- In percona_bug1218330.test, consider replacing MTR variables
with SQL variables. Then the need to hide the exact values by
disable_query_log is removed from the adaptive_hash_mem_2 point
onwards, resulting in an easier-to-understand test result file.
- btr_search_info_get_ref_count header comment in btr0sea.c has
"the ." Also, this header comment should be synced with
btr0sea.h after editing.
- It seems that now every index structure in memory create must
be accompanied by a btr_search_index_init() call. This is OK,
but I'm wondering if there is any way to merge-proof this
further. For example, poison index->search_latch and
search_table pointers in dict_mem_fill_index_struct() if
UNIV_DEBUG?
- The ha_innodb.cc conflict needs either you fixing it, if
possible on GCA, otherwise me fixing it on merge. Note that
the snippet which was supposed to go to
innodb_release_stat_resources() went to
innodb_srv_conc_force_exit_innodb() instead.
- Why do innobase_release_stat_resources(),
innobase_commit_ordered(), and innobase_commit() appear to have
new calls to trx_search_latch_release_if_reserved() added
anyway, if they are? Or is it a bzr mismerge and only comments
next to the calls have been added?
- Diff line 1273: "These latches"
- It appears that sync_thread_levels_nonempty_trx() is dead code
now, and all its calls may be replaced by
sync_thread_levels_nonempty_gen(FALSE).
- trx->search_latch_timeout now becomes a never-changing constant
BTR_SEA_TIMEOUT. It can be removed and
I_S.INNODB_TRX.TRX_ADAPTIVE_HASH_TIMEOUT field can return
BTR_SEA_TIMEOUT directly, always.
- Not fully sure but it appears that there are some
locking/unlocking assymetries wrt UNIV_SEARCH_DEBUG, preceding
this MP. Not if actionable, probably no need to waste time on
this. One suspicious example is the target code of
release_search_latch goto label.
- In percona_ bug1218330. test, consider replacing MTR variables query_log is removed from the adaptive_hash_mem_2 point to-understand test result file. info_get_ ref_count header comment in btr0sea.c has index_init( ) call. This is OK, fill_index_ struct( ) if release_ stat_resources( ) went to srv_conc_ force_exit_ innodb( ) instead. release_ stat_resources( ), commit_ ordered( ), and innobase_commit() appear to have latch_release_ if_reserved( ) added levels_ nonempty_ trx() is dead code thread_ levels_ nonempty_ gen(FALSE) . latch_timeout now becomes a never-changing constant SEA_TIMEOUT. It can be removed and S.INNODB_ TRX.TRX_ ADAPTIVE_ HASH_TIMEOUT field can return SEA_TIMEOUT directly, always. unlocking assymetries wrt UNIV_SEARCH_DEBUG, preceding search_ latch goto label.
with SQL variables. Then the need to hide the exact values by
disable_
onwards, resulting in an easier-
- btr_search_
"the ." Also, this header comment should be synced with
btr0sea.h after editing.
- It seems that now every index structure in memory create must
be accompanied by a btr_search_
but I'm wondering if there is any way to merge-proof this
further. For example, poison index->search_latch and
search_table pointers in dict_mem_
UNIV_DEBUG?
- The ha_innodb.cc conflict needs either you fixing it, if
possible on GCA, otherwise me fixing it on merge. Note that
the snippet which was supposed to go to
innodb_
innodb_
- Why do innobase_
innobase_
new calls to trx_search_
anyway, if they are? Or is it a bzr mismerge and only comments
next to the calls have been added?
- Diff line 1273: "These latches"
- It appears that sync_thread_
now, and all its calls may be replaced by
sync_
- trx->search_
BTR_
I_
BTR_
- Not fully sure but it appears that there are some
locking/
this MP. Not if actionable, probably no need to waste time on
this. One suspicious example is the target code of
release_