[Maria-developers] IB: trx_t:: start_time and start_time_micro
Hello! Different semantics of these 2 members doesn't look good: time_t start_time; /*!< time the state last time became TRX_STATE_ACTIVE */ ib_uint64_t start_time_micro; /*!< start time of transaction in microseconds */ Either `start_time_micro` should be drawn to the base of `start_time` or it should be explicitly commented that it has nothing to do with `start_time`. I.e. there is nothing wrong on algorithmic level, but such data members may (and certainly would) lead developers to misuse of `start_time_micro`. -- All the best, Aleksey Midenkov @midenok
Hi Aleksey, I think that we should remove trx_t::start_time, and always use trx_t::start_time_micro. I shortly discussed this with Vladislav Vaintroub, and he mentioned that the microsecond timestamp in THD::start_utime is updated at the start of each statement. I also looked at his MDEV-12345 performance fix <https://github.com/MariaDB/server/commit/e5b67a46bcb75a751f5906f19694971bc803d67c>. It looks like trx_start_low() is already doing the right thing, except that the trx_t::start_time should be removed. I'd welcome a contributed patch that removes trx_t:.start_time. Or if that really is too tricky, convert start_time_micro to start_time. Marko On Thu, Nov 16, 2017 at 11:29 AM, Aleksey Midenkov <midenok@gmail.com> wrote:
Hello!
Different semantics of these 2 members doesn't look good:
time_t start_time; /*!< time the state last time became TRX_STATE_ACTIVE */ ib_uint64_t start_time_micro; /*!< start time of transaction in microseconds */
Either `start_time_micro` should be drawn to the base of `start_time` or it should be explicitly commented that it has nothing to do with `start_time`. I.e. there is nothing wrong on algorithmic level, but such data members may (and certainly would) lead developers to misuse of `start_time_micro`.
-- All the best,
Aleksey Midenkov @midenok
_______________________________________________ Mailing list: https://launchpad.net/~maria-developers Post to : maria-developers@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-developers More help : https://help.launchpad.net/ListHelp
-- Marko Mäkelä, Lead Developer InnoDB MariaDB Corporation DON’T MISS M|18 MariaDB User Conference February 26 - 27, 2018 New York City https://m18.mariadb.com/
Hello Marko! On Tue, Nov 21, 2017 at 11:59 AM, Marko Mäkelä <marko.makela@mariadb.com> wrote:
Hi Aleksey,
I think that we should remove trx_t::start_time, and always use trx_t::start_time_micro.
If that will be the case, then start_time_micro must be the absolute time, not the relative one like it is now. -- All the best, Aleksey Midenkov @midenok
I shortly discussed this with Vladislav Vaintroub, and he mentioned that the microsecond timestamp in THD::start_utime is updated at the start of each statement. I also looked at his MDEV-12345 performance fix. It looks like trx_start_low() is already doing the right thing, except that the trx_t::start_time should be removed.
I'd welcome a contributed patch that removes trx_t:.start_time. Or if that really is too tricky, convert start_time_micro to start_time.
Marko
On Thu, Nov 16, 2017 at 11:29 AM, Aleksey Midenkov <midenok@gmail.com> wrote:
Hello!
Different semantics of these 2 members doesn't look good:
time_t start_time; /*!< time the state last time became TRX_STATE_ACTIVE */ ib_uint64_t start_time_micro; /*!< start time of transaction in microseconds */
Either `start_time_micro` should be drawn to the base of `start_time` or it should be explicitly commented that it has nothing to do with `start_time`. I.e. there is nothing wrong on algorithmic level, but such data members may (and certainly would) lead developers to misuse of `start_time_micro`.
-- All the best,
Aleksey Midenkov @midenok
_______________________________________________ Mailing list: https://launchpad.net/~maria-developers Post to : maria-developers@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-developers More help : https://help.launchpad.net/ListHelp
-- Marko Mäkelä, Lead Developer InnoDB MariaDB Corporation
DON’T MISS
M|18
MariaDB User Conference
February 26 - 27, 2018
New York City
participants (2)
-
Aleksey Midenkov -
Marko Mäkelä