void mur_tp_resolve_time(jnl_tm_t max_lvrec_time)
{
int reg_total;
reg_ctl_list *rctl, *rctl_top;
jnl_tm_t reg_tp_resolve_time;
reg_total = murgbl.reg_total;
if (mur_options.forward)
{
if (!mur_options.verify)
{
jgbl.mur_tp_resolve_time = mur_ctl[0].lvrec_time;
for (rctl = mur_ctl + 1, rctl_top = mur_ctl + reg_total; rctl < rctl_top; rctl++)
{
if (rctl->lvrec_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = rctl->lvrec_time;
}
assert(jgbl.mur_tp_resolve_time);
} else
jgbl.mur_tp_resolve_time = 0; /* verify continues till the beginning of journal file */
return;
}
mur_sort_files();
jgbl.mur_tp_resolve_time = MAXUINT4;
assert(max_lvrec_time == mur_ctl[reg_total - 1].lvrec_time);
for (rctl = mur_ctl, rctl_top = mur_ctl + reg_total; rctl < rctl_top; rctl++)
{
/* Assumption : It is guaranteed to see an EPOCH in every
* "rctl->jctl->jfh->epoch_interval + MAX_EPOCH_DELAY" seconds. */
assert(max_lvrec_time >= rctl->jctl->jfh->epoch_interval + MAX_EPOCH_DELAY);
reg_tp_resolve_time = max_lvrec_time - rctl->jctl->jfh->epoch_interval - MAX_EPOCH_DELAY;
if (rctl->lvrec_time > reg_tp_resolve_time)
reg_tp_resolve_time = rctl->lvrec_time;
if (reg_tp_resolve_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = reg_tp_resolve_time;
assert(!mur_options.update || NULL != rctl->csd);
if (mur_options.update && rctl->recov_interrupted && rctl->csd->intrpt_recov_tp_resolve_time &&
rctl->csd->intrpt_recov_tp_resolve_time < jgbl.mur_tp_resolve_time)
/* Previous backward recovery/rollback was interrupted.
* Update tp_resolve_time to reflect the minimum of the previous and
* current recovery/rollback's turn-around-points.
* It is possible that both rctl->csd->intrpt_recov_resync_seqno and
* rctl->csd->intrpt_recov_tp_resolve_time are zero in case previous recover
* was killed after mur_open_files, (which sets csd->recov_interrupted)
* but before mur_back_process() which would have set csd->intrpt_recov_tp_resolve_time */
jgbl.mur_tp_resolve_time = rctl->csd->intrpt_recov_tp_resolve_time;
}
if (mur_options.since_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = (jnl_tm_t)mur_options.since_time;
if (FENCE_NONE == mur_options.fences && !mur_options.since_time_specified && !murgbl.intrpt_recovery)
jgbl.mur_tp_resolve_time = max_lvrec_time;
}
void mur_tp_resolve_time(jnl_tm_t max_lvrec_time)
{
int reg_total;
reg_ctl_list *rctl, *rctl_top;
jnl_tm_t reg_tp_resolve_time;
reg_total = murgbl.reg_total;
if (mur_options.forward)
{
if (!mur_options.verify)
{
jgbl.mur_tp_resolve_time = mur_ctl[0].lvrec_time;
for (rctl = mur_ctl + 1, rctl_top = mur_ctl + reg_total; rctl < rctl_top; rctl++)
{
if (rctl->lvrec_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = rctl->lvrec_time;
}
assert(jgbl.mur_tp_resolve_time);
} else
jgbl.mur_tp_resolve_time = 0; /* verify continues till the beginning of journal file */
return;
}
mur_sort_files();
jgbl.mur_tp_resolve_time = MAXUINT4;
assert(max_lvrec_time == mur_ctl[reg_total - 1].lvrec_time);
for (rctl = mur_ctl, rctl_top = mur_ctl + reg_total; rctl < rctl_top; rctl++)
{
/* Assumption : It is guaranteed to see an EPOCH in every
* "rctl->jctl->jfh->epoch_interval + MAX_EPOCH_DELAY" seconds. */
assert(max_lvrec_time >= rctl->jctl->jfh->epoch_interval + MAX_EPOCH_DELAY);
/* Calculate this region's TP resolve time based on the maximum of the last valid record across regions. If the
* region is properly closed (typically this means that the journal file's crash field is FALSE. But, with online
* rollback, crash field being TRUE, does not mean the journal file is crashed (as long as the shared memory for
* that region existed when the region was opened). So, use jctl->properly_closed to determine whether the journal
* file for the region is really crashed. If it is properly closed, the region's TP resolve time is the
* max_lvrec_time. If not, we need to go back by an epoch interval in addition to the MAX_EPOCH_DELAY.
*/
if (!rctl->jctl->properly_closed)
reg_tp_resolve_time = max_lvrec_time - rctl->jctl->jfh->epoch_interval - MAX_EPOCH_DELAY;
else
reg_tp_resolve_time = max_lvrec_time;
if (rctl->lvrec_time > reg_tp_resolve_time)
reg_tp_resolve_time = rctl->lvrec_time;
if (reg_tp_resolve_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = reg_tp_resolve_time;
assert(!mur_options.update || NULL != rctl->csd);
if (mur_options.update && rctl->recov_interrupted && rctl->csd->intrpt_recov_tp_resolve_time &&
rctl->csd->intrpt_recov_tp_resolve_time < jgbl.mur_tp_resolve_time)
/* Previous backward recovery/rollback was interrupted.
* Update tp_resolve_time to reflect the minimum of the previous and
* current recovery/rollback's turn-around-points.
* It is possible that both rctl->csd->intrpt_recov_resync_seqno and
* rctl->csd->intrpt_recov_tp_resolve_time are zero in case previous recover
* was killed after mur_open_files, (which sets csd->recov_interrupted)
* but before mur_back_process() which would have set csd->intrpt_recov_tp_resolve_time */
jgbl.mur_tp_resolve_time = rctl->csd->intrpt_recov_tp_resolve_time;
}
if (mur_options.since_time < jgbl.mur_tp_resolve_time)
jgbl.mur_tp_resolve_time = (jnl_tm_t)mur_options.since_time;
if (FENCE_NONE == mur_options.fences && !mur_options.since_time_specified && !murgbl.intrpt_recovery)
jgbl.mur_tp_resolve_time = max_lvrec_time;
}
