diff options
| -rw-r--r-- | sql/slave.h | 26 |
1 files changed, 24 insertions, 2 deletions
diff --git a/sql/slave.h b/sql/slave.h index f61891acc91..2b1af08fb7a 100644 --- a/sql/slave.h +++ b/sql/slave.h @@ -170,10 +170,32 @@ typedef struct st_relay_log_info /* Handling of the relay_log_space_limit optional constraint. ignore_log_space_limit is used to resolve a deadlock between I/O and SQL - threads, it makes the I/O thread temporarily forget about the constraint + threads, the SQL thread sets it to unblock the I/O thread and make it + temporarily forget about the constraint. It is declared volatile because we + have this loop in the I/O thread (slave.cc): + while (rli->log_space_limit < rli->log_space_total && + !(slave_killed=io_slave_killed(thd,mi)) && + !rli->ignore_log_space_limit) + pthread_cond_wait(&rli->log_space_cond, &rli->log_space_lock); + According to Monty, on some systems pthread_cond_wait() could be inline, + and so the loop be optimized by the compiler, so rli->ignore_log_space_limit + could be considered constant and the loop never ends, even if the SQL thread + has set rli->ignore_log_space_limit to 1 (and called + pthread_cond_broadcast()) to break the loop in the I/O thread. + By declaring it volatile, we are sure that the variable will not be + considered constant and that the loop can be broken. + This is the same for all bool variables used by a thread to inform another + thread that something has changed: thd->killed, rli->abort_slave, + MYSQL_LOG::log_type; they are all volatile. + Quoting: + <serg> while (a>0) { wait_for_condition } + <serg> here a should be volatile ? + <monty> serg: in most system no, but on some yes + <serg> on what systems ? + <monty> On any system where pthread_mutex is a macro. */ ulonglong log_space_limit,log_space_total; - bool ignore_log_space_limit; + volatile bool ignore_log_space_limit; /* InnoDB internally stores the master log position it has processed |