Differences between revisions 15 and 17 (spanning 2 versions)
Revision 15 as of 2017-05-24 14:00:06
Size: 5004
Revision 17 as of 2017-05-29 10:01:54
Size: 5244
Comment: Typo.
Deletions are marked like this. Additions are marked like this.
Line 5: Line 5:

=== 2017-05-29 ===

 * Petter receive a box of disks from Ola, and insert one of them in the slow previously held by /dev/sdc on file01. It is a Western Digital WD2002FYPS-02W380 T TB disk with S/N WCAVY5484097, WWN 50014EE204F8523A.

Driftslogg for Frikanalen

Her føres større endringer relatert til drift av systemet i omvendt kronologisk rekkefølge.



  • Petter receive a box of disks from Ola, and insert one of them in the slow previously held by /dev/sdc on file01. It is a Western Digital WD2002FYPS-02W380 T TB disk with S/N WCAVY5484097, WWN 50014EE204F8523A.


  • Tore discovers a disk has failed over to a hot spare on file01, uses sudo ledctl locate=/dev/sdc to identify it.

  • Tore contacts Petter on IRC and report that /dev/sdc on the new storage server is having problems and should be blinking red. Petter have a look in the machine room, and find a disk blinking as predicted. It is pulled out. It is a Wester Digital WD2002FYPS-01U1B0 2 TB disk with S/N WCAVY0503173, WWN 50014EE20311A302 dated 2009-07-19.
  • While in the machine room Petter notice disk in slot 8 in the lower shelf of the old storage server is also blinking red.
  • Tore: The issue with simula is known, but I don't want to tempt fate by touching that crappy RAID controller any more than absolutely necessary. With sdc now out of the way until monday, sds has been promoted to a full member of the RAID (using sudo zpool detach file01-zpool0 sdc, in accordance with Oracle documentation) and it is no longer marked as DEGRADED.


* Power outage in the USIT machine room where Frikanalen keep its server installation. The entire room was without power. The initial power outage started 09:45 and the UPS(es) ran out of power 15-20 minutes later. The diesel power generator failed to start, and shortly after 10:00 all power was out. Petter noticed a while later and tried to call Ola, David, Tore and Benjamin. When things started to recover around 11:30, he notified IRC. As NUUG is located in the same room, email to frikanalen@nuug.no was down too.


- Machine named filserver installed in rack by Marius Halden and Ola Tellesbø

- First 20 disks installed in machine - Basic Debian OS installed by David Noble, connected to internet, fail2ban installed. Tested booting.

- operations.log created for keeping a local log of things done with the machine

- Granted ssh and sudo access to toresbe and halden; based on ssh pubkeys found on github.org

- Granted non-root ssh access to pere and halden; again based on ssh pubkeys found on github.org

- Confirmed working management subsystem using ipmitool - Confirmed SATA hotplug works as intended

- Tested that rcs works correctly on operations.log


- Configure BMC. root@file01:~# ipmitool lan print shows current configuration. Ethernet is connected to borch eth1; but this is probably not configured either.

- recommended by pere: Test online remove of disk. Test remove of one of the power sources.


Checksum errors på /dev/sdu. OS-devicet /dev/sdu blir sport tilbake av Petter til RAID-device 33:4 vha. 'ls -l /dev/disk/by-path/|grep sdu', 'megaclisas-status |grep u20' og til slutt søk etter 'Z1E6BMRQST2000DM001-1CH164 i "megacli -PDList -aALL'.

Vi prøvde først å fastslå hvorvidt problemet hadde vært midlertidig ved å kjøre en scrub (checksumming av all data); ytterligere korrupsjon ble påvist; disken erklært upålitelig. Disk ble fjernet fra kontrolleren av Tore vha. 'megacli -pdoffline -physdrv[33:4] -a0', 'megacli -pdmarkmissing -physdrv[33:4] -a0' og 'megacli -pdprprmv -physdrv[33:4] -a0' kjørt (usikkert om noen av disse er unødvendige?).

Petter bytter disk, og ny disk satt inn av Tore med 'megacli -CfgLdAdd -r0 [33:4] -a0' og 'zpool replace -f tank sdu'.


Petter kjører 'apt-get upgrade' på simula og oppgraderer alle Debian pakker og restarter tjenester som trenger det. Det trengs reboot for å aktivere ny kjerne.


Petter forsøker å aktivere den tredje nye disken som fortsatt var markert som FAILED av RAID-kontrolleren, ved å kjøre 'zpool offline tank 13666938400593999646', 'megacli -PDOffline -PhysDrv[33:7] -a0', 'megacli -PDOnline -PhysDrv[33:7] -a0' og til slutt 'zpool replace -f tank 13666938400593999646 sdad' etter å ha funnet riktig device-navn med dmesg. Resilver skal ta 11 timer. Uvisst hvorfor, så starter en resilver av de to andre nye diskene samtidig.


Petter kjører 'zpool replace -f tank 16694028023172514029 sdx' for å ta inn den andre av de nye diskene.


Petter kjører 'megacli -CfgLdAdd -r0 [33:12] -a0' og 'megacli -CfgLdAdd -r0 [33:5] -a0' for å gjøre diskene synlige for Linux, og 'zpool replace -f tank sdw' for å ta inn en av de nye diskene i ZFS. Resilver går galt, og vi oppdager at vi har mistet en fil (/tank/new_media/media/625464/broadcast/God sondag - tv_gs_140 airdate 2015_10_25.avi).


Petter ga David Noble sudo-tilgang på simula for å kunne aktivere erstatningsdiskene.


Ola og Petter byttet to disker i MD1000-hyllene til Frikanalen, disk 7 og 12 i nedre diskhylle blinket orange og ble byttet med to disker Ola hadde fått fra Tore.

grupper/video/frikanalen/driftslogg (last edited 2019-12-09 16:29:58 by PetterReinholdtsen)