Details
-
Improvement
-
Resolution: Fixed
-
Minor
-
Lustre 2.15.0
-
3
-
9223372036854775807
Description
For debugging purposes, it would be useful to store the ctime for an llog object when it is created. Otherwise, the objects currently do not set any timestamp on the log even though it was recently created (e.g. the 39xx logs, the low-numbered 2015 objects were created with an older Lustre version):
9851 4 -rw-r--r-- 1 root root 8 Nov 7 2013 /mnt/tmp/O/1/LAST_ID 1850 20 -rw-r--r-- 1 root root 19584 May 6 2015 /mnt/tmp/O/1/d4/164 98363 12 -rw-r--r-- 1 root root 8448 Dec 31 1969 /mnt/tmp/O/1/d4/3972 1665 20 -rw-r--r-- 1 root root 17792 May 6 2015 /mnt/tmp/O/1/d0/160 1849 12 -rw-r--r-- 1 root root 8256 Apr 23 2015 /mnt/tmp/O/1/d30/158 1861 20 -rw-r--r-- 1 root root 17472 May 6 2015 /mnt/tmp/O/1/d6/166 98365 12 -rw-r--r-- 1 root root 8448 Dec 31 1969 /mnt/tmp/O/1/d6/3974 2130157 12 -rw-r--r-- 1 root root 8384 Dec 31 1969 /mnt/tmp/O/1/d8/3976
The same with logs in CONFIGS:
# debugfs -c -R "ls -l CONFIGS" /dev/loop0 debugfs 1.46.6-wc1 (10-Jan-2023) 13 40755 (2) 0 0 4096 5-Mar-2024 12:58 . 2 40755 (2) 0 0 4096 5-Mar-2024 12:58 .. 14 100644 (1) 0 0 12288 5-Mar-2024 12:58 mountdata 82 100644 (1) 0 0 8192 31-Dec-1969 17:00 nodemap 85 100644 (1) 0 0 11008 31-Dec-1969 17:00 params 86 100644 (1) 0 0 69600 31-Dec-1969 17:00 testfs-client 87 100644 (1) 0 0 75560 31-Dec-1969 17:00 testfs-MDT0000 154 100644 (1) 0 0 75024 31-Dec-1969 17:00 testfs-MDT0001 156 100644 (1) 0 0 74488 31-Dec-1969 17:00 testfs-MDT0002 158 100644 (1) 0 0 73952 31-Dec-1969 17:00 testfs-MDT0003 160 100644 (1) 0 0 17992 31-Dec-1969 17:00 testfs-OST0000 161 100644 (1) 0 0 17680 31-Dec-1969 17:00 testfs-OST0001 169 100644 (1) 0 0 17144 31-Dec-1969 17:00 testfs-OST0002 170 100644 (1) 0 0 16608 31-Dec-1969 17:00 testfs-OST0003
Similarly, it would also be useful to update the mtime when a catalog is modified, since this doesn't happen very often and otherwise they show the timestamp when the file was created (maybe not even that for newer versions of Lustre):
4 -rw-r--r--. 1 root 160 May 6 2015 CATALOGS 180 -rw-r--r--. 1 root 176576 Nov 7 2013 changelog_catalog 12 -rw-r--r--. 1 root 8384 Nov 7 2013 changelog_users
There is no overhead to update the mtime when new records are added to an llog file, since the new inode size/blocks needs to be written anyway.
squalfof please check https://review.whamcloud.com/c/fs/lustre-release/+/57510
I benchmarked with ldiskfs, you can use sanity/60a as a reference.