Check for XRAY_PATCH_AWS_SDK env var, and do not patch aws_sdk if unset

[barrucadu]

I'll open an issue on aws-xray-sdk if other people agree with my conclusion.

---

Since applying the last round of govuk_app_config upgrades, asset-manager and publishing-api have been using far more memory:

The memory shoots up as soon as the app is deployed, so I believe this is an overhead imposed at initialisation-time, rather than some sort of leak which manifests when requests are made.

After doing some debugging (commenting out lines of code and crossing my fingers), I narrowed this down to having X-Ray patch the aws_sdk gem. If that gem is patched, the problem occurs; if it is not, even if net_http is patched, the memory usage seems much closer to before.

It's a shame to lose tracking for stuff which goes through aws_sdk, but that's only a small amount of stuff (asset uploads from the asset-manger, and nothing(!) from the publishing-api), and I think we can assume that AWS is not the bottleneck in our architecture.

---

Trello card

[barrucadu]

Here are some more graphs. The spikes are when the app was deployed.

asset-manager continues to look good, memory usage has been creeping up but it's not yet at the point it was before X-Ray, and it's quite slow:

publishing-api was good, then a branch with X-Ray instrumenting aws_sdk was deployed. Memory usage is below what it was, but is still rather higher than it was without X-Ray instrumenting aws_sdk:

[bevanloon]

Sounds like you've got a cause and effect here.