0

We have a server with 8 core CPU and 32 gigs RAM. About 1200 active users. We use GitLab rake task for the backups..

sudo gitlab-rake gitlab:backup:create

This process takes around 80 minutes and GitLab is unsuable during this time intermittently as the CPU and RAM are completely full. See the below image.

Load average at 40 and RAM consumption 100 percent.

enter image description here

The CPU/RAM is consumed fully when the final tar is getting created. The size of our backups is going to increase each passing day. Is there a way to optimize the performance of this job? The final tar size is 17496780800 bytes (17.5GB).

IOSTAT command output when job is running:

Linux 3.10.0-957.10.1.el7.x86_64 (abcdefghi)  02/07/19        _x86_64_        (8 CPU)

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           5.01    0.00    2.05    3.14    0.00   89.80
Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
fd0               0.00     0.00    0.00    0.00     0.00     0.00     8.00     0.00   42.20   42.20    0.00  42.20   0.00
sda               0.04     0.27    1.88   26.14     0.10     0.32    30.77     0.24    8.41    4.42    8.70   0.14   0.38
sdc               0.00     0.36    0.28    0.56     0.00     0.03    79.07     0.00    0.72    0.83    0.67   0.43   0.04
sdd               0.01     0.40    2.51    1.19     0.08     0.11   107.56     0.01    2.23    2.85    0.90   0.63   0.23
sdb               0.01     0.49    0.01    0.14     0.00     0.00    38.10     0.00    0.54    0.27    0.57   0.39   0.01
dm-0              0.00     0.00    0.44    0.02     0.03     0.00   122.65     0.00    2.26    2.34    0.45   0.81   0.04
dm-1              0.00     0.00    0.03    0.63     0.00     0.00     8.68     0.00    1.63    0.53    1.68   0.09   0.01
dm-2              0.00     0.00    0.01    0.41     0.00     0.02    85.97     0.00    1.13    0.64    1.15   0.42   0.02
dm-3              0.00     0.00    1.18   24.94     0.07     0.29    28.24     0.23    8.98    6.22    9.11   0.12   0.32
dm-4              0.00     0.00    0.25    0.65     0.01     0.00    26.20     0.00    1.28    1.47    1.21   0.34   0.03
dm-5              0.00     0.00    0.02    0.00     0.00     0.00    34.01     0.00    2.21    1.17  233.00   1.40   0.00
dm-6              0.00     0.00    0.02    0.37     0.00     0.00    15.53     0.00    1.97    4.37    1.87   0.32   0.01

SWAP WHEN THE JOB IS RUNNING

enter image description here

3
  • Run this job when the server is not in use. You can;t do much if you have so huge amount of data and (probably) a lot of small files – Romeo Ninov Jul 2 '19 at 6:10
  • What is your storage backend? Does your GitLab have enough resources? Does it go on swap while creating the tar? Perhaps you should consider backing up to a network location and/or SSD or other device. – Miuku Jul 2 '19 at 7:11
  • Storage is the File System on the same server. A bit of swap is used while creating the tar (Between 0 to 100MB). We have 4 GB swap. I have added other details regarding resources etc. to the question. – Koshur Jul 2 '19 at 8:35
0

It's not clear if you backend storage is SATA or SSD or what. I would recommend you migrate to SSD. We had similar issues when running on SATA. Once we migrated to SSD, the issues vanished.

0

This is likely happening because all of the normal Gitlab operations are reading & writing to the same disk as the Backup operations. Is it possible to store the backup image on another disk or to an NFS-mounted filesystem? That will at least move the I/O onto a different subsystem.

It might be worth trying the STRATEGY=copy backup strategy, which will copy the data to another location first before running the tar/gzip-- gzip in particular can be resource intensive. Since the copy should be fairly fast, the Gitlab server shouldn't be impacted for very long.

Look to see if you can apply flags to Gzip. Try reducing the Gzip compression level-- this will result in a moderately larger backup file, but will reduce system load-- sometimes significantly.

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.