Details
-
Bug
-
Status: Closed (View Workflow)
-
Blocker
-
Resolution: Fixed
-
10.8.1
Description
In MDEV-14425, a check that log writes succeed was inadvertently removed. Before that change, the code looked like this:
void log_t::file::write(os_offset_t offset, span<byte> buf) |
{
|
srv_stats.os_log_pending_writes.inc();
|
if (const dberr_t err= fd.write(offset, buf)) |
ib::fatal() << "write(" << fd.get_path() << ") returned " << err; |
srv_stats.os_log_pending_writes.dec();
|
srv_stats.os_log_written.add(buf.size());
|
srv_stats.log_writes.inc();
|
log_sys.n_log_ios++;
|
}
|
The counters were removed or relocated on purpose, but the error handling must be retained. If a log write fails, the only other reasonable alternative to crashing might be to switch to read-only mode.
Attachments
Issue Links
- is caused by
-
MDEV-14425 Change the InnoDB redo log format to reduce write amplification
-
- Closed
-
- relates to
-
MDEV-26603 asynchronous redo log write
-
- Closed
-