Re: lots of data written, but no space occupied

[André Weidemann]

On 23.02.2018 00:50, Chris Murphy wrote:

The only thing I can think of is something's updating metadata due to
relatime mount option. Maybe try noatime? At 9GiB, really it's 4.5GiB
because whatever is being written is being doubled by raid1 profile
and multiple devices. There is a case of wandering trees where a
little bit of change can get amplified, but this requires an
instigator, the file system won't do it on its own.

What were the mkfs options? Are you using the default 16KiB nodesize?

I found the cause...  And "pebkac" is the only way to describe it.



In the end it only took a small effort to figure it out.

I then grep inside the syslog using:
"tail -f /var/log-ram/syslog|egrep -v "btrfs-transacti|kwork"|grep sda"
until something like this fell out:




Thanks anyway.
  André

As I wrote in my initial email, I installed collectd. And I was sure to have disabled the rrdtool plugin, which apparently I hadn't. So there was no additional space occupied, because collectd only updated already existing files. I knew that calling "echo 1 > /proc/sys/vm/block_dump" would dump all file access to syslog. Since /var/log/syslog was on the effected drives, I created a ram-disk and had rsyslog log to the RAM-disk. I did this so I wouldn't see my own writes to the log while running "grep sda". Feb 23 15:05:35 server2 kernel: [496223.186943] collectd(604): dirtied inode 870979 (smart_attribute-temperature-celsius-2.rrd) on sda2

smime.p7s
Description: S/MIME Cryptographic Signature