Summary: | LOR in zfs | ||
---|---|---|---|
Product: | Base System | Reporter: | Dan Langille <dvl> |
Component: | kern | Assignee: | freebsd-fs (Nobody) <fs> |
Status: | New --- | ||
Severity: | Affects Only Me | ||
Priority: | --- | ||
Version: | 10.3-RELEASE | ||
Hardware: | Any | ||
OS: | Any |
Description
Dan Langille
2016-12-28 16:18:11 UTC
In case it's relevant: The server has 3x LSI SAS 2008 cards. It boots off zfs, a 10-drive raidz2. There is a second 10-drive raidz3. 8 drives from one zpool are all on one HBA. 8 drives from the other zpool are on another HBA. Two drives from each zpool are on a third HBA (i.e. that HBA has two drives from one pool and two drives from the other). The zroot zpool has been around for several years, but was recently moved from one box to another, getting a new M/B & new HBAs. Shortly afterwards, the raidz3 was added to the box. Since that addition, the system has frozen most nights at about 0301, more or less. The host has 17 jails, all of which have stock /etc/crontab entries (i.e. daily periodic runs). By freeze, I mean: - ssh to the server fails to connect - login at console accepts password but does not present prompt, but responds to CTL-T (https://twitter.com/DLangille/status/804738019857199105) - Backups jobs which used this server as the destination, continued to work. - nagios flips out over the services provided by the jails/host - postfix stops responding (both host and jails) - tail -F /var/log/messages : in an existing ssh session continued to stream entries refs: * https://twitter.com/DLangille/media (lots of screen shots0 * https://twitter.com/search?q=%23frozenserver&src=typd * http://dan.langille.org/2016/12/09/system-freezes-up-with-lots-of-sonewconn-listen-queue-overflow/ * http://dan.langille.org/2016/12/14/server-freeze-2014-12-14/ Last night, I invoked the server freeze by doing a 'tar -czf'. I took a dump from that, and will provide upon request. See the output of 'show lockedvnods' here: https://twitter.com/DLangille/status/813870668513153024 |