A free space test on Ceph

First, the environment

Node Overview mon:ceph-node01 ceph-node02 ceph-node03osd:ceph-node01 ceph-node02 ceph-node03mds:ceph-node01 CEPH-NODE02 operating system: Ubuntu 14.10 Each OSD host has an OSD with a usable capacity of 15GB per OSD.

Second, the test process

1. Ceph-s View Overview

[email protected]:~# ceph -s    cluster  9ae8eb40-4f71-49ec-aa77-eda1cb6edbc3     health health_ok      monmap e1: 3 mons at {ceph-node01=192.168.239.161:6789/0,ceph-node02= 192.168.239.162:6789/0,ceph-node03=192.168.239.163:6789/0}, election epoch 50, quorum  0,1,2 ceph-node01,ceph-node02,ceph-node03     mdsmap e19: 1/1/1  up {0=ceph-node01=up:active}, 1 up:standby     osdmap e24:  3 osds: 3 up, 3 in      pgmap v202: 192  pgs, 3 pools, 672 kB data, 21 objects             3204 mb used, 45917 mb / 49122 mb  avail                  192 active+clean 

2. DF View available space

[email protected]:~# df -pmfilesystem                                          1048576-blocks  used  Available Capacity Mounted on/dev/sda1                                                     12277  1474     10804       13% /none                                                              1     0          1       0% /sys/fs/cgroupudev                                                           1959      1      1959       1%  /devtmpfs                                                          394      2       393        1% /runnone                                                              5     0          5       0% /run/locknone                                                           1970      0      1970       0% /run/shmnone                                                             100     0       100        0% /run/user/dev/sdb                                                      16374  1067     15308       7%  /data192.168.239.161,192.168.239.162,192.168.239.163:/           49120  3208     45912        7% /mnt/cephfs

the above shown The available space is certainly wrong, according to Ceph three principle, the real usable space should be less than 15GB, the following method to write 16GB files to verify.

3. DD Write file

Mount the file system to/MNT/CEPHFS, generate 8 dd files with a script, 2GB per file, in order to burst the OSD.

The script content is simple

#!/bin/bashcount=0max=8while [$count-lt $max];d o printf "Writing test${count}.dat\n" dd If=/dev/zero bs=1m count= 2048 Of=test${count}.dat ((count++)) done

Third, test results

Write to the last file, three nodes on the Ceph-mon process can not be seen with PS, check the log file, there are the following prompt:

2015-01-03 21:13:55.066943 7f0da98ce700  0 [email protected] (leader). Data_health ($)  update_stats avail 5% total 16766976 used 15915768 avail  8512082015-01-03 21:13:55.067245 7f0da98ce700 -1 [email protected] (leader). Data_ Health ( reached critical levels of available space on local ) Monitor storage -- shutdown!2015-01-03 21:13:55.067266 7f0da98ce700  0  ** shutdown via data health service **2015-01-03 21:13:55.067292  7f0da7ec9700 -1 [email protected] (leader)  e1 *** Got Signal  Interrupt ***2015-01-03 21:13:55.067300 7f0da7ec9700  1 [email protected] ( Leader)  e1 shutdown2015-01-03 21:13:55.067338 7f0da7ec9700  0 quorum  service shutdown2015-01-03  21:13:55.067339 7f0da7ec9700  0 [email protected] (shutdown). Health   healthmonitor::service_shutdown 1 services2015-01-03 21:13:55.067340 7f0da7ec9700   0 quorum service shutdown

It seems that the ceph-mon process itself out, the common local file system is almost full, the error is the application, why Ceph-mon designed to kill themselves?

A free space test on Ceph