summaryrefslogtreecommitdiff
path: root/storage
diff options
context:
space:
mode:
authorBrave Galera Crew <devel@codership.com>2019-01-23 15:30:00 +0400
committerSergey Vojtovich <svoj@mariadb.org>2019-01-23 15:30:00 +0400
commit36a2a185fe18d31a644da46cfabd9757a379280c (patch)
tree00ca186ce2cfdc3ab7e4979336a384e2b51c5aa9 /storage
parent382115b99297ceaa4c3067f79efb5c2515013be5 (diff)
downloadmariadb-git-36a2a185fe18d31a644da46cfabd9757a379280c.tar.gz
Galera4
Diffstat (limited to 'storage')
-rw-r--r--storage/innobase/buf/buf0dump.cc4
-rw-r--r--storage/innobase/handler/ha_innodb.cc490
-rw-r--r--storage/innobase/handler/ha_innodb.h33
-rw-r--r--storage/innobase/include/trx0trx.h5
-rw-r--r--storage/innobase/lock/lock0lock.cc55
-rw-r--r--storage/innobase/row/row0ins.cc30
-rw-r--r--storage/innobase/row/row0sel.cc10
-rw-r--r--storage/innobase/row/row0upd.cc6
-rw-r--r--storage/innobase/srv/srv0conc.cc8
-rw-r--r--storage/innobase/srv/srv0srv.cc4
-rw-r--r--storage/innobase/srv/srv0start.cc2
-rw-r--r--storage/innobase/trx/trx0roll.cc11
-rw-r--r--storage/innobase/trx/trx0rseg.cc69
-rw-r--r--storage/innobase/trx/trx0trx.cc7
14 files changed, 341 insertions, 393 deletions
diff --git a/storage/innobase/buf/buf0dump.cc b/storage/innobase/buf/buf0dump.cc
index c8e7236dc93..2747a6fa338 100644
--- a/storage/innobase/buf/buf0dump.cc
+++ b/storage/innobase/buf/buf0dump.cc
@@ -822,7 +822,7 @@ DECLARE_THREAD(buf_dump_thread)(void*)
if (srv_buffer_pool_load_at_startup) {
#ifdef WITH_WSREP
- if (!wsrep_recovery) {
+ if (!get_wsrep_recovery()) {
#endif /* WITH_WSREP */
buf_load();
#ifdef WITH_WSREP
@@ -856,7 +856,7 @@ DECLARE_THREAD(buf_dump_thread)(void*)
"Dumping of buffer pool not started"
" as load was incomplete");
#ifdef WITH_WSREP
- } else if (wsrep_recovery) {
+ } else if (get_wsrep_recovery()) {
#endif /* WITH_WSREP */
} else {
buf_dump(FALSE/* do complete dump at shutdown */);
diff --git a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
index 2690ff9f736..186c3ee3154 100644
--- a/storage/innobase/handler/ha_innodb.cc
+++ b/storage/innobase/handler/ha_innodb.cc
@@ -131,6 +131,9 @@ void close_thread_tables(THD* thd);
#define tdc_size 400
#endif
+#include <mysql/plugin.h>
+#include <mysql/service_wsrep.h>
+
#include "ha_innodb.h"
#include "i_s.h"
#include "sync0sync.h"
@@ -138,9 +141,6 @@ void close_thread_tables(THD* thd);
#include <string>
#include <sstream>
-#include <mysql/plugin.h>
-#include <mysql/service_wsrep.h>
-
#ifdef WITH_WSREP
#include "dict0priv.h"
#include <mysql/service_md5.h>
@@ -148,21 +148,6 @@ void close_thread_tables(THD* thd);
extern MYSQL_PLUGIN_IMPORT MYSQL_BIN_LOG mysql_bin_log;
-static inline wsrep_ws_handle_t*
-wsrep_ws_handle(THD* thd, const trx_t* trx) {
- return wsrep_ws_handle_for_trx(wsrep_thd_ws_handle(thd),
- (wsrep_trx_id_t)trx->id);
-}
-
-extern TC_LOG* tc_log;
-extern void wsrep_cleanup_transaction(THD *thd);
-static int
-wsrep_abort_transaction(handlerton* hton, THD *bf_thd, THD *victim_thd,
- my_bool signal);
-static void
-wsrep_fake_trx_id(handlerton* hton, THD *thd);
-static int innobase_wsrep_set_checkpoint(handlerton* hton, const XID* xid);
-static int innobase_wsrep_get_checkpoint(handlerton* hton, XID* xid);
#endif /* WITH_WSREP */
/** to force correct commit order in binlog */
@@ -1843,8 +1828,13 @@ thd_to_trx_id(
{
return(thd_to_trx(thd)->id);
}
-#endif /* WITH_WSREP */
+static int
+wsrep_abort_transaction(handlerton* hton, THD *bf_thd, THD *victim_thd,
+ my_bool signal);
+static int innobase_wsrep_set_checkpoint(handlerton* hton, const XID* xid);
+static int innobase_wsrep_get_checkpoint(handlerton* hton, XID* xid);
+#endif /* WITH_WSREP */
/********************************************************************//**
Increments innobase_active_counter and every INNOBASE_WAKE_INTERVALth
time calls srv_active_wake_master_thread. This function should be used
@@ -4161,13 +4151,12 @@ static int innodb_init(void* p)
innobase_hton->show_status = innobase_show_status;
innobase_hton->flags =
HTON_SUPPORTS_EXTENDED_KEYS | HTON_SUPPORTS_FOREIGN_KEYS
- | HTON_NATIVE_SYS_VERSIONING;
+ | HTON_NATIVE_SYS_VERSIONING | HTON_WSREP_REPLICATION;
#ifdef WITH_WSREP
innobase_hton->abort_transaction=wsrep_abort_transaction;
innobase_hton->set_checkpoint=innobase_wsrep_set_checkpoint;
innobase_hton->get_checkpoint=innobase_wsrep_get_checkpoint;
- innobase_hton->fake_trx_id=wsrep_fake_trx_id;
#endif /* WITH_WSREP */
innobase_hton->tablefile_extensions = ha_innobase_exts;
@@ -4502,6 +4491,14 @@ innobase_commit_ordered_2(
trx->flush_log_later = true;
}
+#ifdef WITH_WSREP
+ /* If the transaction is not run in 2pc, we must assign wsrep
+ XID here in order to get it written in rollback segment. */
+ if (wsrep_on(thd)) {
+ thd_get_xid(thd, (MYSQL_XID*)trx->xid);
+ }
+#endif /* WITH_WSREP */
+
innobase_commit_low(trx);
if (!read_only) {
@@ -4704,6 +4701,15 @@ innobase_rollback(
dberr_t error;
+#ifdef WITH_WSREP
+ /* If trx was assigned wsrep XID in prepare phase and the
+ trx is being rolled back due to BF abort, clear XID in order
+ to avoid writing it to rollback segment out of order. The XID
+ will be reassigned when the transaction is replayed. */
+ if (wsrep_is_wsrep_xid(trx->xid)) {
+ trx->xid->null();
+ }
+#endif /* WITH_WSREP */
if (rollback_trx
|| !thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
@@ -5109,7 +5115,7 @@ static void innobase_kill_query(handlerton*, THD* thd, enum thd_kill_levels)
{
DBUG_ENTER("innobase_kill_query");
#ifdef WITH_WSREP
- if (wsrep_thd_get_conflict_state(thd) != NO_CONFLICT) {
+ if (wsrep_on(thd) && wsrep_thd_is_aborting(thd)) {
/* if victim has been signaled by BF thread and/or aborting
is already progressing, following query aborting is not necessary
any more.
@@ -8107,9 +8113,9 @@ ha_innobase::write_row(
wsrep_thd_query(m_user_thd) :
(char *)"void");
error= DB_SUCCESS;
- wsrep_thd_set_conflict_state(
- m_user_thd, MUST_ABORT);
- innobase_srv_conc_exit_innodb(m_prebuilt);
+ wsrep_thd_self_abort(m_user_thd);
+ innobase_srv_conc_exit_innodb(
+ m_prebuilt);
/* jump straight to func exit over
* later wsrep hooks */
goto func_exit;
@@ -8179,10 +8185,14 @@ report_error:
#ifdef WITH_WSREP
if (!error_result
&& wsrep_on(m_user_thd)
- && wsrep_thd_exec_mode(m_user_thd) == LOCAL_STATE
+ && wsrep_thd_is_local(m_user_thd)
&& !wsrep_consistency_check(m_user_thd)
- && !wsrep_thd_ignore_table(m_user_thd)) {
- if (wsrep_append_keys(m_user_thd, false, record, NULL)) {
+ && (thd_sql_command(m_user_thd) != SQLCOM_CREATE_TABLE)
+ && (thd_sql_command(m_user_thd) != SQLCOM_LOAD ||
+ thd_binlog_format(m_user_thd) == BINLOG_FORMAT_ROW)) {
+ if (wsrep_append_keys(m_user_thd, WSREP_SERVICE_KEY_EXCLUSIVE,
+ record,
+ NULL)) {
DBUG_PRINT("wsrep", ("row key failed"));
error_result = HA_ERR_INTERNAL_ERROR;
goto wsrep_error;
@@ -8868,13 +8878,17 @@ func_exit:
innobase_active_small();
#ifdef WITH_WSREP
- if (error == DB_SUCCESS &&
- wsrep_thd_exec_mode(m_user_thd) == LOCAL_STATE &&
- wsrep_on(m_user_thd) &&
- !wsrep_thd_ignore_table(m_user_thd)) {
+ if (error == DB_SUCCESS
+ && wsrep_on(m_user_thd)
+ && wsrep_thd_is_local(m_user_thd)) {
+
DBUG_PRINT("wsrep", ("update row key"));
- if (wsrep_append_keys(m_user_thd, false, old_row, new_row)) {
+ if (wsrep_append_keys(m_user_thd,
+ wsrep_protocol_version >= 4
+ ? WSREP_SERVICE_KEY_UPDATE
+ : WSREP_SERVICE_KEY_EXCLUSIVE,
+ old_row, new_row)){
WSREP_DEBUG("WSREP: UPDATE_ROW_KEY FAILED");
DBUG_PRINT("wsrep", ("row key failed"));
err = HA_ERR_INTERNAL_ERROR;
@@ -8934,11 +8948,14 @@ ha_innobase::delete_row(
innobase_active_small();
#ifdef WITH_WSREP
- if (error == DB_SUCCESS &&
- wsrep_thd_exec_mode(m_user_thd) == LOCAL_STATE &&
- wsrep_on(m_user_thd) &&
- !wsrep_thd_ignore_table(m_user_thd)) {
- if (wsrep_append_keys(m_user_thd, false, record, NULL)) {
+ if (error == DB_SUCCESS
+ && wsrep_on(m_user_thd)
+ && wsrep_thd_is_local(m_user_thd)
+ && !wsrep_thd_ignore_table(m_user_thd)) {
+
+ if (wsrep_append_keys(m_user_thd, WSREP_SERVICE_KEY_EXCLUSIVE,
+ record,
+ NULL)) {
DBUG_PRINT("wsrep", ("delete fail"));
error = (dberr_t) HA_ERR_INTERNAL_ERROR;
goto wsrep_error;
@@ -10121,6 +10138,22 @@ wsrep_dict_foreign_find_index(
ibool check_charsets,
ulint check_null);
+inline
+const char*
+wsrep_key_type_to_str(Wsrep_service_key_type type)
+{
+ switch (type) {
+ case WSREP_SERVICE_KEY_SHARED:
+ return "shared";
+ case WSREP_SERVICE_KEY_REFERENCE:
+ return "reference";
+ case WSREP_SERVICE_KEY_UPDATE:
+ return "update";
+ case WSREP_SERVICE_KEY_EXCLUSIVE:
+ return "exclusive";
+ };
+ return "unknown";
+}
extern dberr_t
wsrep_append_foreign_key(
@@ -10130,17 +10163,17 @@ wsrep_append_foreign_key(
const rec_t* rec, /*!<in: clustered index record */
dict_index_t* index, /*!<in: clustered index */
ibool referenced, /*!<in: is check for referenced table */
- ibool shared) /*!<in: is shared access */
+ Wsrep_service_key_type key_type) /*!< in: access type of this key
+ (shared, exclusive, reference...) */
{
ut_a(trx);
THD* thd = (THD*)trx->mysql_thd;
ulint rcode = DB_SUCCESS;
char cache_key[513] = {'\0'};
int cache_key_len=0;
- bool const copy = true;
if (!wsrep_on(trx->mysql_thd) ||
- wsrep_thd_exec_mode(thd) != LOCAL_STATE) {
+ wsrep_thd_is_local(trx->mysql_thd) == false) {
return DB_SUCCESS;
}
@@ -10230,8 +10263,8 @@ wsrep_append_foreign_key(
if (rcode != DB_SUCCESS) {
WSREP_ERROR(
"FK key set failed: " ULINTPF
- " (" ULINTPF " " ULINTPF "), index: %s %s, %s",
- rcode, referenced, shared,
+ " (" ULINTPF "%s), index: %s %s, %s",
+ rcode, referenced, wsrep_key_type_to_str(key_type),
(index) ? index->name() : "void index",
(index && index->table) ? index->table->name.m_name :
"void table",
@@ -10250,7 +10283,7 @@ wsrep_append_foreign_key(
#ifdef WSREP_DEBUG_PRINT
ulint j;
fprintf(stderr, "FK parent key, table: %s %s len: %lu ",
- cache_key, (shared) ? "shared" : "exclusive", len+1);
+ cache_key, wsrep_key_type_to_str(key_type), len+1);
for (j=0; j<len+1; j++) {
fprintf(stderr, " %hhX, ", key[j]);
}
@@ -10269,7 +10302,8 @@ wsrep_append_foreign_key(
wsrep_buf_t wkey_part[3];
wsrep_key_t wkey = {wkey_part, 3};
- if (!wsrep_prepare_key(
+ if (!wsrep_prepare_key_for_innodb(
+ thd,
(const uchar*)cache_key,
cache_key_len + 1,
(const uchar*)key, len+1,
@@ -10280,17 +10314,7 @@ wsrep_append_foreign_key(
wsrep_thd_query(thd) : "void");
return DB_ERROR;
}
-
- wsrep_t *wsrep= get_wsrep();
-
- rcode = (int)wsrep->append_key(
- wsrep,
- wsrep_ws_handle(thd, trx),
- &wkey,
- 1,
- shared ? WSREP_KEY_SHARED : WSREP_KEY_EXCLUSIVE,
- copy);
-
+ rcode = wsrep_thd_append_key(thd, &wkey, 1, key_type);
if (rcode) {
DBUG_PRINT("wsrep", ("row key failed: " ULINTPF, rcode));
WSREP_ERROR("Appending cascaded fk row key failed: %s, "
@@ -10311,16 +10335,19 @@ wsrep_append_key(
TABLE_SHARE *table_share,
const char* key,
uint16_t key_len,
- bool shared
+ Wsrep_service_key_type key_type /*!< in: access type of this key
+ (shared, exclusive, semi...) */
)
{
DBUG_ENTER("wsrep_append_key");
- bool const copy = true;
+ DBUG_PRINT("enter",
+ ("thd: %lu trx: %lld", thd_get_thread_id(thd),
+ (long long)trx->id));
#ifdef WSREP_DEBUG_PRINT
- fprintf(stderr, "%s conn %ld, trx %llu, keylen %d, table %s\n Query: %s ",
- (shared) ? "Shared" : "Exclusive",
- thd_get_thread_id(thd), (long long)trx->id, key_len,
- table_share->table_name.str, wsrep_thd_query(thd));
+ fprintf(stderr, "%s conn %lu, trx " TRX_ID_FMT ", keylen %d, key %s.%s\n",
+ wsrep_key_type_to_str(key_type),
+ thd_get_thread_id(thd), trx->id, key_len,
+ table_share->table_name.str, key);
for (int i=0; i<key_len; i++) {
fprintf(stderr, "%hhX, ", key[i]);
}
@@ -10329,7 +10356,8 @@ wsrep_append_key(
wsrep_buf_t wkey_part[3];
wsrep_key_t wkey = {wkey_part, 3};
- if (!wsrep_prepare_key(
+ if (!wsrep_prepare_key_for_innodb(
+ thd,
(const uchar*)table_share->table_cache_key.str,
table_share->table_cache_key.length,
(const uchar*)key, key_len,
@@ -10341,15 +10369,7 @@ wsrep_append_key(
DBUG_RETURN(HA_ERR_INTERNAL_ERROR);
}
- wsrep_t *wsrep= get_wsrep();
-
- int rcode = (int)wsrep->append_key(
- wsrep,
- wsrep_ws_handle(thd, trx),
- &wkey,
- 1,
- shared ? WSREP_KEY_SHARED : WSREP_KEY_EXCLUSIVE,
- copy);
+ int rcode = wsrep_thd_append_key(thd, &wkey, 1, key_type);
if (rcode) {
DBUG_PRINT("wsrep", ("row key failed: %d", rcode));
WSREP_WARN("Appending row key failed: %s, %d",
@@ -10390,16 +10410,30 @@ int
ha_innobase::wsrep_append_keys(
/*===========================*/
THD *thd,
- bool shared,
+ Wsrep_service_key_type key_type, /*!< in: access type of this row
+ operation:
+ (shared, exclusive, reference...) */
const uchar* record0, /* in: row in MySQL format */
const uchar* record1) /* in: row in MySQL format */
{
+ /* Sanity check: newly inserted records should always be passed with
+ EXCLUSIVE key type, all the rest are expected to carry a pre-image
+ */
+ ut_a(record1 != NULL || key_type == WSREP_SERVICE_KEY_EXCLUSIVE);
+
int rcode;
DBUG_ENTER("wsrep_append_keys");
bool key_appended = false;
trx_t *trx = thd_to_trx(thd);
+#ifdef WSREP_DEBUG_PRINT
+ fprintf(stderr, "%s conn %lu, trx " TRX_ID_FMT ", table %s\nSQL: %s\n",
+ wsrep_key_type_to_str(key_type),
+ thd_get_thread_id(thd), trx->id,
+ table_share->table_name.str, wsrep_thd_query(thd));
+#endif
+
if (table_share && table_share->tmp_table != NO_TMP_TABLE) {
WSREP_DEBUG("skipping tmp table DML: THD: %lu tmp: %d SQL: %s",
thd_get_thread_id(thd),
@@ -10422,7 +10456,7 @@ ha_innobase::wsrep_append_keys(
if (!is_null) {
rcode = wsrep_append_key(
thd, trx, table_share, keyval,
- len, shared);
+ len, key_type);
if (rcode) {
DBUG_RETURN(rcode);
@@ -10440,68 +10474,91 @@ ha_innobase::wsrep_append_keys(
KEY* key_info = table->key_info + i;
if (key_info->flags & HA_NOSAME) {
hasPK = true;
+ break;
}
}
for (i=0; i<table->s->keys; ++i) {
- uint len;
- char keyval0[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
- char keyval1[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
- char* key0 = &keyval0[1];
- char* key1 = &keyval1[1];
KEY* key_info = table->key_info + i;
- ibool is_null;
dict_index_t* idx = innobase_get_index(i);
dict_table_t* tab = (idx) ? idx->table : NULL;
+ /* keyval[] shall contain an ordinal number at byte 0
+ and the actual key data shall be written at byte 1.
+ Hence the total data length is the key length + 1 */
+ char keyval0[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
+ char keyval1[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
keyval0[0] = (char)i;
keyval1[0] = (char)i;
+ char* key0 = &keyval0[1];
+ char* key1 = &keyval1[1];
if (!tab) {
WSREP_WARN("MariaDB-InnoDB key mismatch %s %s",
table->s->table_name.str,
key_info->name.str);
}
- /* !hasPK == table with no PK, must append all non-unique keys */
+ /* !hasPK == table with no PK,
+ must append all non-unique keys */
if (!hasPK || key_info->flags & HA_NOSAME ||
((tab &&
referenced_by_foreign_key2(tab, idx)) ||
(!tab && referenced_by_foreign_key()))) {
- len = wsrep_store_key_val_for_row(
+ ibool is_null0;
+ uint len0 = wsrep_store_key_val_for_row(
thd, table, i, key0,
WSREP_MAX_SUPPORTED_KEY_LENGTH,
- record0, &is_null);
- if (!is_null) {
- rcode = wsrep_append_key(
- thd, trx, table_share,
- keyval0, len+1, shared);
-
- if (rcode) {
- DBUG_RETURN(rcode);
- }
-
- if (key_info->flags & HA_NOSAME || shared)
- key_appended = true;
- } else {
- WSREP_DEBUG("NULL key skipped: %s",
- wsrep_thd_query(thd));
- }
+ record0, &is_null0);
if (record1) {
- len = wsrep_store_key_val_for_row(
+ ibool is_null1;
+ uint len1 = wsrep_store_key_val_for_row(
thd, table, i, key1,
WSREP_MAX_SUPPORTED_KEY_LENGTH,
- record1, &is_null);
+ record1, &is_null1);
+
+ if (is_null0 != is_null1 ||
+ len0 != len1 ||
+ memcmp(key0, key1, len0)) {
+ /* This key has chaged. If it
+ is unique, this is an exclusive
+ operation -> upgrade key type */
+ if (key_info->flags & HA_NOSAME) {
+ key_type = WSREP_SERVICE_KEY_EXCLUSIVE;
+ }
- if (!is_null && memcmp(key0, key1, len)) {
- rcode = wsrep_append_key(
+ if (!is_null1) {
+ rcode = wsrep_append_key(
thd, trx, table_share,
- keyval1, len+1, shared);
- if (rcode) DBUG_RETURN(rcode);
+ keyval1,
+ /* for len1+1 see keyval1
+ initialization comment */
+ len1+1, key_type);
+ if (rcode)
+ DBUG_RETURN(rcode);
+ }
}
}
+
+ if (!is_null0) {
+ rcode = wsrep_append_key(
+ thd, trx, table_share,
+ /* for len0+1 see keyval0
+ initialization comment */
+ keyval0, len0+1, key_type);
+ if (rcode)
+ DBUG_RETURN(rcode);
+
+ if (key_info->flags & HA_NOSAME ||
+ key_type == WSREP_SERVICE_KEY_SHARED||
+ key_type == WSREP_SERVICE_KEY_REFERENCE)
+ key_appended = true;
+ } else {
+ WSREP_DEBUG("NULL key skipped: %s",
+ wsrep_thd_query(thd));
+ }
}
}
}
@@ -10515,7 +10572,7 @@ ha_innobase::wsrep_append_keys(
if ((rcode = wsrep_append_key(thd, trx, table_share,
(const char*) digest, 16,
- shared))) {
+ key_type))) {
DBUG_RETURN(rcode);
}
@@ -10524,7 +10581,7 @@ ha_innobase::wsrep_append_keys(
digest, record1, table, m_prebuilt);
if ((rcode = wsrep_append_key(thd, trx, table_share,
(const char*) digest,
- 16, shared))) {
+ 16, key_type))) {
DBUG_RETURN(rcode);
}
}
@@ -15405,8 +15462,7 @@ ha_innobase::external_lock(
if (!skip) {
#ifdef WITH_WSREP
- if (!wsrep_on(thd) || wsrep_thd_exec_mode(thd) == LOCAL_STATE)
- {
+ if (!wsrep_on(thd) || wsrep_thd_is_local(m_user_thd)) {
#endif /* WITH_WSREP */
my_error(ER_BINLOG_STMT_MODE_AND_ROW_ENGINE, MYF(0),
" InnoDB is limited to row-logging when"
@@ -16920,6 +16976,14 @@ innobase_rollback_by_xid(
}
if (trx_t* trx = trx_get_trx_by_xid(xid)) {
+#ifdef WITH_WSREP
+ /* If a wsrep transaction is being rolled back during
+ the recovery, we must clear the xid in order to avoid
+ writing serialisation history for rolled back transaction. */
+ if (wsrep_is_wsrep_xid(trx->xid)) {
+ trx->xid->null();
+ }
+#endif /* WITH_WSREP */
int ret = innobase_rollback_trx(trx);
trx_deregister_from_2pc(trx);
ut_ad(!trx->will_lock);
@@ -18520,123 +18584,32 @@ wsrep_innobase_kill_one_trx(
bf_trx ? bf_trx->id : 0);
DBUG_RETURN(1);
}
-
WSREP_LOG_CONFLICT(bf_thd, thd, TRUE);
-
+ wsrep_thd_LOCK(thd);
WSREP_DEBUG("BF kill (" ULINTPF ", seqno: " INT64PF
"), victim: (%lu) trx: " TRX_ID_FMT,
signal, bf_seqno,
thd_get_thread_id(thd),
victim_trx->id);
- WSREP_DEBUG("Aborting query: %s conf %d trx: %" PRId64,
+ WSREP_DEBUG("Aborting query: %s conf %s trx: %lld",
(thd && wsrep_thd_query(thd)) ? wsrep_thd_query(thd) : "void",
- wsrep_thd_conflict_state(thd, FALSE),
- wsrep_thd_ws_handle(thd)->trx_id);
-
- wsrep_thd_LOCK(thd);
- DBUG_EXECUTE_IF("sync.wsrep_after_BF_victim_lock",
- {
- const char act[]=
- "now "
- "wait_for signal.wsrep_after_BF_victim_lock";
- DBUG_ASSERT(!debug_sync_set_action(bf_thd,
- STRING_WITH_LEN(act)));
- };);
-
-
- if (wsrep_thd_query_state(thd) == QUERY_EXITING) {
- WSREP_DEBUG("kill trx EXITING for " TRX_ID_FMT,
- victim_trx->id);
- wsrep_thd_UNLOCK(thd);
- DBUG_RETURN(0);
- }
-
- if (wsrep_thd_exec_mode(thd) != LOCAL_STATE) {
- WSREP_DEBUG("withdraw for BF trx: " TRX_ID_FMT ", state: %d",
- victim_trx->id,
- wsrep_thd_get_conflict_state(thd));
- }
-
- switch (wsrep_thd_get_conflict_state(thd)) {
- case NO_CONFLICT:
- wsrep_thd_set_conflict_state(thd, MUST_ABORT);
- break;
- case MUST_ABORT:
- WSREP_DEBUG("victim " TRX_ID_FMT " in MUST ABORT state",
- victim_trx->id);
- wsrep_thd_UNLOCK(thd);
- wsrep_thd_awake(thd, signal);
- DBUG_RETURN(0);
- break;
- case ABORTED:
- case ABORTING: // fall through
- default:
- WSREP_DEBUG("victim " TRX_ID_FMT " in state %d",
- victim_trx->id, wsrep_thd_get_conflict_state(thd));
- wsrep_thd_UNLOCK(thd);
- DBUG_RETURN(0);
- break;
- }
-
- switch (wsrep_thd_query_state(thd)) {
- case QUERY_COMMITTING:
- enum wsrep_status rcode;
-
- WSREP_DEBUG("kill query for: %ld",
- thd_get_thread_id(thd));
- WSREP_DEBUG("kill trx QUERY_COMMITTING for " TRX_ID_FMT,
- victim_trx->id);
-
- if (wsrep_thd_exec_mode(thd) == REPL_RECV) {
- wsrep_abort_slave_trx(bf_seqno,
- wsrep_thd_trx_seqno(thd));
- } else {
- wsrep_t *wsrep= get_wsrep();
- rcode = wsrep->abort_pre_commit(
- wsrep, bf_seqno,
- (wsrep_trx_id_t)wsrep_thd_ws_handle(thd)->trx_id
- );
-
- switch (rcode) {
- case WSREP_WARNING:
- WSREP_DEBUG("cancel commit warning: "
- TRX_ID_FMT,
- victim_trx->id);
- wsrep_thd_UNLOCK(thd);
- wsrep_thd_awake(thd, signal);
- DBUG_RETURN(1);
- break;
- case WSREP_OK:
- break;
- default:
- WSREP_ERROR(
- "cancel commit bad exit: %d "
- TRX_ID_FMT,
- rcode, victim_trx->id);
- /* unable to interrupt, must abort */
- /* note: kill_mysql() will block, if we cannot.
- * kill the lock holder first.
- */
- abort();
- break;
- }
- }
- wsrep_thd_UNLOCK(thd);
- wsrep_thd_awake(thd, signal);
- break;
- case QUERY_EXEC:
- /* it is possible that victim trx is itself waiting for some
- * other lock. We need to cancel this waiting
- */
- WSREP_DEBUG("kill trx QUERY_EXEC for " TRX_ID_FMT,
- victim_trx->id);
-
- victim_trx->lock.was_chosen_as_deadlock_victim= TRUE;
-
+ wsrep_thd_transaction_state_str(thd),
+ wsrep_thd_transaction_id(thd));
+
+ /*
+ * we mark with was_chosen_as_deadlock_victim transaction,
+ * which is already marked as BF victim
+ * lock_sys is held until this vicitm has aborted
+ */
+ victim_trx->lock.was_chosen_as_wsrep_victim = TRUE;
+
+ wsrep_thd_UNLOCK(thd);
+ if (wsrep_thd_bf_abort(bf_thd, thd, signal))
+ {
if (victim_trx->lock.wait_lock) {
- WSREP_DEBUG("victim has wait flag: %ld",
- thd_get_thread_id(thd));
+ WSREP_DEBUG("victim has wait flag: %lu",
+ thd_get_thread_id(thd));
lock_t* wait_lock = victim_trx->lock.wait_lock;
if (wait_lock) {
@@ -18644,67 +18617,7 @@ wsrep_innobase_kill_one_trx(
victim_trx->lock.was_chosen_as_deadlock_victim= TRUE;
lock_cancel_waiting_and_release(wait_lock);
}
-
- wsrep_thd_UNLOCK(thd);
- wsrep_thd_awake(thd, signal);
- } else {
- /* abort currently executing query */
- DBUG_PRINT("wsrep",("sending KILL_QUERY to: %lu",
- thd_get_thread_id(thd)));
- WSREP_DEBUG("kill query for: %ld",
- thd_get_thread_id(thd));
- /* Note that innobase_kill_query will take lock_mutex
- and trx_mutex */
- wsrep_thd_UNLOCK(thd);
- wsrep_thd_awake(thd, signal);
-
- /* for BF thd, we need to prevent him from committing */
- if (wsrep_thd_exec_mode(thd) == REPL_RECV) {
- wsrep_abort_slave_trx(bf_seqno,
- wsrep_thd_trx_seqno(thd));
- }
- }
- break;
- case QUERY_IDLE:
- {
- WSREP_DEBUG("kill IDLE for " TRX_ID_FMT, victim_trx->id);
-
- if (wsrep_thd_exec_mode(thd) == REPL_RECV) {
- WSREP_DEBUG("kill BF IDLE, seqno: %lld",
- (long long)wsrep_thd_trx_seqno(thd));
- wsrep_thd_UNLOCK(thd);
- wsrep_abort_slave_trx(bf_seqno,
- wsrep_thd_trx_seqno(thd));
- DBUG_RETURN(0);
}
- /* This will lock thd from proceeding after net_read() */
- wsrep_thd_set_conflict_state(thd, ABORTING);
-
- wsrep_lock_rollback();
-
- if (wsrep_aborting_thd_contains(thd)) {
- WSREP_WARN("duplicate thd aborter %lu",
- (ulong) thd_get_thread_id(thd));
- } else {
- wsrep_aborting_thd_enqueue(thd);
- DBUG_PRINT("wsrep",("enqueuing trx abort for %lu",
- thd_get_thread_id(thd)));
- WSREP_DEBUG("enqueuing trx abort for (%lu)",
- thd_get_thread_id(thd));
- }
-
- DBUG_PRINT("wsrep",("signalling wsrep rollbacker"));
- WSREP_DEBUG("signaling aborter");
- wsrep_unlock_rollback();
- wsrep_thd_UNLOCK(thd);
-
- break;
- }
- default:
- WSREP_WARN("bad wsrep query state: %d",
- wsrep_thd_query_state(thd));
- wsrep_thd_UNLOCK(thd);
- break;
}
DBUG_RETURN(0);
@@ -18724,26 +18637,22 @@ wsrep_abort_transaction(
trx_t* victim_trx = thd_to_trx(victim_thd);
trx_t* bf_trx = (bf_thd) ? thd_to_trx(bf_thd) : NULL;
- WSREP_DEBUG("abort transaction: BF: %s victim: %s victim conf: %d",
+ WSREP_DEBUG("abort transaction: BF: %s victim: %s victim conf: %s",
wsrep_thd_query(bf_thd),
wsrep_thd_query(victim_thd),
- wsrep_thd_conflict_state(victim_thd, FALSE));
+ wsrep_thd_transaction_state_str(victim_thd));
if (victim_trx) {
lock_mutex_enter();
trx_mutex_enter(victim_trx);
- int rcode = wsrep_innobase_kill_one_trx(bf_thd, bf_trx,
- victim_trx, signal);
- lock_mutex_exit();
+ int rcode= wsrep_innobase_kill_one_trx(bf_thd, bf_trx,
+ victim_trx, signal);
trx_mutex_exit(victim_trx);
+ lock_mutex_exit();
wsrep_srv_conc_cancel_wait(victim_trx);
DBUG_RETURN(rcode);
} else {
- WSREP_DEBUG("victim does not have transaction");
- wsrep_thd_LOCK(victim_thd);
- wsrep_thd_set_conflict_state(victim_thd, MUST_ABORT);
- wsrep_thd_UNLOCK(victim_thd);
- wsrep_thd_awake(victim_thd, signal);
+ wsrep_thd_bf_abort(bf_thd, victim_thd, signal);
}
DBUG_RETURN(-1);
@@ -18779,15 +18688,6 @@ innobase_wsrep_get_checkpoint(
trx_rseg_read_wsrep_checkpoint(*xid);
return 0;
}
-
-static void wsrep_fake_trx_id(handlerton *, THD *thd)
-{
- trx_id_t trx_id = trx_sys.get_new_trx_id();
- WSREP_DEBUG("innodb fake trx id: " TRX_ID_FMT " thd: %s",
- trx_id, wsrep_thd_query(thd));
- wsrep_ws_handle_for_trx(wsrep_thd_ws_handle(thd), trx_id);
-}
-
#endif /* WITH_WSREP */
/* plugin options */
diff --git a/storage/innobase/handler/ha_innodb.h b/storage/innobase/handler/ha_innodb.h
index 6f0c5b535fb..71212785df4 100644
--- a/storage/innobase/handler/ha_innodb.h
+++ b/storage/innobase/handler/ha_innodb.h
@@ -16,6 +16,10 @@ this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA
*****************************************************************************/
+#ifdef WITH_WSREP
+#include "wsrep_api.h"
+#include <mysql/service_wsrep.h>
+#endif /* WITH_WSREP */
/* The InnoDB handler: the interface between MySQL and InnoDB. */
@@ -444,8 +448,11 @@ protected:
dict_index_t* innobase_get_index(uint keynr);
#ifdef WITH_WSREP
- int wsrep_append_keys(THD *thd, bool shared,
- const uchar* record0, const uchar* record1);
+ int wsrep_append_keys(
+ THD *thd,
+ Wsrep_service_key_type key_type,
+ const uchar* record0,
+ const uchar* record1);
#endif
/** Builds a 'template' to the prebuilt struct.
@@ -571,23 +578,7 @@ thd_get_work_part_info(
struct trx_t;
#ifdef WITH_WSREP
#include <mysql/service_wsrep.h>
-//extern "C" int wsrep_trx_order_before(void *thd1, void *thd2);
-
-extern "C" bool wsrep_thd_is_wsrep_on(THD *thd);
-
-
-extern "C" void wsrep_thd_set_exec_mode(THD *thd, enum wsrep_exec_mode mode);
-extern "C" void wsrep_thd_set_query_state(
- THD *thd, enum wsrep_query_state state);
-
-extern "C" void wsrep_thd_set_trx_to_replay(THD *thd, uint64 trx_id);
-
-extern "C" uint32 wsrep_thd_wsrep_rand(THD *thd);
-extern "C" time_t wsrep_thd_query_start(THD *thd);
-extern "C" query_id_t wsrep_thd_query_id(THD *thd);
-extern "C" query_id_t wsrep_thd_wsrep_last_query_id(THD *thd);
-extern "C" void wsrep_thd_set_wsrep_last_query_id(THD *thd, query_id_t id);
-#endif
+#endif /* WITH_WSREP */
extern const struct _ft_vft ft_vft_result;
@@ -621,10 +612,6 @@ innobase_index_name_is_reserved(
be created. */
MY_ATTRIBUTE((nonnull(1), warn_unused_result));
-#ifdef WITH_WSREP
-//extern "C" int wsrep_trx_is_aborting(void *thd_ptr);
-#endif
-
/** Parse hint for table and its indexes, and update the information
in dictionary.
@param[in] thd Connection thread
diff --git a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h
index 34d61100a9b..8bf64c28441 100644
--- a/storage/innobase/include/trx0trx.h
+++ b/storage/innobase/include/trx0trx.h
@@ -577,6 +577,11 @@ struct trx_lock_t {
lock_sys.mutex. Otherwise, this may
only be modified by the thread that is
serving the running transaction. */
+#ifdef WITH_WSREP
+ bool was_chosen_as_wsrep_victim;
+ /*!< high priority wsrep thread has
+ marked this trx to abort */
+#endif /* WITH_WSREP */
/** Pre-allocated record locks */
struct {
diff --git a/storage/innobase/lock/lock0lock.cc b/storage/innobase/lock/lock0lock.cc
index 7fc986bdbca..493f73cc148 100644
--- a/storage/innobase/lock/lock0lock.cc
+++ b/storage/innobase/lock/lock0lock.cc
@@ -762,9 +762,7 @@ lock_rec_has_to_wait(
<< wsrep_thd_query(lock2->trx->mysql_thd);
}
- if (wsrep_trx_order_before(trx->mysql_thd,
- lock2->trx->mysql_thd)
- && (type_mode & LOCK_MODE_MASK) == LOCK_X
+ if ((type_mode & LOCK_MODE_MASK) == LOCK_X
&& (lock2->type_mode & LOCK_MODE_MASK) == LOCK_X) {
if (for_locking || wsrep_debug) {
/* exclusive lock conflicts are not
@@ -774,12 +772,11 @@ lock_rec_has_to_wait(
<< type_mode
<< " supremum: " << lock_is_on_supremum
<< "conflicts states: my "
- << wsrep_thd_conflict_state(
- trx->mysql_thd, FALSE)
+ << wsrep_thd_transaction_state_str(
+ trx->mysql_thd)
<< " locked "
- << wsrep_thd_conflict_state(
- lock2->trx->mysql_thd,
- FALSE);
+ << wsrep_thd_transaction_state_str(
+ lock2->trx->mysql_thd);
lock_rec_print(stderr, lock2);
ib::info() << " SQL1: "
<< wsrep_thd_query(trx->mysql_thd)
@@ -1098,11 +1095,14 @@ wsrep_kill_victim(
return;
}
- my_bool bf_this = wsrep_thd_is_BF(trx->mysql_thd, FALSE);
+ if (!wsrep_thd_is_BF(trx->mysql_thd, FALSE)) {
+ return;
+ }
+
my_bool bf_other = wsrep_thd_is_BF(lock->trx->mysql_thd, TRUE);
- if ((bf_this && !bf_other) ||
- (bf_this && bf_other && wsrep_trx_order_before(
+ if ((!bf_other) ||
+ (wsrep_thd_order_before(
trx->mysql_thd, lock->trx->mysql_thd))) {
if (lock->trx->lock.que_state == TRX_QUE_LOCK_WAIT) {
@@ -1113,11 +1113,7 @@ wsrep_kill_victim(
is in the queue*/
} else if (lock->trx != trx) {
if (wsrep_log_conflicts) {
- if (bf_this) {
- ib::info() << "*** Priority TRANSACTION:";
- } else {
- ib::info() << "*** Victim TRANSACTION:";
- }
+ ib::info() << "*** Priority TRANSACTION:";
trx_print_latched(stderr, trx, 3000);
@@ -1426,7 +1422,7 @@ lock_rec_create_low(
lock_t *prev = NULL;
while (hash && wsrep_thd_is_BF(hash->trx->mysql_thd, TRUE)
- && wsrep_trx_order_before(hash->trx->mysql_thd,
+ && wsrep_thd_order_before(hash->trx->mysql_thd,
trx->mysql_thd)) {
prev = hash;
hash = (lock_t *)hash->hash;
@@ -1840,15 +1836,15 @@ lock_rec_add_to_queue(
ib::info() << "WSREP BF lock conflict for my lock:\n BF:" <<
((wsrep_thd_is_BF(trx->mysql_thd, FALSE)) ? "BF" : "normal") << " exec: " <<
- wsrep_thd_exec_mode(trx->mysql_thd) << " conflict: " <<
- wsrep_thd_conflict_state(trx->mysql_thd, false) << " seqno: " <<
+ wsrep_thd_client_state_str(trx->mysql_thd) << " conflict: " <<
+ wsrep_thd_transaction_state_str(trx->mysql_thd) << " seqno: " <<
wsrep_thd_trx_seqno(trx->mysql_thd) << " SQL: " <<
wsrep_thd_query(trx->mysql_thd);
trx_t* otrx = other_lock->trx;
ib::info() << "WSREP other lock:\n BF:" <<
((wsrep_thd_is_BF(otrx->mysql_thd, FALSE)) ? "BF" : "normal") << " exec: " <<
- wsrep_thd_exec_mode(otrx->mysql_thd) << " conflict: " <<
- wsrep_thd_conflict_state(otrx->mysql_thd, false) << " seqno: " <<
+ wsrep_thd_client_state_str(otrx->mysql_thd) << " conflict: " <<
+ wsrep_thd_transaction_state_str(otrx->mysql_thd) << " seqno: " <<
wsrep_thd_trx_seqno(otrx->mysql_thd) << " SQL: " <<
wsrep_thd_query(otrx->mysql_thd);
}
@@ -4949,8 +4945,8 @@ lock_rec_queue_validate(
if (!lock_get_wait(other_lock) ) {
ib::info() << "WSREP impl BF lock conflict for my impl lock:\n BF:" <<
((wsrep_thd_is_BF(impl_trx->mysql_thd, FALSE)) ? "BF" : "normal") << " exec: " <<
- wsrep_thd_exec_mode(impl_trx->mysql_thd) << " conflict: " <<
- wsrep_thd_conflict_state(impl_trx->mysql_thd, false) << " seqno: " <<
+ wsrep_thd_client_state_str(impl_trx->mysql_thd) << " conflict: " <<
+ wsrep_thd_transaction_state_str(impl_trx->mysql_thd) << " seqno: " <<
wsrep_thd_trx_seqno(impl_trx->mysql_thd) << " SQL: " <<
wsrep_thd_query(impl_trx->mysql_thd);
@@ -4958,8 +4954,8 @@ lock_rec_queue_validate(
ib::info() << "WSREP other lock:\n BF:" <<
((wsrep_thd_is_BF(otrx->mysql_thd, FALSE)) ? "BF" : "normal") << " exec: " <<
- wsrep_thd_exec_mode(otrx->mysql_thd) << " conflict: " <<
- wsrep_thd_conflict_state(otrx->mysql_thd, false) << " seqno: " <<
+ wsrep_thd_client_state_str(otrx->mysql_thd) << " conflict: " <<
+ wsrep_thd_transaction_state_str(otrx->mysql_thd) << " seqno: " <<
wsrep_thd_trx_seqno(otrx->mysql_thd) << " SQL: " <<
wsrep_thd_query(otrx->mysql_thd);
}
@@ -6395,6 +6391,12 @@ lock_trx_handle_wait(
/*=================*/
trx_t* trx) /*!< in/out: trx lock state */
{
+#ifdef WITH_WSREP
+ /* We already own mutexes */
+ if (trx->lock.was_chosen_as_wsrep_victim) {
+ return lock_trx_handle_wait_low(trx);
+ }
+#endif /* WITH_WSREP */
lock_mutex_enter();
trx_mutex_enter(trx);
dberr_t err = lock_trx_handle_wait_low(trx);
@@ -6992,6 +6994,9 @@ DeadlockChecker::trx_rollback()
trx_t* trx = m_wait_lock->trx;
print("*** WE ROLL BACK TRANSACTION (1)\n");
+#ifdef WITH_WSREP
+ wsrep_handle_SR_rollback(m_start->mysql_thd, trx->mysql_thd);
+#endif
trx_mutex_enter(trx);
diff --git a/storage/innobase/row/row0ins.cc b/storage/innobase/row/row0ins.cc
index e2ac60c1103..ed72f6d482b 100644
--- a/storage/innobase/row/row0ins.cc
+++ b/storage/innobase/row/row0ins.cc
@@ -44,6 +44,9 @@ Created 4/20/1996 Heikki Tuuri
#include "buf0lru.h"
#include "fts0fts.h"
#include "fts0types.h"
+#ifdef WITH_WSREP
+#include "wsrep_mysqld.h"
+#endif /* WITH_WSREP */
/*************************************************************************
IMPORTANT NOTE: Any operation that generates redo MUST check that there
@@ -1043,7 +1046,7 @@ dberr_t wsrep_append_foreign_key(trx_t *trx,
const rec_t* clust_rec,
dict_index_t* clust_index,
ibool referenced,
- ibool shared);
+ Wsrep_service_key_type key_type);
#endif /* WITH_WSREP */
/*********************************************************************//**
@@ -1430,7 +1433,7 @@ row_ins_foreign_check_on_constraint(
#ifdef WITH_WSREP
err = wsrep_append_foreign_key(trx, foreign, clust_rec, clust_index,
- FALSE, FALSE);
+ FALSE, WSREP_SERVICE_KEY_EXCLUSIVE);
if (err != DB_SUCCESS) {
fprintf(stderr,
"WSREP: foreign key append failed: %d\n", err);
@@ -1811,7 +1814,10 @@ row_ins_check_foreign_constraint(
rec,
check_index,
check_ref,
- (upd_node) ? TRUE : FALSE);
+ (upd_node != NULL
+ && wsrep_protocol_version < 4)
+ ? WSREP_SERVICE_KEY_SHARED
+ : WSREP_SERVICE_KEY_REFERENCE);
#endif /* WITH_WSREP */
goto end_scan;
} else if (foreign->type != 0) {
@@ -3228,9 +3234,27 @@ row_ins_clust_index_entry(
n_uniq = dict_index_is_unique(index) ? index->n_uniq : 0;
+#ifdef WITH_WSREP
+ const bool skip_locking
+ = wsrep_thd_skip_locking(thr_get_trx(thr)->mysql_thd);
+ ulint flags = index->table->no_rollback() ? BTR_NO_ROLLBACK
+ : (index->table->is_temporary() || skip_locking)
+ ? BTR_NO_LOCKING_FLAG : 0;
+#ifdef UNIV_DEBUG
+ if (skip_locking && strcmp(wsrep_get_sr_table_name(),
+ index->table->name.m_name)) {
+ WSREP_ERROR("Record locking is disabled in this thread, "
+ "but the table being modified is not "
+ "`%s`: `%s`.", wsrep_get_sr_table_name(),
+ index->table->name.m_name);
+ ut_error;
+ }
+#endif /* UNIV_DEBUG */
+#else
ulint flags = index->table->no_rollback() ? BTR_NO_ROLLBACK
: index->table->is_temporary()
? BTR_NO_LOCKING_FLAG : 0;
+#endif /* WITH_WSREP */
const ulint orig_n_fields = entry->n_fields;
/* Try first optimistic descent to the B-tree */
diff --git a/storage/innobase/row/row0sel.cc b/storage/innobase/row/row0sel.cc
index 658785a5c03..8d7824bfc56 100644
--- a/storage/innobase/row/row0sel.cc
+++ b/storage/innobase/row/row0sel.cc
@@ -54,6 +54,9 @@ Created 12/19/1997 Heikki Tuuri
#include "buf0lru.h"
#include "srv0srv.h"
#include "srv0mon.h"
+#ifdef WITH_WSREP
+#include "mysql/service_wsrep.h" /* For wsrep_thd_skip_locking */
+#endif
/* Maximum number of rows to prefetch; MySQL interface has another parameter */
#define SEL_MAX_N_PREFETCH 16
@@ -4442,6 +4445,13 @@ row_search_mvcc(
set_also_gap_locks = FALSE;
}
+#ifdef WITH_WSREP
+ else if (wsrep_thd_skip_locking(trx->mysql_thd)) {
+ ut_ad(!strcmp(wsrep_get_sr_table_name(),
+ prebuilt->table->name.m_name));
+ set_also_gap_locks = FALSE;
+ }
+#endif /* WITH_WSREP */
/* Note that if the search mode was GE or G, then the cursor
naturally moves upward (in fetch next) in alphabetical order,
diff --git a/storage/innobase/row/row0upd.cc b/storage/innobase/row/row0upd.cc
index 591ae238c43..a60210c77c8 100644
--- a/storage/innobase/row/row0upd.cc
+++ b/storage/innobase/row/row0upd.cc
@@ -2458,7 +2458,7 @@ row_upd_sec_index_entry(
err = DB_SUCCESS;
break;
case DB_DEADLOCK:
- if (wsrep_debug) {
+ if (wsrep_get_debug()) {
ib::warn() << "WSREP: sec index FK check fail for deadlock"
<< " index " << index->name
<< " table " << index->table->name;
@@ -2773,7 +2773,7 @@ check_fk:
err = DB_SUCCESS;
break;
case DB_DEADLOCK:
- if (wsrep_debug) {
+ if (wsrep_get_debug()) {
ib::warn() << "WSREP: sec index FK check fail for deadlock"
<< " index " << index->name
<< " table " << index->table->name;
@@ -3002,7 +3002,7 @@ row_upd_del_mark_clust_rec(
err = DB_SUCCESS;
break;
case DB_DEADLOCK:
- if (wsrep_debug) {
+ if (wsrep_get_debug()) {
ib::warn() << "WSREP: sec index FK check fail for deadlock"
<< " index " << index->name
<< " table " << index->table->name;
diff --git a/storage/innobase/srv/srv0conc.cc b/storage/innobase/srv/srv0conc.cc
index 6beeee58fae..b2b464e31fa 100644
--- a/storage/innobase/srv/srv0conc.cc
+++ b/storage/innobase/srv/srv0conc.cc
@@ -119,8 +119,8 @@ srv_conc_enter_innodb_with_atomics(
ulint sleep_in_us;
#ifdef WITH_WSREP
if (wsrep_on(trx->mysql_thd) &&
- wsrep_trx_is_aborting(trx->mysql_thd)) {
- if (wsrep_debug) {
+ wsrep_thd_is_aborting(trx->mysql_thd)) {
+ if (wsrep_get_debug()) {
ib::info() <<
"srv_conc_enter due to MUST_ABORT";
}
@@ -308,14 +308,14 @@ wsrep_srv_conc_cancel_wait(
srv_conc_enter_innodb_with_atomics(). No need to cancel here,
thr will wake up after os_sleep and let to enter innodb
*/
- if (wsrep_debug) {
+ if (wsrep_get_debug()) {
ib::info() << "WSREP: conc slot cancel, no atomics";
}
#else
// JAN: TODO: MySQL 5.7
//os_fast_mutex_lock(&srv_conc_mutex);
if (trx->wsrep_event) {
- if (wsrep_debug) {
+ if (wsrep_get_debug()) {
ib::info() << "WSREP: conc slot cancel";
}
os_event_set(trx->wsrep_event);
diff --git a/storage/innobase/srv/srv0srv.cc b/storage/innobase/srv/srv0srv.cc
index 890bd969670..3b5461aae25 100644
--- a/storage/innobase/srv/srv0srv.cc
+++ b/storage/innobase/srv/srv0srv.cc
@@ -76,10 +76,6 @@ Created 10/8/1995 Heikki Tuuri
#include <my_service_manager.h>
-#ifdef WITH_WSREP
-extern int wsrep_debug;
-extern int wsrep_trx_is_aborting(void *thd_ptr);
-#endif
/* The following is the maximum allowed duration of a lock wait. */
UNIV_INTERN ulong srv_fatal_semaphore_wait_threshold = DEFAULT_SRV_FATAL_SEMAPHORE_TIMEOUT;
diff --git a/storage/innobase/srv/srv0start.cc b/storage/innobase/srv/srv0start.cc
index 54b0e75fd80..e15afcf9cd7 100644
--- a/storage/innobase/srv/srv0start.cc
+++ b/storage/innobase/srv/srv0start.cc
@@ -2328,7 +2328,7 @@ files_checked:
Create the dump/load thread only when not running with
--wsrep-recover.
*/
- if (!wsrep_recovery) {
+ if (!get_wsrep_recovery()) {
#endif /* WITH_WSREP */
/* Create the buffer pool dump/load thread */
diff --git a/storage/innobase/trx/trx0roll.cc b/storage/innobase/trx/trx0roll.cc
index 4b0684d1735..22b72a14304 100644
--- a/storage/innobase/trx/trx0roll.cc
+++ b/storage/innobase/trx/trx0roll.cc
@@ -178,6 +178,11 @@ trx_rollback_to_savepoint(
partial rollback requested, or NULL for
complete rollback */
{
+#ifdef WITH_WSREP
+ if (savept == NULL && wsrep_on(trx->mysql_thd)) {
+ wsrep_handle_SR_rollback(NULL, trx->mysql_thd);
+ }
+#endif /* WITH_WSREP */
ut_ad(!trx_mutex_own(trx));
trx_start_if_not_started_xa(trx, true);
@@ -446,12 +451,8 @@ trx_rollback_to_savepoint_for_mysql_low(
trx_mark_sql_stat_end(trx);
trx->op_info = "";
-
#ifdef WITH_WSREP
- if (wsrep_on(trx->mysql_thd) &&
- trx->lock.was_chosen_as_deadlock_victim) {
- trx->lock.was_chosen_as_deadlock_victim = FALSE;
- }
+ trx->lock.was_chosen_as_wsrep_victim = FALSE;
#endif
return(err);
}
diff --git a/storage/innobase/trx/trx0rseg.cc b/storage/innobase/trx/trx0rseg.cc
index f0793140277..239df440c2b 100644
--- a/storage/innobase/trx/trx0rseg.cc
+++ b/storage/innobase/trx/trx0rseg.cc
@@ -43,6 +43,33 @@ static long long wsrep_seqno = -1;
/** The latest known WSREP XID UUID */
static unsigned char wsrep_uuid[16];
+/** Write the WSREP XID information into rollback segment header.
+@param[in,out] rseg_header rollback segment header
+@param[in] xid WSREP XID
+@param[in,out] mtr mini transaction */
+static void
+trx_rseg_write_wsrep_checkpoint(
+ trx_rsegf_t* rseg_header,
+ const XID* xid,
+ mtr_t* mtr)
+{
+ mlog_write_ulint(TRX_RSEG_WSREP_XID_FORMAT + rseg_header,
+ uint32_t(xid->formatID),
+ MLOG_4BYTES, mtr);
+
+ mlog_write_ulint(TRX_RSEG_WSREP_XID_GTRID_LEN + rseg_header,
+ uint32_t(xid->gtrid_length),
+ MLOG_4BYTES, mtr);
+
+ mlog_write_ulint(TRX_RSEG_WSREP_XID_BQUAL_LEN + rseg_header,
+ uint32_t(xid->bqual_length),
+ MLOG_4BYTES, mtr);
+
+ mlog_write_string(TRX_RSEG_WSREP_XID_DATA + rseg_header,
+ reinterpret_cast<const byte*>(xid->data),
+ XIDDATASIZE, mtr);
+}
+
/** Update the WSREP XID information in rollback segment header.
@param[in,out] rseg_header rollback segment header
@param[in] xid WSREP XID
@@ -60,29 +87,15 @@ trx_rseg_update_wsrep_checkpoint(
long long xid_seqno = wsrep_xid_seqno(xid);
const byte* xid_uuid = wsrep_xid_uuid(xid);
- if (!memcmp(xid_uuid, wsrep_uuid, sizeof wsrep_uuid)) {
+ if (xid_seqno != -1
+ && !memcmp(xid_uuid, wsrep_uuid, sizeof wsrep_uuid)) {
ut_ad(xid_seqno > wsrep_seqno);
} else {
memcpy(wsrep_uuid, xid_uuid, sizeof wsrep_uuid);
}
wsrep_seqno = xid_seqno;
#endif /* UNIV_DEBUG */
-
- mlog_write_ulint(TRX_RSEG_WSREP_XID_FORMAT + rseg_header,
- uint32_t(xid->formatID),
- MLOG_4BYTES, mtr);
-
- mlog_write_ulint(TRX_RSEG_WSREP_XID_GTRID_LEN + rseg_header,
- uint32_t(xid->gtrid_length),
- MLOG_4BYTES, mtr);
-
- mlog_write_ulint(TRX_RSEG_WSREP_XID_BQUAL_LEN + rseg_header,
- uint32_t(xid->bqual_length),
- MLOG_4BYTES, mtr);
-
- mlog_write_string(TRX_RSEG_WSREP_XID_DATA + rseg_header,
- reinterpret_cast<const byte*>(xid->data),
- XIDDATASIZE, mtr);
+ trx_rseg_write_wsrep_checkpoint(rseg_header, xid, mtr);
}
/** Update WSREP checkpoint XID in first rollback segment header
@@ -97,6 +110,13 @@ void trx_rseg_update_wsrep_checkpoint(const XID* xid)
mtr_t mtr;
mtr.start();
+ const byte* xid_uuid = wsrep_xid_uuid(xid);
+ /* We must make check against wsrep_uuid here, the
+ trx_rseg_update_wsrep_checkpoint() writes over wsrep_uuid with
+ xid contents in debug mode and the memcmp() will never give nonzero
+ result. */
+ const bool must_clear_rsegs = memcmp(wsrep_uuid, xid_uuid,
+ sizeof wsrep_uuid);
const trx_rseg_t* rseg = trx_sys.rseg_array[0];
trx_rsegf_t* rseg_header = trx_rsegf_get(rseg->space, rseg->page_no,
@@ -107,10 +127,11 @@ void trx_rseg_update_wsrep_checkpoint(const XID* xid)
trx_rseg_update_wsrep_checkpoint(rseg_header, xid, &mtr);
- const byte* xid_uuid = wsrep_xid_uuid(xid);
- if (memcmp(wsrep_uuid, xid_uuid, sizeof wsrep_uuid)) {
+ if (must_clear_rsegs) {
+ XID null_xid;
+ memset(&null_xid, 0, sizeof null_xid);
+ null_xid.null();
memcpy(wsrep_uuid, xid_uuid, sizeof wsrep_uuid);
-
/* Because the UUID part of the WSREP XID differed
from current_xid_uuid, the WSREP group UUID was
changed, and we must reset the XID in all rollback
@@ -118,10 +139,12 @@ void trx_rseg_update_wsrep_checkpoint(const XID* xid)
for (ulint rseg_id = 1; rseg_id < TRX_SYS_N_RSEGS; ++rseg_id) {
if (const trx_rseg_t* rseg =
trx_sys.rseg_array[rseg_id]) {
- trx_rseg_update_wsrep_checkpoint(
+ trx_rseg_write_wsrep_checkpoint(
trx_rsegf_get(rseg->space,
- rseg->page_no, &mtr),
- xid, &mtr);
+ rseg->page_no,
+ &mtr),
+ &null_xid,
+ &mtr);
}
}
}
diff --git a/storage/innobase/trx/trx0trx.cc b/storage/innobase/trx/trx0trx.cc
index 333c7fe90eb..7be760c6221 100644
--- a/storage/innobase/trx/trx0trx.cc
+++ b/storage/innobase/trx/trx0trx.cc
@@ -1441,11 +1441,8 @@ trx_commit_in_memory(
trx_mutex_enter(trx);
trx->dict_operation = TRX_DICT_OP_NONE;
-
#ifdef WITH_WSREP
- if (trx->mysql_thd && wsrep_on(trx->mysql_thd)) {
- trx->lock.was_chosen_as_deadlock_victim = FALSE;
- }
+ trx->lock.was_chosen_as_wsrep_victim = FALSE;
#endif
DBUG_LOG("trx", "Commit in memory: " << trx);
@@ -2155,7 +2152,7 @@ static my_bool trx_get_trx_by_xid_callback(rw_trx_hash_element_t *element,
transaction needs a valid trx->xid for
invoking trx_sys_update_wsrep_checkpoint(). */
if (!wsrep_is_wsrep_xid(trx->xid))
-#endif
+#endif /* WITH_WSREP */
/* Invalidate the XID, so that subsequent calls will not find it. */
trx->xid->null();
arg->trx= trx;