update_vertica detects scheduler and readahead wrong

ivan_kovalenko

We're updating vertica from 6.1.2 to 7. When performing /opt/vertica/sbin/update_vertica i do recieve following errors permanently
FAIL (S0150): <A href="https://my.vertica.com/docs/7.0.x/HTML/index.htm#cshid=S0150" target="_blank" rel="nofollow noopener noreferrer">https://my.vertica.com/docs/7.0.x/HTML/index.htm#cshid=S0150</A><BR /> These disks do not have 'deadline' or 'noop' IO scheduling: '/dev/sdb1'<BR /> ('sda') = 'cfq', '/dev/sda1' ('sda') = 'cfq', '/dev/sda3' ('sda') =<BR /> 'cfq'<BR /> FAIL (S0020): <A href="https://my.vertica.com/docs/7.0.x/HTML/index.htm#cshid=S0020" target="_blank" rel="nofollow noopener noreferrer">https://my.vertica.com/docs/7.0.x/HTML/index.htm#cshid=S0020</A><BR /> Readahead size of /dev/sda1 is too low for typical systems: 256 < 2048<BR /> Readahead size of /dev/sda3 is too low for typical systems: 256 < 2048<BR /><BR /><BR /> But in the same time
# mount|grep sdb1
/dev/sdb1 on /opt type ext4 (rw)
# cat /sys/block/sdb/queue/scheduler
noop anticipatory [deadline] cfq

and /dev/sda1 and /dev/sda3 are used not by vertica but OS. Why that is a FAIL failure? Should not it be WARNING?

[Deleted User]

Hi,

We actually have an open New Feature request, ver-28840, related to the scope of the checks. Specifically highlighted was the scheduler. It was suggested that instead of checking the i/o scheduler for all volumes. installer should only check the i/o scheduler for volumes that host the datadir. On a new install this is not known so the check of all volumes is appropriate, but on upgrade it should be known and could be scoped accordingly. Similarly for readahead. I believe these are FAIL vs. WARN because of the performance implications. The "--force-threshold NONE" can be used to bypass.

ivan_kovalenko

I've modified threshold and successfully updated vertica. But that was confusing a bit reading FAILed checks.

Thanks for reply.

haim_ari

B.T.W that is failure-threshold not force