Trying to deploy GitLab-CE into AWS. Have selected m7i.8xlarge as my instance-type. This instance-type is provisional – mostly to support speedier migration from on-prem to AWS. Will likely dial-back to something less expensive and more right-sized once we get metrics from the migrated service.
I am deploying using a (cloud-init) userData payload that I’ve generalized so it supports EL 8 & 9 distros and Amazon Linux 2023. Basically, the userData encapsulates the installation-guidance for RHEL and Amazon Linux 2023.
Installation runs like a champ on RHEL, Rocky and OEL. Primary annoyance being that the gitlab-ce RPM’s %post script takes nearly 15 minutes to run (used pstree to see that it’s bogging on the gitlab-ctl upgrade step).
Same automation becomes unusably-slow on AL2023. Just like on ELx-based deployments, the slowdown is in the %post script’s gitlab-ctl upgrade step. However, instead of taking fifteen-ish minutes, it’s taking over sixty minutes. Seeing similar slowness when using gitlab-ctl reconfigure as part of a migration-process to move from an on-prem installation that’s using local storage for BLOBs.
Only reason I’m bothering with AL2023 is that the observed transfer-speeds between ELx-based EC2s is less than 60% the throughput observed between AL2023-based EC2s. Haven’t sorted out why EL9’s network-performance (S3) is so gawdawful, but that’s (likely) not an issue for this forum.
Hoping someone’s run into this, before, and there’s an easily-resolved gotcha that’s not listed in the Amazon Linux 2023 installation-instructions that will get me past this.
Host details:
- GitLab CE: 17.<latest>
- Host OS: varies – listed above
- Hosting Environment: AWS
- CPU: 32 (
/proc/cpuinfoshows: Intel(R) Xeon(R) Platinum 8488C @ 2.4GHz) - RAM: 128GiB
- Disk: 500GiB (
gp3@ 1000 IOPS and 500GiB/s throughput)