2

Before I start the question I will just state that I am no sysadmin, but a layman seeking help.

I'm using Linux 3.10.0-514.26.2.el7.x86_64 and Amazon Web Services external backup service. In order for it to work I need a backup directory on my server, where all the files will be zipped, copied to and accessed from by the external backup service.

The question is - should I mount a separate partition on that directory? Backups can take a lot of HD space, and having a separate partition for this purpose should in theory secure my day-to-day server operations and websites from issues with low amount of free disk space.

But is it the right approach? Is there a better one?

4
  • 1
    Eh? What "external backup service" are you using? Mar 18, 2018 at 18:39
  • @Michael Hampton aws backup and recovery
    – P.Pal
    Mar 19, 2018 at 8:56
  • If you're improvising and just want to make sure you have some location where you want to dump files and be sure you don't fill the rest of your HD, you can make an image file, run mkfs.ext4 on it and mount it over loopback. But, in the end, using separate EBS volumes, LVM, partitions, etc (which depending on the situation), is better long term.
    – Halfgaar
    Mar 19, 2018 at 9:47
  • It's cloud. Don't partition, mount another disk.
    – Sammitch
    Mar 19, 2018 at 19:06

2 Answers 2

1

I would respectfully disagree with another answer and say that proper partitioning is essential to maintaining Linux servers in so many ways (cannot say much about Windows as that is not my experience).

I do always use separate partition for backups and this has been incredibly helpful not only for planning the space properly, but in recovery process. Let me outline below (forgive bad formatting as I type this from my phone):

  1. Separate partition allows for a capacity isolation where if something eats up all free space on that partition it won’t affect the rest of system. Think about /var/log for example. I’ve seen servers where users unintentionally broke logrotate and logs used up 100% of root (or that could happen with sudden increase in traffic for example).

  2. Separate partition on a separate disk in case of AWS will allow you to mount it to another instance and restore your data there (e.g. for forensic investigation)

  3. (Not only backup related) separate partition will let you to set noexec property when you mount it to minimize possibly intrusion (in fact that should be done for most of partitions on the system except those where your executables located)

Now since you mention that this is AWS system what I would recommend is instead of mounting separate EBS disk on the server is to utilize S3fs extension and mount an S3 bucket as a partition for backups. The benefit of this is great durability of data on s3.

Note 2 points: you MUST always monitor successful backup execution and you MUST periodically test data recoverability (read what happened to Gitlab for example)

Additionally if you decide to go with another EBS disk - avoid using LVM at all cost as 1 - fragmentation of LVM partition across multiple disks easily can cause loss of data (unfortunately LVM is not as good yet as its authors want it to be) and 2 - you can now grow EBS disks on AWS and thus you can add more space without LVM fragmentation.

6
  • Could you elaborate on the Gitlab issue you mentioned? Googling it didn't help and I believe the best way to learn is from examples.
    – P.Pal
    Mar 19, 2018 at 9:00
  • @P.Pal I've added link to their postmortem, most interesting part begins at chapter "Broken recovery procedures". But I would recommend reading it all as it is invaluable experience from which everyone can learn something new I believe. The major lesson from their failure should be -- always test recovery procedure, on a regular basis. Mar 19, 2018 at 9:06
  • Separate partitions also help against loss due to file system corruption.
    – Halfgaar
    Mar 19, 2018 at 9:45
  • @Halfgaar this is true but out of scope for this question Mar 19, 2018 at 9:46
  • I would like to nuance the LVM comment a bit. I wouldn't span volume groups over multiple physical disks indeed, but if you're not doing that, LVM can be very powerful. So, avoid it to cluster disks together, but don't avoid it all costs.
    – Halfgaar
    Mar 19, 2018 at 9:49
0

No. In your case partitions will only make life harder in the long run. You just multiply the points of possible future* failure.

The solution to low disk space class of problems is two-fold:

  1. Monitoring (immediate reaction).
  2. Capacity planning (in the long run).

If you have these implemeted and then you'd find you that some very mysterious activity can suddenly spike the size of your backups, you would have a (shaky) argument to introduce partitioning.

[*] Your first duty as a sysadmin (which you in fact are) is to be much more pessimistic than your users.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.