[systemd-devel] systemd and lots of kernel devices

[Kay Sievers]

Just a note: last week we needed to fix the event queue size in
systemd to be able to handle many udev events happen at the same time.
Systemd hit an assert and switched to the rescue console.

Systemd keeps internal track about all block devices (udev rules tag
all block devices) and creates management units for them. To check how
we behave with a large number of devices, I used scsi_debug (1.5 GHz
Dual Core laptop):
  $ time modprobe scsi_debug max_luns=16 add_host=16 num_parts=4 num_tgts=16
  14m

We have ~20.000 block devices now:
  $ find /sys/class/block/ | wc -l
  20489

The logs look fine, and 'systemctl' outputs all these units just fine:
  $ systemctl | wc -l
  20592

Seems we pass all that without any errors now. People with 'insane'
setups, please check now and report any still existing failures.

Oh, during such tests, never try to run a desktop, or hal, at the same
time, the box will not survive. :)

Cheers,
Kay