Details
-
Bug
-
Status: Open (View Workflow)
-
Critical
-
Resolution: Unresolved
-
13.0, 13.1
-
Can result in hang or crash
Description
mleich shared an rr replay trace where InnoDB intentionally crashes itself because we failed to allocate an innodb_log_archive log file. In this case, we did not have errno=ENOSPC, but instead, errno=EINTR was set in the following stack trace:
#0 0x00007d8af591c6f4 in fallocate64 (fd=fd@entry=3, mode=mode@entry=0, offset=offset@entry=0, len=len@entry=25993216) at ../sysdeps/unix/sysv/linux/fallocate64.c:27
|
#1 0x00005e439db8eeb6 in os_file_set_size (name=0x758ae00035f0 "ib_00000000018ca000.log", file=file@entry=3, size=25993216, is_sparse=is_sparse@entry=false)
|
at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/os/os0file.cc:1582
|
#2 0x00005e439db45839 in log_t::archive_create (this=this@entry=0x5e439f4e29c0 <log_sys>, lsn=<optimized out>, ex=ex@entry=true) at /usr/include/c++/13/bits/basic_string.h:222
|
#3 0x00005e439db45dba in log_t::archive_new_write (this=this@entry=0x5e439f4e29c0 <log_sys>, buf=0x758ae5e00200 "e", buf@entry=0x758ae5e00000 "\032\02675", length=1024, length@entry=1536,
|
offset=offset@entry=25992704) at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/log/log0log.cc:1331
|
#4 0x00005e439db46676 in log_t::write_buf<(log_t::resizing_and_latch)1> (this=0x5e439f4e29c0 <log_sys>) at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/log/log0log.cc:1822
|
#5 log_writer () at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/log/log0log.cc:1970
|
#6 0x00005e439db43342 in log_write_up_to (lsn=lsn@entry=25993777, durable=durable@entry=true, callback=<optimized out>, callback@entry=0x0)
|
at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/log/log0log.cc:1929
|
#7 0x00005e439dce656b in log_checkpoint_low (oldest_lsn=oldest_lsn@entry=25993113, end_lsn=end_lsn@entry=25993113) at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/buf/buf0flu.cc:2333
|
#8 0x00005e439dce8960 in buf_flush_sync_for_checkpoint (lsn=lsn@entry=25993113) at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/buf/buf0flu.cc:2597
|
#9 0x00005e439dce8d79 in buf_flush_page_cleaner () at /data/Server/MDEV-14992C_MDEV-40932/storage/innobase/buf/buf0flu.cc:2893
|
The Linux documentation says:
|
fallocate(2) ERRORS |
EINTR A signal was caught during execution; see signal(7).
|
It was the intention of MDEV-37949 to intentionally crash when we run out of space, but not when there is an attempt to shut down the server while we happen to be creating a new archive log file. For normal data file extension, we do want os_file_set_size() to terminate during server shutdown.
Each invocation of os_file_set_size() needs to be reviewed and adjusted for this. I think that it could be simplest to add a parameter to the function that requests it to retry on soft errors, such as EINTR.
Attachments
Issue Links
- is caused by
-
MDEV-37949 Implement innodb_log_archive
-
- Closed
-
