Publish step of CI is busted

[tsibley]

Failing since March 2021, with the same failure in May and just now.

Pretty cryptic. It doesn't seem to get past the initial setting of distribution=$(…) in ./publish since otherwise I'd expect to see more output:

conda/publish

Lines 6 to 20 in e525970

	echo "-> Finding CloudFront distribution"
	distribution=$(
	aws cloudfront list-distributions \
	--query 'DistributionList.Items[?contains(Aliases.Items, `data.nextstrain.org`)] | [0].Id' \
	--output text
	)
	if [[ -z $distribution || $distribution == None ]]; then
	echo "Unable to find CloudFront distribution id for data.nextstrain.org" >&2
	echo
	echo "Are your AWS CLI credentials for the right account?" >&2
	exit 1
	fi
	echo "-> Uploading to S3"

The sole message output is clearly a Python repr() string, but without any other context. I thought it might be an AWS CLI v1 vs. v2 thing, but in local testing with v2 the commands seem to work just fine. Might still be related, though.

Exit status 255 from Bash likely indicates that something exited out of range (<0 or ≥256). Typically I've seen this with programs that exit(-1).

I've manually run ./publish locally for the latest changes for now.

[jameshadfield]

Possibly unrelated, but for GitHub actions you need (or needed) to have AWS_DEFAULT_REGION: us-east-1. I found this in an obscure GitHub thread while debugging nextstrain/ncov#671. I have a feeling the nextstrain cli defaults to this without it needing to be explicitly set, but here the publish script uses aws commands.

[tsibley]

Hmm. An explicit default AWS region certainly wasn't always required since the automation worked for a long time. But it's possible the OS image upgraded to a newer version of the AWS CLI at some point which requires it. Can check if we continue to debug this.