[Bug 241476] zfs checksum errors with FreeBSD 11.3 and mps driver

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=241476

            Bug ID: 241476
           Summary: zfs checksum errors with FreeBSD 11.3 and mps driver
           Product: Base System
           Version: 11.3-RELEASE
          Hardware: amd64
                OS: Any
            Status: New
          Severity: Affects Some People
          Priority: ---
         Component: kern
          Assignee: bugs at FreeBSD.org
          Reporter: spam+bugs.freebsd.org at dentarg.net

Created attachment 208573
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=208573&action=edit
Output from "mpsutil show all"

We have two servers, Jon and Samwell.

The servers are SuperMicro SuperStorage Server 2027R-E1R24L with the X9DR7-LN4F
motherboard, 2x Xeon E5-2620v2 and 256GB ECC RAM. The motherboard has the LSI
SAS 2308 controller. Both servers have 22x 960GB Samsung SSD (SV843, SM843T,
SM863, SM863a).

Each server has a zpool ("db") with 11 mirrors consisting of the SSDs. They
boot of another zpool. The servers are using the mps driver.

The servers are configured to scrub regularly:

$ cat /etc/periodic.conf
daily_scrub_zfs_enable="YES"
daily_scrub_zfs_default_threshold="7" # days between scrubs
daily_status_zfs_enable="YES"

We upgraded both servers from FreeBSD 11.2 to 11.3 at 2019-09-04.

Every 10th minute a cronjob checks that the pools are healthy, if not, we get
an alert. After going to FreeBSD 11.3, both servers have alerted twice:

2019-09-20 04:10 Jon
2019-10-12 08:20 Jon

2019-10-09 05:10 Samwell
2019-10-24 04:10 Samwell

about the error https://illumos.org/msg/ZFS-8000-8A (see attached files for
details; outputs from zpool status)

So far we have only seen errors reported for what I understand is metadata,
example:

errors: Permanent errors have been detected in the following files:

        <0x16dc>:<0x498>


Issuing another "zpool scrub" makes zpool status say "errors: No known data
errors" and then a "zpool clear" clears the checksum error counts.

We see no errors in /var/log/messages. Only "ZFS: vdev state changed" rows when
a scrub starts.

I've seen and read bug #239801, but opted to open a new bug as we are using a
different driver.

-- 
You are receiving this mail because:
You are the assignee for the bug.