/*创建一个connection evict cache*/
int __wt_cache_create(WT_SESSION_IMPL* session, const char* cfg[])
{
WT_CACHE *cache;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
conn = S2C(session);
WT_RET(__wt_calloc_one(session, &conn->cache));
cache = conn->cache;
/*对cache进行配置*/
WT_RET(__wt_cache_config(session, 0, cfg));
if (cache->eviction_target >= cache->eviction_trigger)
WT_ERR_MSG(session, EINVAL, "eviction target must be lower than the eviction trigger");
/*创建evict cond信号量*/
WT_ERR(__wt_cond_alloc(session, "cache eviction server", 0, &cache->evict_cond));
WT_ERR(__wt_cond_alloc(session, "eviction waiters", 0, &cache->evict_waiter_cond));
WT_ERR(__wt_spin_init(session, &cache->evict_lock, "cache eviction"));
WT_ERR(__wt_spin_init(session, &cache->evict_walk_lock, "cache walk"));
/* Allocate the LRU eviction queue. */
cache->evict_slots = WT_EVICT_WALK_BASE + WT_EVICT_WALK_INCR;
WT_ERR(__wt_calloc_def(session, cache->evict_slots, &cache->evict));
/*初始化cache stat统计模块*/
__wt_cache_stats_update(session);
return 0;
err:
WT_RET(__wt_cache_destroy(session));
return ret;
}
/*
* __wt_logmgr_create --
* Initialize the log subsystem (before running recovery).
*/
int
__wt_logmgr_create(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CONNECTION_IMPL *conn;
WT_LOG *log;
bool run;
conn = S2C(session);
/* Handle configuration. */
WT_RET(__logmgr_config(session, cfg, &run, false));
/* If logging is not configured, we're done. */
if (!run)
return (0);
FLD_SET(conn->log_flags, WT_CONN_LOG_ENABLED);
/*
* Logging is on, allocate the WT_LOG structure and open the log file.
*/
WT_RET(__wt_calloc_one(session, &conn->log));
log = conn->log;
WT_RET(__wt_spin_init(session, &log->log_lock, "log"));
WT_RET(__wt_spin_init(session, &log->log_slot_lock, "log slot"));
WT_RET(__wt_spin_init(session, &log->log_sync_lock, "log sync"));
WT_RET(__wt_spin_init(session, &log->log_writelsn_lock,
"log write LSN"));
WT_RET(__wt_rwlock_alloc(session,
&log->log_archive_lock, "log archive lock"));
if (FLD_ISSET(conn->direct_io, WT_FILE_TYPE_LOG))
log->allocsize =
WT_MAX((uint32_t)conn->buffer_alignment, WT_LOG_ALIGN);
else
log->allocsize = WT_LOG_ALIGN;
WT_INIT_LSN(&log->alloc_lsn);
WT_INIT_LSN(&log->ckpt_lsn);
WT_INIT_LSN(&log->first_lsn);
WT_INIT_LSN(&log->sync_lsn);
/*
* We only use file numbers for directory sync, so this needs to
* initialized to zero.
*/
WT_ZERO_LSN(&log->sync_dir_lsn);
WT_INIT_LSN(&log->trunc_lsn);
WT_INIT_LSN(&log->write_lsn);
WT_INIT_LSN(&log->write_start_lsn);
log->fileid = 0;
WT_RET(__wt_cond_alloc(
session, "log sync", false, &log->log_sync_cond));
WT_RET(__wt_cond_alloc(
session, "log write", false, &log->log_write_cond));
WT_RET(__wt_log_open(session));
WT_RET(__wt_log_slot_init(session));
return (0);
}
/*
* __wt_rwlock_init --
* Initialize a read/write lock.
*/
int
__wt_rwlock_init(WT_SESSION_IMPL *session, WT_RWLOCK *l)
{
l->u.v = 0;
l->stat_read_count_off = l->stat_write_count_off = -1;
l->stat_app_usecs_off = l->stat_int_usecs_off = -1;
WT_RET(__wt_cond_alloc(session, "rwlock wait", &l->cond_readers));
WT_RET(__wt_cond_alloc(session, "rwlock wait", &l->cond_writers));
return (0);
}
/*
* __wt_sweep_create --
* Start the handle sweep thread.
*/
int
__wt_sweep_create(WT_SESSION_IMPL *session)
{
WT_CONNECTION_IMPL *conn;
conn = S2C(session);
/* Set first, the thread might run before we finish up. */
F_SET(conn, WT_CONN_SERVER_SWEEP);
WT_RET(__wt_open_internal_session(
conn, "sweep-server", 1, 1, &conn->sweep_session));
session = conn->sweep_session;
/*
* Handle sweep does enough I/O it may be called upon to perform slow
* operations for the block manager.
*/
F_SET(session, WT_SESSION_CAN_WAIT);
WT_RET(__wt_cond_alloc(
session, "handle sweep server", 0, &conn->sweep_cond));
WT_RET(__wt_thread_create(
session, &conn->sweep_tid, __sweep_server, session));
conn->sweep_tid_set = 1;
return (0);
}
/*
* __wt_checkpoint_create --
* Start the checkpoint server thread.
*/
int
__wt_checkpoint_create(WT_CONNECTION_IMPL *conn, const char *cfg[])
{
WT_SESSION_IMPL *session;
int run;
session = conn->default_session;
/* Handle configuration. */
WT_RET(__ckpt_server_config(session, cfg, &run));
/* If not configured, we're done. */
if (!run)
return (0);
/* The checkpoint server gets its own session. */
WT_RET(__wt_open_session(conn, 1, NULL, NULL, &conn->ckpt_session));
conn->ckpt_session->name = "checkpoint-server";
WT_RET(
__wt_cond_alloc(session, "checkpoint server", 0, &conn->ckpt_cond));
/*
* Start the thread.
*/
WT_RET(__wt_thread_create(
session, &conn->ckpt_tid, __ckpt_server, conn->ckpt_session));
conn->ckpt_tid_set = 1;
return (0);
}
/*
* __ckpt_server_start --
* Start the checkpoint server thread.
*/
static int
__ckpt_server_start(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
/* Nothing to do if the server is already running. */
if (conn->ckpt_session != NULL)
return (0);
F_SET(conn, WT_CONN_SERVER_CHECKPOINT);
/* The checkpoint server gets its own session. */
WT_RET(__wt_open_internal_session(
conn, "checkpoint-server", 1, 1, &conn->ckpt_session));
session = conn->ckpt_session;
/*
* Checkpoint does enough I/O it may be called upon to perform slow
* operations for the block manager.
*/
F_SET(session, WT_SESSION_CAN_WAIT);
WT_RET(
__wt_cond_alloc(session, "checkpoint server", 0, &conn->ckpt_cond));
/*
* Start the thread.
*/
WT_RET(__wt_thread_create(
session, &conn->ckpt_tid, __ckpt_server, session));
conn->ckpt_tid_set = 1;
return (0);
}
/*
* __ckpt_server_start --
* Start the checkpoint server thread.
*/
static int
__ckpt_server_start(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
session = conn->default_session;
/* Nothing to do if the server is already running. */
if (conn->ckpt_session != NULL)
return (0);
F_SET(conn, WT_CONN_SERVER_CHECKPOINT);
/* The checkpoint server gets its own session. */
WT_RET(__wt_open_internal_session(
conn, "checkpoint-server", 1, 1, &conn->ckpt_session));
WT_RET(
__wt_cond_alloc(session, "checkpoint server", 0, &conn->ckpt_cond));
/*
* Start the thread.
*/
WT_RET(__wt_thread_create(
session, &conn->ckpt_tid, __ckpt_server, conn->ckpt_session));
conn->ckpt_tid_set = 1;
return (0);
}
/*
* __wt_cache_create --
* Create the underlying cache.
*/
int
__wt_cache_create(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CACHE *cache;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
conn = S2C(session);
WT_ASSERT(session, conn->cache == NULL);
WT_RET(__wt_calloc_one(session, &conn->cache));
cache = conn->cache;
/* Use a common routine for run-time configuration options. */
WT_RET(__wt_cache_config(session, 0, cfg));
/*
* The target size must be lower than the trigger size or we will never
* get any work done.
*/
if (cache->eviction_target >= cache->eviction_trigger)
WT_ERR_MSG(session, EINVAL,
"eviction target must be lower than the eviction trigger");
WT_ERR(__wt_cond_alloc(session,
"cache eviction server", 0, &cache->evict_cond));
WT_ERR(__wt_cond_alloc(session,
"eviction waiters", 0, &cache->evict_waiter_cond));
WT_ERR(__wt_spin_init(session, &cache->evict_lock, "cache eviction"));
WT_ERR(__wt_spin_init(session, &cache->evict_walk_lock, "cache walk"));
/* Allocate the LRU eviction queue. */
cache->evict_slots = WT_EVICT_WALK_BASE + WT_EVICT_WALK_INCR;
WT_ERR(__wt_calloc_def(session, cache->evict_slots, &cache->evict));
/*
* We get/set some values in the cache statistics (rather than have
* two copies), configure them.
*/
__wt_cache_stats_update(session);
return (0);
err: WT_RET(__wt_cache_destroy(session));
return (ret);
}
/*
* __wt_cache_create --
* Create the underlying cache.
*/
int
__wt_cache_create(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CACHE *cache;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
conn = S2C(session);
WT_ASSERT(session, conn->cache == NULL);
WT_RET(__wt_calloc_one(session, &conn->cache));
cache = conn->cache;
/* Use a common routine for run-time configuration options. */
WT_RET(__wt_cache_config(session, false, cfg));
/*
* The lowest possible page read-generation has a special meaning, it
* marks a page for forcible eviction; don't let it happen by accident.
*/
cache->read_gen = WT_READGEN_START_VALUE;
/*
* The target size must be lower than the trigger size or we will never
* get any work done.
*/
if (cache->eviction_target >= cache->eviction_trigger)
WT_ERR_MSG(session, EINVAL,
"eviction target must be lower than the eviction trigger");
WT_ERR(__wt_cond_auto_alloc(session, "cache eviction server",
false, 10000, WT_MILLION, &cache->evict_cond));
WT_ERR(__wt_cond_alloc(session,
"eviction waiters", false, &cache->evict_waiter_cond));
WT_ERR(__wt_spin_init(session, &cache->evict_lock, "cache eviction"));
WT_ERR(__wt_spin_init(session, &cache->evict_walk_lock, "cache walk"));
/* Allocate the LRU eviction queue. */
cache->evict_slots = WT_EVICT_WALK_BASE + WT_EVICT_WALK_INCR;
WT_ERR(__wt_calloc_def(session,
cache->evict_slots, &cache->evict_queue));
/*
* We get/set some values in the cache statistics (rather than have
* two copies), configure them.
*/
__wt_cache_stats_update(session);
return (0);
err: WT_RET(__wt_cache_destroy(session));
return (ret);
}
/*
* __lsm_tree_open --
* Open an LSM tree structure.
*/
static int
__lsm_tree_open(
WT_SESSION_IMPL *session, const char *uri, WT_LSM_TREE **treep)
{
WT_DECL_RET;
WT_LSM_TREE *lsm_tree;
WT_ASSERT(session, F_ISSET(session, WT_SESSION_SCHEMA_LOCKED));
/* Make sure no one beat us to it. */
TAILQ_FOREACH(lsm_tree, &S2C(session)->lsmqh, q)
if (strcmp(uri, lsm_tree->name) == 0) {
*treep = lsm_tree;
return (0);
}
/* Try to open the tree. */
WT_RET(__wt_calloc_def(session, 1, &lsm_tree));
WT_ERR(__wt_rwlock_alloc(session, "lsm tree", &lsm_tree->rwlock));
WT_ERR(__wt_cond_alloc(session, "lsm ckpt", 0, &lsm_tree->work_cond));
WT_ERR(__lsm_tree_set_name(session, lsm_tree, uri));
WT_ERR(__wt_lsm_meta_read(session, lsm_tree));
/*
* Sanity check the configuration. Do it now since this is the first
* time we have the LSM tree configuration.
*/
WT_ERR(__lsm_tree_open_check(session, lsm_tree));
if (lsm_tree->nchunks == 0) {
F_SET(lsm_tree, WT_LSM_TREE_NEED_SWITCH);
WT_ERR(__wt_lsm_tree_switch(session, lsm_tree));
}
/* Set the generation number so cursors are opened on first usage. */
lsm_tree->dsk_gen = 1;
/* Now the tree is setup, make it visible to others. */
lsm_tree->refcnt = 1;
TAILQ_INSERT_HEAD(&S2C(session)->lsmqh, lsm_tree, q);
F_SET(lsm_tree, WT_LSM_TREE_OPEN);
WT_ERR(__lsm_tree_start_worker(session, lsm_tree));
*treep = lsm_tree;
if (0) {
err: WT_TRET(__lsm_tree_discard(session, lsm_tree));
}
return (ret);
}
/*
* __wt_sweep_create --
* Start the handle sweep thread.
*/
int
__wt_sweep_create(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
session = conn->default_session;
F_SET(conn, WT_CONN_SERVER_SWEEP);
WT_RET(__wt_open_session(conn, 1, NULL, NULL, &conn->sweep_session));
conn->sweep_session->name = "sweep-server";
session = conn->sweep_session;
WT_RET(__wt_cond_alloc(
session, "handle sweep server", 0, &conn->sweep_cond));
WT_RET(__wt_thread_create(
session, &conn->sweep_tid, __sweep_server, session));
conn->sweep_tid_set = 1;
return (0);
}
/*
* __wt_thread_group_create --
* Create a new thread group, assumes incoming group structure is
* zero initialized.
*/
int
__wt_thread_group_create(
WT_SESSION_IMPL *session, WT_THREAD_GROUP *group, const char *name,
uint32_t min, uint32_t max, uint32_t flags,
int (*run_func)(WT_SESSION_IMPL *session, WT_THREAD *context))
{
WT_DECL_RET;
bool cond_alloced;
/* Check that the structure is initialized as expected */
WT_ASSERT(session, group->alloc == 0);
cond_alloced = false;
__wt_verbose(session, WT_VERB_THREAD_GROUP,
"Creating thread group: %p", (void *)group);
WT_RET(__wt_rwlock_alloc(session, &group->lock, "Thread group"));
WT_ERR(__wt_cond_alloc(
session, "Thread group cond", false, &group->wait_cond));
cond_alloced = true;
__wt_writelock(session, group->lock);
group->run_func = run_func;
group->name = name;
WT_TRET(__thread_group_resize(session, group, min, max, flags));
__wt_writeunlock(session, group->lock);
/* Cleanup on error to avoid leaking resources */
err: if (ret != 0) {
if (cond_alloced)
WT_TRET(__wt_cond_destroy(session, &group->wait_cond));
__wt_rwlock_destroy(session, &group->lock);
}
return (ret);
}
/*
* __wt_connection_init --
* Structure initialization for a just-created WT_CONNECTION_IMPL handle.
*/
int
__wt_connection_init(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
u_int i;
session = conn->default_session;
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++) {
TAILQ_INIT(&conn->dhhash[i]); /* Data handle hash lists */
TAILQ_INIT(&conn->fhhash[i]); /* File handle hash lists */
}
TAILQ_INIT(&conn->dhqh); /* Data handle list */
TAILQ_INIT(&conn->dlhqh); /* Library list */
TAILQ_INIT(&conn->dsrcqh); /* Data source list */
TAILQ_INIT(&conn->fhqh); /* File list */
TAILQ_INIT(&conn->collqh); /* Collator list */
TAILQ_INIT(&conn->compqh); /* Compressor list */
TAILQ_INIT(&conn->encryptqh); /* Encryptor list */
TAILQ_INIT(&conn->extractorqh); /* Extractor list */
TAILQ_INIT(&conn->lsmqh); /* WT_LSM_TREE list */
/* Setup the LSM work queues. */
TAILQ_INIT(&conn->lsm_manager.switchqh);
TAILQ_INIT(&conn->lsm_manager.appqh);
TAILQ_INIT(&conn->lsm_manager.managerqh);
/* Random numbers. */
__wt_random_init(&session->rnd);
/* Configuration. */
WT_RET(__wt_conn_config_init(session));
/* Statistics. */
WT_RET(__wt_stat_connection_init(session, conn));
/* Spinlocks. */
WT_RET(__wt_spin_init(session, &conn->api_lock, "api"));
WT_SPIN_INIT_TRACKED(session, &conn->checkpoint_lock, checkpoint);
WT_SPIN_INIT_TRACKED(session, &conn->dhandle_lock, handle_list);
WT_RET(__wt_spin_init(session, &conn->encryptor_lock, "encryptor"));
WT_RET(__wt_spin_init(session, &conn->fh_lock, "file list"));
WT_RET(__wt_spin_init(session, &conn->las_lock, "lookaside table"));
WT_SPIN_INIT_TRACKED(session, &conn->metadata_lock, metadata);
WT_RET(__wt_spin_init(session, &conn->reconfig_lock, "reconfigure"));
WT_SPIN_INIT_TRACKED(session, &conn->schema_lock, schema);
WT_SPIN_INIT_TRACKED(session, &conn->table_lock, table);
WT_RET(__wt_spin_init(session, &conn->turtle_lock, "turtle file"));
/* Read-write locks */
WT_RET(__wt_rwlock_alloc(
session, &conn->hot_backup_lock, "hot backup"));
WT_RET(__wt_calloc_def(session, WT_PAGE_LOCKS, &conn->page_lock));
WT_CACHE_LINE_ALIGNMENT_VERIFY(session, conn->page_lock);
for (i = 0; i < WT_PAGE_LOCKS; ++i)
WT_RET(
__wt_spin_init(session, &conn->page_lock[i], "btree page"));
/* Setup the spin locks for the LSM manager queues. */
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.app_lock, "LSM application queue lock"));
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.manager_lock, "LSM manager queue lock"));
WT_RET(__wt_spin_init(
session, &conn->lsm_manager.switch_lock, "LSM switch queue lock"));
WT_RET(__wt_cond_alloc(
session, "LSM worker cond", false, &conn->lsm_manager.work_cond));
/*
* Generation numbers.
*
* Start split generations at one. Threads publish this generation
* number before examining tree structures, and zero when they leave.
* We need to distinguish between threads that are in a tree before the
* first split has happened, and threads that are not in a tree.
*/
conn->split_gen = 1;
/*
* Block manager.
* XXX
* If there's ever a second block manager, we'll want to make this
* more opaque, but for now this is simpler.
*/
WT_RET(__wt_spin_init(session, &conn->block_lock, "block manager"));
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++)
TAILQ_INIT(&conn->blockhash[i]);/* Block handle hash lists */
TAILQ_INIT(&conn->blockqh); /* Block manager list */
return (0);
}
/*
* __wt_connection_init --
* Structure initialization for a just-created WT_CONNECTION_IMPL handle.
*/
int
__wt_connection_init(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
u_int i;
session = conn->default_session;
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++) {
SLIST_INIT(&conn->dhhash[i]); /* Data handle hash lists */
SLIST_INIT(&conn->fhhash[i]); /* File handle hash lists */
}
SLIST_INIT(&conn->dhlh); /* Data handle list */
TAILQ_INIT(&conn->dlhqh); /* Library list */
TAILQ_INIT(&conn->dsrcqh); /* Data source list */
SLIST_INIT(&conn->fhlh); /* File list */
TAILQ_INIT(&conn->collqh); /* Collator list */
TAILQ_INIT(&conn->compqh); /* Compressor list */
TAILQ_INIT(&conn->extractorqh); /* Extractor list */
TAILQ_INIT(&conn->lsmqh); /* WT_LSM_TREE list */
/* Setup the LSM work queues. */
TAILQ_INIT(&conn->lsm_manager.switchqh);
TAILQ_INIT(&conn->lsm_manager.appqh);
TAILQ_INIT(&conn->lsm_manager.managerqh);
/* Configuration. */
WT_RET(__wt_conn_config_init(session));
/* Statistics. */
__wt_stat_init_connection_stats(&conn->stats);
/* Locks. */
WT_RET(__wt_spin_init(session, &conn->api_lock, "api"));
WT_RET(__wt_spin_init(session, &conn->checkpoint_lock, "checkpoint"));
WT_RET(__wt_spin_init(session, &conn->dhandle_lock, "data handle"));
WT_RET(__wt_spin_init(session, &conn->fh_lock, "file list"));
WT_RET(__wt_spin_init(session, &conn->hot_backup_lock, "hot backup"));
WT_RET(__wt_spin_init(session, &conn->reconfig_lock, "reconfigure"));
WT_RET(__wt_spin_init(session, &conn->schema_lock, "schema"));
WT_RET(__wt_spin_init(session, &conn->table_lock, "table creation"));
WT_RET(__wt_calloc_def(session, WT_PAGE_LOCKS(conn), &conn->page_lock));
for (i = 0; i < WT_PAGE_LOCKS(conn); ++i)
WT_RET(
__wt_spin_init(session, &conn->page_lock[i], "btree page"));
/* Setup the spin locks for the LSM manager queues. */
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.app_lock, "LSM application queue lock"));
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.manager_lock, "LSM manager queue lock"));
WT_RET(__wt_spin_init(
session, &conn->lsm_manager.switch_lock, "LSM switch queue lock"));
WT_RET(__wt_cond_alloc(
session, "LSM worker cond", 0, &conn->lsm_manager.work_cond));
/*
* Generation numbers.
*
* Start split generations at one. Threads publish this generation
* number before examining tree structures, and zero when they leave.
* We need to distinguish between threads that are in a tree before the
* first split has happened, and threads that are not in a tree.
*/
conn->split_gen = 1;
/*
* Block manager.
* XXX
* If there's ever a second block manager, we'll want to make this
* more opaque, but for now this is simpler.
*/
WT_RET(__wt_spin_init(session, &conn->block_lock, "block manager"));
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++)
SLIST_INIT(&conn->blockhash[i]);/* Block handle hash lists */
SLIST_INIT(&conn->blocklh); /* Block manager list */
return (0);
}
/*
* __wt_logmgr_open --
* Start the log service threads.
*/
int
__wt_logmgr_open(WT_SESSION_IMPL *session)
{
WT_CONNECTION_IMPL *conn;
uint32_t session_flags;
conn = S2C(session);
/* If no log thread services are configured, we're done. */
if (!FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED))
return (0);
F_SET(conn, WT_CONN_SERVER_LOG);
/*
* Start the log close thread. It is not configurable.
* If logging is enabled, this thread runs.
*/
session_flags = WT_SESSION_NO_DATA_HANDLES;
WT_RET(__wt_open_internal_session(conn,
"log-close-server", false, session_flags, &conn->log_file_session));
WT_RET(__wt_cond_alloc(
conn->log_file_session, "log close server", &conn->log_file_cond));
/*
* Start the log file close thread.
*/
WT_RET(__wt_thread_create(conn->log_file_session,
&conn->log_file_tid, __log_file_server, conn->log_file_session));
conn->log_file_tid_set = true;
/*
* Start the log write LSN thread. It is not configurable.
* If logging is enabled, this thread runs.
*/
WT_RET(__wt_open_internal_session(conn, "log-wrlsn-server",
false, session_flags, &conn->log_wrlsn_session));
WT_RET(__wt_cond_auto_alloc(conn->log_wrlsn_session,
"log write lsn server", 10000, WT_MILLION, &conn->log_wrlsn_cond));
WT_RET(__wt_thread_create(conn->log_wrlsn_session,
&conn->log_wrlsn_tid, __log_wrlsn_server, conn->log_wrlsn_session));
conn->log_wrlsn_tid_set = true;
/*
* If a log server thread exists, the user may have reconfigured
* archiving or pre-allocation. Signal the thread. Otherwise the
* user wants archiving and/or allocation and we need to start up
* the thread.
*/
if (conn->log_session != NULL) {
WT_ASSERT(session, conn->log_cond != NULL);
WT_ASSERT(session, conn->log_tid_set == true);
__wt_cond_signal(session, conn->log_cond);
} else {
/* The log server gets its own session. */
WT_RET(__wt_open_internal_session(conn,
"log-server", false, session_flags, &conn->log_session));
WT_RET(__wt_cond_auto_alloc(conn->log_session,
"log server", 50000, WT_MILLION, &conn->log_cond));
/*
* Start the thread.
*/
WT_RET(__wt_thread_create(conn->log_session,
&conn->log_tid, __log_server, conn->log_session));
conn->log_tid_set = true;
}
return (0);
}
/*
* __wt_open_session --
* Allocate a session handle. The internal parameter is used for sessions
* opened by WiredTiger for its own use.
*/
int
__wt_open_session(WT_CONNECTION_IMPL *conn, int internal,
WT_EVENT_HANDLER *event_handler, const char *config,
WT_SESSION_IMPL **sessionp)
{
static WT_SESSION stds = {
NULL,
__session_close,
__session_reconfigure,
__session_open_cursor,
__session_create,
__session_compact,
__session_drop,
__session_rename,
__session_salvage,
__session_truncate,
__session_upgrade,
__session_verify,
__session_begin_transaction,
__session_commit_transaction,
__session_rollback_transaction,
__session_checkpoint,
__session_msg_printf
};
WT_DECL_RET;
WT_SESSION_IMPL *session, *session_ret;
uint32_t i;
session = conn->default_session;
session_ret = NULL;
__wt_spin_lock(session, &conn->api_lock);
/* Find the first inactive session slot. */
for (session_ret = conn->sessions,
i = 0; i < conn->session_size; ++session_ret, ++i)
if (!session_ret->active)
break;
if (i == conn->session_size)
WT_ERR_MSG(session, WT_ERROR,
"only configured to support %d thread contexts",
conn->session_size);
/*
* If the active session count is increasing, update it. We don't worry
* about correcting the session count on error, as long as we don't mark
* this session as active, we'll clean it up on close.
*/
if (i >= conn->session_cnt) /* Defend against off-by-one errors. */
conn->session_cnt = i + 1;
session_ret->id = i;
session_ret->iface = stds;
session_ret->iface.connection = &conn->iface;
WT_ERR(__wt_cond_alloc(session, "session", 0, &session_ret->cond));
__wt_event_handler_set(session_ret,
event_handler == NULL ? session->event_handler : event_handler);
TAILQ_INIT(&session_ret->cursors);
TAILQ_INIT(&session_ret->btrees);
/* Initialize transaction support. */
WT_ERR(__wt_txn_init(session_ret));
/*
* The session's hazard reference memory isn't discarded during normal
* session close because access to it isn't serialized. Allocate the
* first time we open this session.
*/
if (session_ret->hazard == NULL)
WT_ERR(__wt_calloc(session, conn->hazard_max,
sizeof(WT_HAZARD), &session_ret->hazard));
/*
* Set an initial size for the hazard array. It will be grown as
* required up to hazard_max. The hazard_size is reset on close, since
* __wt_hazard_close ensures the array is cleared - so it is safe to
* reset the starting size on each open.
*/
session_ret->hazard_size = WT_HAZARD_INCR;
/*
* Public sessions are automatically closed during WT_CONNECTION->close.
* If the session handles for internal threads were to go on the public
* list, there would be complex ordering issues during close. Set a
* flag to avoid this: internal sessions are not closed automatically.
*/
if (internal)
F_SET(session_ret, WT_SESSION_INTERNAL);
/*
* Configuration: currently, the configuration for open_session is the
* same as session.reconfigure, so use that function.
*/
if (config != NULL)
WT_ERR(
__session_reconfigure((WT_SESSION *)session_ret, config));
/*
* Publish: make the entry visible to server threads. There must be a
* barrier for two reasons, to ensure structure fields are set before
* any other thread will consider the session, and to push the session
* count to ensure the eviction thread can't review too few slots.
*/
WT_PUBLISH(session_ret->active, 1);
STATIC_ASSERT(offsetof(WT_SESSION_IMPL, iface) == 0);
*sessionp = session_ret;
err: __wt_spin_unlock(session, &conn->api_lock);
return (ret);
}
/*
* __wt_open_session --
* Allocate a session handle. The internal parameter is used for sessions
* opened by WiredTiger for its own use.
*/
int
__wt_open_session(WT_CONNECTION_IMPL *conn, int internal,
WT_EVENT_HANDLER *event_handler, const char *config,
WT_SESSION_IMPL **sessionp)
{
static WT_SESSION stds = {
NULL,
__session_close,
__session_open_cursor,
__session_create,
__session_drop,
__session_rename,
__session_salvage,
__session_sync,
__session_truncate,
__session_upgrade,
__session_verify,
__session_begin_transaction,
__session_commit_transaction,
__session_rollback_transaction,
__session_checkpoint,
__session_dumpfile,
__session_msg_printf
};
WT_SESSION_IMPL *session, *session_ret;
uint32_t slot;
int ret;
WT_UNUSED(config);
ret = 0;
session = &conn->default_session;
session_ret = NULL;
__wt_spin_lock(session, &conn->spinlock);
/* Check to see if there's an available session slot. */
if (conn->session_cnt == conn->session_size - 1)
WT_ERR_MSG(session, WT_ERROR,
"WiredTiger only configured to support %d thread contexts",
conn->session_size);
/*
* The session reference list is compact, the session array is not.
* Find the first empty session slot.
*/
for (slot = 0, session_ret = conn->session_array;
session_ret->iface.connection != NULL;
++session_ret, ++slot)
;
/* Session entries are re-used, clear the old contents. */
WT_CLEAR(*session_ret);
WT_ERR(__wt_cond_alloc(session, "session", 1, &session_ret->cond));
session_ret->iface = stds;
session_ret->iface.connection = &conn->iface;
WT_ASSERT(session, session->event_handler != NULL);
session_ret->event_handler = session->event_handler;
session_ret->hazard = conn->hazard + slot * conn->hazard_size;
TAILQ_INIT(&session_ret->cursors);
TAILQ_INIT(&session_ret->btrees);
if (event_handler != NULL)
session_ret->event_handler = event_handler;
/*
* Public sessions are automatically closed during WT_CONNECTION->close.
* If the session handles for internal threads were to go on the public
* list, there would be complex ordering issues during close. Set a
* flag to avoid this: internal sessions are not closed automatically.
*/
if (internal)
F_SET(session_ret, WT_SESSION_INTERNAL);
/*
* Publish: make the entry visible to server threads. There must be a
* barrier to ensure the structure fields are set before any other
* thread can see the session.
*/
WT_PUBLISH(conn->sessions[conn->session_cnt++], session_ret);
STATIC_ASSERT(offsetof(WT_CONNECTION_IMPL, iface) == 0);
*sessionp = session_ret;
err:
__wt_spin_unlock(session, &conn->spinlock);
return (ret);
}
/*
* __wt_logmgr_open --
* Start the log service threads.
*/
int
__wt_logmgr_open(WT_SESSION_IMPL *session)
{
WT_CONNECTION_IMPL *conn;
conn = S2C(session);
/* If no log thread services are configured, we're done. */
if (!FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED))
return (0);
/*
* Start the log close thread. It is not configurable.
* If logging is enabled, this thread runs.
*/
WT_RET(__wt_open_internal_session(
conn, "log-close-server", 0, 0, &conn->log_file_session));
WT_RET(__wt_cond_alloc(conn->log_file_session,
"log close server", 0, &conn->log_file_cond));
/*
* Start the log file close thread.
*/
WT_RET(__wt_thread_create(conn->log_file_session,
&conn->log_file_tid, __log_file_server, conn->log_file_session));
conn->log_file_tid_set = 1;
/*
* Start the log write LSN thread. It is not configurable.
* If logging is enabled, this thread runs.
*/
WT_RET(__wt_open_internal_session(
conn, "log-wrlsn-server", 0, 0, &conn->log_wrlsn_session));
WT_RET(__wt_cond_alloc(conn->log_wrlsn_session,
"log write lsn server", 0, &conn->log_wrlsn_cond));
WT_RET(__wt_thread_create(conn->log_wrlsn_session,
&conn->log_wrlsn_tid, __log_wrlsn_server, conn->log_wrlsn_session));
conn->log_wrlsn_tid_set = 1;
/* If no log thread services are configured, we're done. */
if (!FLD_ISSET(conn->log_flags,
(WT_CONN_LOG_ARCHIVE | WT_CONN_LOG_PREALLOC)))
return (0);
/*
* If a log server thread exists, the user may have reconfigured
* archiving or pre-allocation. Signal the thread. Otherwise the
* user wants archiving and/or allocation and we need to start up
* the thread.
*/
if (conn->log_session != NULL) {
WT_ASSERT(session, conn->log_cond != NULL);
WT_ASSERT(session, conn->log_tid_set != 0);
WT_RET(__wt_cond_signal(session, conn->log_cond));
} else {
/* The log server gets its own session. */
WT_RET(__wt_open_internal_session(
conn, "log-server", 0, 0, &conn->log_session));
WT_RET(__wt_cond_alloc(conn->log_session,
"log server", 0, &conn->log_cond));
/*
* Start the thread.
*/
WT_RET(__wt_thread_create(conn->log_session,
&conn->log_tid, __log_server, conn->log_session));
conn->log_tid_set = 1;
}
return (0);
}
/*
* __wt_connection_init --
* Structure initialization for a just-created WT_CONNECTION_IMPL handle.
*/
int
__wt_connection_init(WT_CONNECTION_IMPL *conn)
{
WT_SESSION_IMPL *session;
u_int i;
session = conn->default_session;
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++) {
TAILQ_INIT(&conn->dhhash[i]); /* Data handle hash lists */
TAILQ_INIT(&conn->fhhash[i]); /* File handle hash lists */
}
TAILQ_INIT(&conn->dhqh); /* Data handle list */
TAILQ_INIT(&conn->dlhqh); /* Library list */
TAILQ_INIT(&conn->dsrcqh); /* Data source list */
TAILQ_INIT(&conn->fhqh); /* File list */
TAILQ_INIT(&conn->collqh); /* Collator list */
TAILQ_INIT(&conn->compqh); /* Compressor list */
TAILQ_INIT(&conn->encryptqh); /* Encryptor list */
TAILQ_INIT(&conn->extractorqh); /* Extractor list */
TAILQ_INIT(&conn->lsmqh); /* WT_LSM_TREE list */
/* Setup the LSM work queues. */
TAILQ_INIT(&conn->lsm_manager.switchqh);
TAILQ_INIT(&conn->lsm_manager.appqh);
TAILQ_INIT(&conn->lsm_manager.managerqh);
/* Random numbers. */
__wt_random_init(&session->rnd);
/* Configuration. */
WT_RET(__wt_conn_config_init(session));
/* Statistics. */
WT_RET(__wt_stat_connection_init(session, conn));
/* Spinlocks. */
WT_RET(__wt_spin_init(session, &conn->api_lock, "api"));
WT_SPIN_INIT_TRACKED(session, &conn->checkpoint_lock, checkpoint);
WT_RET(__wt_spin_init(session, &conn->encryptor_lock, "encryptor"));
WT_RET(__wt_spin_init(session, &conn->fh_lock, "file list"));
WT_SPIN_INIT_TRACKED(session, &conn->metadata_lock, metadata);
WT_RET(__wt_spin_init(session, &conn->reconfig_lock, "reconfigure"));
WT_SPIN_INIT_SESSION_TRACKED(session, &conn->schema_lock, schema);
WT_RET(__wt_spin_init(session, &conn->turtle_lock, "turtle file"));
/* Read-write locks */
WT_RWLOCK_INIT_SESSION_TRACKED(session, &conn->dhandle_lock, dhandle);
WT_RET(__wt_rwlock_init(session, &conn->hot_backup_lock));
WT_RWLOCK_INIT_TRACKED(session, &conn->table_lock, table);
/* Setup the spin locks for the LSM manager queues. */
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.app_lock, "LSM application queue lock"));
WT_RET(__wt_spin_init(session,
&conn->lsm_manager.manager_lock, "LSM manager queue lock"));
WT_RET(__wt_spin_init(
session, &conn->lsm_manager.switch_lock, "LSM switch queue lock"));
WT_RET(__wt_cond_alloc(
session, "LSM worker cond", &conn->lsm_manager.work_cond));
/* Initialize the generation manager. */
__wt_gen_init(session);
/*
* Block manager.
* XXX
* If there's ever a second block manager, we'll want to make this
* more opaque, but for now this is simpler.
*/
WT_RET(__wt_spin_init(session, &conn->block_lock, "block manager"));
for (i = 0; i < WT_HASH_ARRAY_SIZE; i++)
TAILQ_INIT(&conn->blockhash[i]);/* Block handle hash lists */
TAILQ_INIT(&conn->blockqh); /* Block manager list */
return (0);
}
