_get_chart_list(nd_uuid_t *host_uuid, void (*dict_cb)(SQL_CHART_DATA *,

104 param = 0;

105 SQL_CHART_DATA chart_data = { 0 };

106 while (sqlite3_step_monitored(res) == SQLITE_ROW) {

107 + if (unlikely(exit_initiated_get()))

108 + break;

109 +

110 if (unlikely(!sqlite3_column_uuid_copy(res, 0, chart_data.chart_id))) {

111 error_report("CTX [%s]: Got invalid chart id in column 0. Ignoring it.", host_guid);

112 continue;

@@ -146,6 +149,9 @@ void ctx_get_dimension_list(nd_uuid_t *host_uuid, void (*dict_cb)(SQL_DIMENSION_

149

150 param = 0;

151 while (sqlite3_step_monitored(res) == SQLITE_ROW) {

152 + if (unlikely(exit_initiated_get()))

153 + break;

154 +

155 if (unlikely(!sqlite3_column_uuid_copy(res, 0, dimension_data.dim_id))) {

156 error_report("CTX [%s]: Got invalid dimension id in column 0. Ignoring it.", host_guid);

157 continue;

@@ -213,6 +219,9 @@ void ctx_get_context_list(nd_uuid_t *host_uuid, void (*dict_cb)(VERSIONED_CONTEX

219 param = 0;

220

221 while (sqlite3_step_monitored(res) == SQLITE_ROW) {

222 + if (unlikely(exit_initiated_get()))

223 + break;

224 +

225 context_data.id = (char *) sqlite3_column_text(res, 0);

226 context_data.version = sqlite3_column_int64(res, 1);

227 context_data.title = (char *) sqlite3_column_text(res, 2);

src/database/sqlite/sqlite_functions.c

+16

@@ -455,6 +455,22 @@ void sql_close_database(sqlite3 *database, const char *database_name)

455

456 extern sqlite3 *db_context_meta;

457

458 +// Close a thread-local sqlite3 handle while serializing against sqlite_library_shutdown().

459 +// If the SQLite library is no longer initialized, the handle is leaked deliberately; the OS

460 +// will reclaim it at process exit, which is strictly safer than crashing inside pcache1.

461 +void sql_close_thread_db_safe(sqlite3 **database)

462 +{

463 + if (unlikely(!database || !*database))

464 + return;

465 +

466 + spinlock_lock(&sqlite_spinlock);

467 + if (sqlite_library_initialized)

468 + (void) sqlite3_close_v2(*database);

469 + spinlock_unlock(&sqlite_spinlock);

470 +

471 + *database = NULL;

472 +}

473 +

474 void sqlite_close_databases(void)

475 {

476 // In case we have statements in the main thread (we should not)

src/database/sqlite/sqlite_functions.h

@@ -166,6 +166,7 @@ int sqlite_library_init(void);

166 void sqlite_library_shutdown(void);

167

168 void sql_close_database(sqlite3 *database, const char *database_name);

169 +void sql_close_thread_db_safe(sqlite3 **database);

170 void sqlite_close_databases(void);

171 uint64_t get_total_database_space(void);

172 int sqlite_release_memory(int bytes);

src/database/sqlite/sqlite_metadata.c

+14 -19

@@ -1895,6 +1895,11 @@ static void restore_host_context(void *arg)

1895 if (!host)

1896 return;

1897

1898 + if (unlikely(exit_initiated_get())) {

1899 + __atomic_store_n(&hclt->finished, true, __ATOMIC_RELEASE);

1900 + return;

1901 + }

1902 +

1903 if (!db_meta_thread) {

1904 if (hclt->db_meta_thread) {

1905 db_meta_thread = hclt->db_meta_thread;

@@ -1903,17 +1908,13 @@ static void restore_host_context(void *arg)

1908 char sqlite_database[FILENAME_MAX + 1];

1909 snprintfz(sqlite_database, sizeof(sqlite_database) - 1, "%s/netdata-meta.db", netdata_configured_cache_dir);

1910 int rc = sqlite3_open_v2(sqlite_database, &db_meta_thread, SQLITE_OPEN_READONLY | SQLITE_OPEN_NOMUTEX, NULL);

1906 - if (rc != SQLITE_OK) {

1907 - sqlite3_close_v2(db_meta_thread);

1908 - db_meta_thread = NULL;

1909 - }

1911 + if (rc != SQLITE_OK)

1912 + sql_close_thread_db_safe(&db_meta_thread);

1913

1914 snprintfz(sqlite_database, sizeof(sqlite_database) - 1, "%s/context-meta.db", netdata_configured_cache_dir);

1915 rc = sqlite3_open_v2(sqlite_database, &db_context_thread, SQLITE_OPEN_READONLY | SQLITE_OPEN_NOMUTEX, NULL);

1913 - if (rc != SQLITE_OK) {

1914 - sqlite3_close_v2(db_context_thread);

1915 - db_context_thread = NULL;

1916 - }

1916 + if (rc != SQLITE_OK)

1917 + sql_close_thread_db_safe(&db_context_thread);

1918

1919 hclt->db_meta_thread = db_meta_thread;

1920 hclt->db_context_thread = db_context_thread;

@@ -2069,11 +2070,8 @@ static void ctx_hosts_load(uv_work_t *req)

2070

2071 if (should_clean_threads) {

2072 for (size_t index = 0; index < max_threads; index++) {

2072 - if (hclt[index].db_meta_thread)

2073 - sqlite3_close_v2(hclt[index].db_meta_thread);

2074 -

2075 - if (hclt[index].db_context_thread)

2076 - sqlite3_close_v2(hclt[index].db_context_thread);

2073 + sql_close_thread_db_safe(&hclt[index].db_meta_thread);

2074 + sql_close_thread_db_safe(&hclt[index].db_context_thread);

2075 }

2076 }

2077

@@ -2090,12 +2088,9 @@ static void ctx_hosts_load(uv_work_t *req)

2088 sync_exec,

2089 load_duration);

2090

2093 - if (db_meta_thread) {

2094 - sqlite3_close_v2(db_meta_thread);

2095 - sqlite3_close_v2(db_context_thread);

2096 - db_meta_thread = NULL;

2097 - db_context_thread = NULL;

2098 - }

2091 + sql_close_thread_db_safe(&db_meta_thread);

2092 + sql_close_thread_db_safe(&db_context_thread);

2093 +

2094 freez(hclt);

2095 worker_is_idle();

2096 }