database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-07-27 18:02:13 +03:00
Code Activity

MDEV-7326: Server deadlock in connection with parallel replication

Browse Source 
The bug occurs when a transaction does a retry after all transactions have
done mark_start_commit() in a batch of group commit from the master. In this
case, the retrying transaction can unmark_start_commit() after the following
batch has already started running and de-allocated the GCO. Then after retry,
the transaction will re-do mark_start_commit() on a de-allocated GCO, and also
wakeup of later GCOs can be lost.

This was seen "in the wild" by a user, even though it is not known exactly
what circumstances can lead to retry of one transaction after all transactions
in a group have reached the commit phase.

The lifetime around GCO was somewhat clunky anyway. With this patch, a GCO
lives until rpl_parallel_entry::last_committed_sub_id has reached the last
transaction in the GCO. This guarantees that the GCO will still be alive when
a transaction does mark_start_commit(). Also, we now loop over the list of
active GCOs for wakeup, to ensure we do not lose a wakeup even in the
problematic case.
This commit is contained in:
Kristian Nielsen
2015-01-07 14:45:39 +01:00
parent 4a3251595c
commit f27817c1d0
6 changed files with 420 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
sql/rpl_parallel.h
View File

@ -39,9 +39,12 @@ struct inuse_relaylog;
rpl_parallel_entry::count_committing_event_groups has reached
gco->next_gco->wait_count.
- When gco->wait_count is reached for a worker and the wait completes,
the worker frees gco->prev_gco; at this point it is guaranteed not to
be needed any longer.
- The gco lives until all its event groups have completed their commit.
This is detected by rpl_parallel_entry::last_committed_sub_id being
greater than or equal gco->last_sub_id. Once this happens, the gco is
freed. Note that since update of last_committed_sub_id can happen
out-of-order, the thread that frees a given gco can be for any later
event group, not necessarily an event group from the gco being freed.
*/
struct group_commit_orderer {
/* Wakeup condition, used with rpl_parallel_entry::LOCK_parallel_entry. */
@ -49,6 +52,16 @@ struct group_commit_orderer {
uint64 wait_count;
group_commit_orderer *prev_gco;
group_commit_orderer *next_gco;
/*
The sub_id of last event group in this the previous GCO.
Only valid if prev_gco != NULL.
*/
uint64 prior_sub_id;
/*
The sub_id of the last event group in this GCO. Only valid when next_gco
is non-NULL.
*/
uint64 last_sub_id;
bool installed;
};
@ -168,7 +181,8 @@ struct rpl_parallel_thread {
LOCK_rpl_thread mutex.
*/
void free_rgi(rpl_group_info *rgi);
group_commit_orderer *get_gco(uint64 wait_count, group_commit_orderer *prev);
group_commit_orderer *get_gco(uint64 wait_count, group_commit_orderer *prev,
uint64 first_sub_id);
/*
Put a gco on the local free list, to be later released to the global free
list by batch_free().