disabling innobase_stats_on_metadata disables ANALYZE

[Stewart Smith]

ha_innobase::analyze(
/*=================*/
 THD* thd, /*!< in: connection thread handle */
 HA_CHECK_OPT* check_opt) /*!< in: currently ignored */
{
 /* Simply call ::info() with all the flags */
 info(HA_STATUS_TIME | HA_STATUS_CONST | HA_STATUS_VARIABLE);

 return(0);
}

versus this code in ::info():

 if (flag & HA_STATUS_TIME) {
  if (innobase_stats_on_metadata) {
   /* In sql_show we call with this flag: update
   then statistics so that they are up-to-date */

   prebuilt->trx->op_info = "updating table statistics";

   dict_update_statistics(ib_table);

   prebuilt->trx->op_info = "returning various info to MySQL";
  }

The solution is pretty simple, in ::info() do something like the following:

if (! innobase_stats_on_metadata)
    dict_update_statistics(prebuilt->table);

It does not seem like it was the intention of innobase_stats_on_metadata to also disable ANALYZE

[Yasufumi Kinoshita (yasufumi-kinoshita)]

XtraDB already has the following code.

        if (flag & HA_STATUS_TIME) {
- if (innobase_stats_on_metadata) {
+ if (innobase_stats_on_metadata
+ || thd_sql_command(user_thd) == SQLCOM_ANALYZE
+ ) {
                        /* In sql_show we call with this flag: update
                        then statistics so that they are up-to-date */

Why was the bug confirmed??

[Michael Widenius (monty)]

Hi!

>>>>> "Stewart" == Stewart Smith <email address hidden> writes:

Stewart> Public bug reported:
Stewart> ha_innobase::analyze(
Stewart> /*=================*/
Stewart> THD* thd, /*!< in: connection thread handle */
Stewart> HA_CHECK_OPT* check_opt) /*!< in: currently ignored */
Stewart> {
Stewart> /* Simply call ::info() with all the flags */
Stewart> info(HA_STATUS_TIME | HA_STATUS_CONST | HA_STATUS_VARIABLE);

Stewart> return(0);
Stewart> }

Stewart> versus this code in ::info():

Stewart> if (flag & HA_STATUS_TIME) {
Stewart> if (innobase_stats_on_metadata) {
Stewart> /* In sql_show we call with this flag: update
Stewart> then statistics so that they are up-to-date */

Stewart> prebuilt->trx->op_info = "updating table
Stewart> statistics";

Stewart> dict_update_statistics(ib_table);

Stewart> prebuilt->trx->op_info = "returning various info to MySQL";
Stewart> }

Stewart> The solution is pretty simple, in ::info() do something like the following:

Stewart> if (! innobase_stats_on_metadata)
Stewart> dict_update_statistics(prebuilt->table);

Stewart> It does not seem like it was the intention of innobase_stats_on_metadata to also disable ANALYZE

Good catch.

Vadim, can you add this fix to xtradb?

Regards,
Monty

[Michael Widenius (monty)]

Hi!

>>>>> "Yasufumi" == Yasufumi Kinoshita <email address hidden> writes:

Yasufumi> XtraDB already has the following code.
Yasufumi> if (flag & HA_STATUS_TIME) {
Yasufumi> - if (innobase_stats_on_metadata) {
Yasufumi> + if (innobase_stats_on_metadata
Yasufumi> + || thd_sql_command(user_thd) == SQLCOM_ANALYZE
Yasufumi> + ) {
Yasufumi> /* In sql_show we call with this flag: update
Yasufumi> then statistics so that they are up-to-date */

Yasufumi> Why was the bug confirmed??

I don't think you should depend on the SQLCOM_ANALYZE status; Someone
may want in the future to call analyze() from other points in the code.

A better way would be to add an engine specific
HA_STATUS_INNODB_ANALYZE flag from ha_innobase::analyze() and test
this in info().

Regards,
Monty

[Yasufumi Kinoshita (yasufumi-kinoshita)]

Monty,

XtraDB currently is maintained as patch cluster form InnoDB-Plugin by only few person.
It can be ignored and conflicted in the future InnoDB-Plugin.
Especially, the first-aid patch for the bug of InnoDB-Plugin should be small and clear to understand later,
and doesn't need to think future call.

This bug should be fixed at the main branch of InnoDB at first.
I will decide after fixing at the main branch when I will meet the conflict through porting the patches to the next version.
I have limited resource only, I don't want to spent them for dim specification which can be ineffectual in the future.

Regards,
Yasufumi

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-2558