| Bug #73087 | slave hits error 1595 when using low master_heartbeat_period | ||
|---|---|---|---|
| Submitted: | 24 Jun 2014 2:04 | Modified: | 26 Jun 2014 19:25 |
| Reporter: | Santosh Praneeth Banda | Email Updates: | |
| Status: | Closed | Impact on me: | |
| Category: | MySQL Server: Replication | Severity: | S2 (Serious) |
| Version: | 5.6.19 5.6.12 | OS: | Any |
| Assigned to: | CPU Architecture: | Any | |
| Tags: | replication | ||
[26 Jun 2014 7:33]
MySQL Verification Team
Hello Santosh, Thank you for the report. I couldn't reproduce this issue with simple replication(5.6.19->5.6.19, also later on included MTS), with moderate load(with as low as 4k max binlog size to trigger frequent log rotate etc) and with low master_heartbeat_period(0.001.. 0.25) etc. Could you please provide repeatable test case and related details? Thanks, Umesh
[26 Jun 2014 19:25]
Santosh Praneeth Banda
Hello Umesh, Sorry for the false bug report. I forget to remove a back port patch from 5.7.2 in my local branch which triggered the bug. Thanks, Santosh
[27 Jun 2014 4:17]
MySQL Verification Team
Thanks you for confirming this. Thanks, Umesh
Description: slave hits error 1595 when using low master_heartbeat_period. slave io_thread resets Format Description Event (FDE) to a lower version on receiving a Rotate Event from master. With low master_heartbeat_period, dump thread may send heartbeat event after sending Rotate event but before sending FDE. Io_thread cannot understand heartbeat event with old version of FDE causing error 1595. Buggy code in sql/rpl_slave.cc /* If mi_description_event is format <4, there is conversion in the relay log to the slave's format (4). And Rotate can mean upgrade or nothing. If upgrade, it's to 5.0 or newer, so we will get a Format_desc, so no need to reset mi_description_event now. And if it's nothing (same master version as before), no need (still using the slave's format). */ Format_description_log_event *old_fdle= mi->get_mi_description_event(); if (old_fdle->binlog_version >= 4) { DBUG_ASSERT(old_fdle->checksum_alg == mi->rli->relay_log.relay_log_checksum_alg); Format_description_log_event *new_fdle= new Format_description_log_event(3); new_fdle->checksum_alg= mi->rli->relay_log.relay_log_checksum_alg; mi->set_mi_description_event(new_fdle); } restarting io_thread fixes the problem. How to repeat: on slave change master to master_heartbeat_period=0.25 slave hits error 1595 on master's binary log rotate. Suggested fix: I don't get the reason for setting mi_description_event to a lower version in the first place. It should not be done to avoid this bug. The comment is the code is not clear too on why resetting mi_description_event is done.