# Problem with Upgrade \[error 137\]

**URL:** <https://meta.discourse.org/t/problem-with-upgrade-error-137/293652>\
**Category:** Self-hosting\
**Tags:** server-resources\
**Created:** [January 31, 2024, 4:33pm UTC](https://meta.discourse.org/t/problem-with-upgrade-error-137/293652 "2024-01-31T16:33:52Z")\
**Posts on this page:** 1\
**Showing post:** 17

<div class="post-metadata">

**Author:** ![merefield](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/merefield/32/176214_2.png) [@merefield](https://meta.discourse.org/u/merefield)\
**Post date:** [February 1, 2024, 8:17am UTC](https://meta.discourse.org/t/problem-with-upgrade-error-137/293652/17 "2024-02-01T08:17:19Z")

</div>

Because running the website (albeit with fewer Unicorns) + performing a build at the same time is costly in memory?

I tried this out on my server (3 VCPU, 4GB):

### Online:

2.21 before upgrade → 1.7 as it frees up memory → 3.5 peak during build (GBs)

Swap also went up ~200MB to peak ~800MB

### Command line (not apples to apples as just after the above)

1.7 before upgrade → drops to ~250MB! → climbs and climbs to peak 3.25 during assets compilation, but generally much lower.

Swap usage was very low during almost the entire process.

Caveat: all figures observed in `htop` so may be inaccurate, probably best to graph it.

I was surprised how “peaky” offline rebuilds were. Assets precompile definitely uses a lot of memory - perhaps more the more VCPUs you have as it might do some of that in parallel?

I was also suprised how little difference the peak was in my case, though swap usage was significantly lower and generally, memory use was _much_ lower for an offline rebuild, remaining in the 100’s of MB during 95% of the process, whereas an online upgrade was minimum 1.7GB.

---

_[View the full topic](https://meta.discourse.org/t/problem-with-upgrade-error-137/293652)._