Zimbra is a email collaboration suite. Its various compontents perform MTA duties, message store, full text indexing. In a large environment, the number of files and I/O operations can really add up. How we ensure the filesystem is ready to support it?
Zimbra offers some guidance for tuning the filesystem, with tips like:
Mount file systems with the
It generally is not important to know the last access time of all the files, so the extra write ops are wasteful.
dirsyncfor ext3/ext4 file systems. (mailstores, indexes, MTA queues)
The mailbox and MTA fsync for files, but it is possible to lose the directory entry for the file in crash. It is also possible to affect this with
chattr -R +d.
More dirsync info.
||Do not update access time|
||Immediately flush directory operations|
The following filesystem creation options are also recommended.
||Use hashed b-trees to speed up lookups in large directories.|
||By default 5% of space is reserved. That can be a lot on a big filesystem.|
||Bytes per Inode or
||Journal size can influence metadata performance, so boost the size.|
How many filesystems do we need? Which filesystems need which options? Let’s see what directories make good candidates for separation.
How fine grained do you split up the filesystems? Below are some key directories within Zimbra which may or may not be candidates for unique filesystems.
|Dir||I/O Type||Latency Sensitivity||Function|
|/opt/zimbra/backup||High write||Low||Nightly dump of all other dirs. (Use a NFS mount)|
|/opt/zimbra/db||Random||High||Message metadata in MySQL. Disambiguates message blob location and tags etc|
|/opt/zimbra/data||Random||High||Data for amavisd, clamav, LDAP, postfix MTA, etc|
|/opt/zimbra/data/amavisd/tmp||Random||High||Temp files created when Amavisd feeds mail to ClamAV and Spamassassin can be sped up with a RAM disk|
|/opt/zimbra/index||High Random||High||Lucene full text index|
|/opt/zimbra/redolog||High Write||High||Transaction log of all activity|
|/opt/zimbra/store||Random||High||Message blob store|
Separating out Filesystems
For starters I’m going to go with:
- /opt/zimbra Nothing special yet. Just noatime.
- /opt/zimbra/backup Backups will go to a NFS filer. I’m not concerned about the filesystem config there. That a job for the NAS admin.
Selecting Filesystem Options
Based on those recommendations, let’s decide which to follow, and how.
I will be working on RHEL 6, and I’m ignoring that fact that anything other than ext4 exists.
The dir_index option speeds up directories with large numbers of files. It turns out this is on by default in
/etc/mke2fs.conf. You can confirm by looking for
dir_index in the filesystem features in the output of
tune2fs. So, nothing to do here.
Filesystem features: has_journal ext_attr resize_inode dir_index filetype needs_recovery extent flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize
Bytes per Inode
Inodes store metadata about files. Every file consumes an inode and you can’t add more inodes. Zimbra stores message blobs as individual files, as opposed to one file per mail folder (ala mbox). Therefore the inode usage can be very high. You may have tons of space free, but if you run out of inodes, pack it in. Use
df -i to examine your inode usage.
-i 10240 mke2fs option. This says create 1 inode for every 10KB of space on the filesystem. This assumes that you expect to fill the filesystem up with 10KB files.
How many inodes do I have on my test filesystem?
[root@zimbra-mbox-10 ~]# tune2fs -l /dev/mapper/VGzstore-LVstore | grep 'Inode count' Inode count: 67108864
If I use the
-i 10240 option, how many inodes will I have then? You can find out with
mke2fs in dry run mode. See line 4 below.
[root@zimbra-mbox-10 ~]# umount /opt/zimbra/store [root@zimbra-mbox-10 ~]# mke2fs -n -i 10240 /dev/mapper/VGzstore-LVstore | grep inodes mke2fs 1.41.12 (17-May-2010) 107479040 inodes, 268435456 blocks 13120 inodes per group
Wow! That is support for up to 107 million files.
How big is your average message? It looks like in my case we are looking at about 190 million emails split across numerous servers, and an average message size of 106KB. So an inode per 10KB is pretty generous. I can probably increase that ratio, and lower the inode count.
Doing the math, it looks like by default I’m getting one inode per 16K. This is borne out in
[defaults] base_features = sparse_super,filetype,resize_inode,dir_index,ext_attr blocksize = 4096 inode_size = 256 inode_ratio = 16384
So the default may be just fine. I could even raise the number, but I won’t.
Journal size reportedly can have a great impact on metadata operations. Zimbra suggests an arbitrary 400M for journal size for filesystems.
If I’ve already created a filesystem, so how do I find out the size of the existing journal?
tune2fs -l doesn’t do the trick.
[root@zimbra-mbox-10 ~]# tune2fs -l /dev/mapper/VGzstore-LVstore tune2fs 1.41.12 (17-May-2010) Filesystem volume name: <none> Last mounted on: <not available> Filesystem UUID: f859c1f6-28b5-4c4b-8bd2-618abcd33607 Filesystem magic number: 0xEF53 Filesystem revision #: 1 (dynamic) Filesystem features: has_journal ext_attr resize_inode dir_index filetype needs_recovery extent flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize Filesystem flags: signed_directory_hash Default mount options: (none) Filesystem state: clean Errors behavior: Continue Filesystem OS type: Linux Inode count: 67108864 Block count: 268435456 Reserved block count: 2684354 Free blocks: 264172519 Free inodes: 67108853 First block: 0 Block size: 4096 Fragment size: 4096 Reserved GDT blocks: 960 Blocks per group: 32768 Fragments per group: 32768 Inodes per group: 8192 Inode blocks per group: 512 Flex block group size: 16 Filesystem created: Mon Feb 2 21:41:09 2015 Last mount time: Mon Feb 2 21:41:38 2015 Last write time: Mon Feb 2 21:41:38 2015 Mount count: 1 Maximum mount count: 32 Last checked: Mon Feb 2 21:41:09 2015 Check interval: 15552000 (6 months) Next check after: Sat Aug 1 22:41:09 2015 Lifetime writes: 16 GB Reserved blocks uid: 0 (user root) Reserved blocks gid: 0 (group root) First inode: 11 Inode size: 256 Required extra isize: 28 Desired extra isize: 28 Journal inode: 8 Default directory hash: half_md4 Directory Hash Seed: 2394798b-a767-4a9c-b92f-dabd75e5f1e4 Journal backup: inode blocks
After a little googling, I found this page that suggested
dumpe2fs. It looks like the default size for this 1TB filesystem was 128M.
[root@zimbra-mbox-10 ~]# dumpe2fs /dev/mapper/VGzstore-LVstore | grep ^Journal dumpe2fs 1.41.12 (17-May-2010) Journal inode: 8 Journal backup: inode blocks Journal features: (none) Journal size: 128M Journal length: 32768 Journal sequence: 0x00000001 Journal start: 0
What size is actually a good size? Well, this article suggested that 256MB may be a good number. Also somewhat arbitrary.
Let’s just go with 256MB for now. I don’t believe you can adjust the journal size after the fact, but in my case it doesn’t matter. I’m going to recreate it.
I’ll of course use the ansible filesystem module and add a value of
-J size=256 in the
So, in the end I’ll be using Ansible filesystem and mount modules to setup filesytems whith a journal size of 256MB, and a reserve of 1%, and use noatime and dirsync mount options. I create a dictionary like this to do that.
zimbra_storage: nfs: backup: export: nfs-server:/zimbra/backup path: /opt/zimbra/backup devices: opt_disk: dev: /dev/sdb size: 500G vg: VGzopt volumes: opt: name: LVzimbra path: /opt/zimbra size: 50G fs_type: ext4 fs_opts: mount_opts: "noatime,dirsync" index: name: LVindex path: /opt/zimbra/index size: 100G fs_type: ext4 fs_opts: "-J size=256" mount_opts: "noatime,dirsync" redo: name: LVredo path: /opt/zimbra/redolog size: 200G fs_type: ext4 fs_opts: mount_opts: "noatime" store_disk: dev: /dev/sdc size: 2T vg: VGzstore volumes: store: name: LVstore path: /opt/zimbra/store size: 1T fs_type: ext4 fs_opts: "-J size=256 -m 1" mount_opts: "noatime,dirsync"
- 30 Oct 2018 » Creating OpenStack Provider Network for Use by a Single Project
- 16 Feb 2018 » Load balancing of OpenShift HA Routers Mind the GARP
- 13 Oct 2017 » OpenShift 3.6 Upgrade Metrics Fails Missing heapster-certs Secret
- 20 Aug 2017 » Installing OpenShift on OpenStack
- 14 Aug 2017 » OpenStack Network Diagram
- 09 Aug 2017 » How to push an image to an unexposed OpenShift Docker registry
- 22 Mar 2017 » Automated Pruning of OpenShift Artifacts; Builds, Deploys, Images
- 29 Jan 2017 » Configuring OpenShift with Multiple Sharded Routers
- 14 Nov 2016 » OpenShift Cluster Metrics and Cassandra Troubleshooting
- 11 Jul 2016 » How to List Tags On Redhat Registry Images