Issue with elixir-phoenix-on-google-compute-engine - google-cloud-platform

Question

I’m trying to deploy to GCP Compute Engine by following this tutorial
https://cloud.google.com/community/tutorials/elixir-phoenix-on-google-compute-engine
Unable to connect to provided external IP after create firewall-rules
There are no errors in following the tutorial. But cannot connect to http://${external_ip}:8080 after creating firewall rules
Build release is already in Google Cloud Storage
I copy hello
gsutil cp _build/prod/rel/hello/bin/hello\
gs://${BUCKET_NAME}/hello-release
instead of hello.run
gsutil cp _build/prod/rel/hello/bin/hello.run \
gs://${BUCKET_NAME}/hello-release
My instance-startup.sh
#!/bin/sh
set -ex
export HOME=/app
mkdir -p ${HOME}
cd ${HOME}
RELEASE_URL=$(curl \
-s "http://metadata.google.internal/computeMetadata/v1/instance/attributes/release-url" \
-H "Metadata-Flavor: Google")
gsutil cp ${RELEASE_URL} hello-release
chmod 755 hello-release
wget https://dl.google.com/cloudsql/cloud_sql_proxy.linux.amd64 \
-O cloud_sql_proxy
chmod +x cloud_sql_proxy
mkdir /tmp/cloudsql
PROJECT_ID=$(curl \
-s "http://metadata.google.internal/computeMetadata/v1/project/project-id" \
-H "Metadata-Flavor: Google")
./cloud_sql_proxy -projects=${PROJECT_ID} -dir=/tmp/cloudsql &
PORT=8080 ./hello-release start
gcloud compute instances get-serial-port-output shows
...
Feb 23 18:02:35 hello-instance startup-script: INFO startup-script: + PORT=8080 ./hello-release start
Feb 23 18:02:35 hello-instance startup-script: INFO startup-script: + ./cloud_sql_proxy -projects= hello -dir=/tmp/cloudsql
Feb 23 18:02:35 hello-instance startup-script: INFO startup-script: 2019/02/23 18:02:35 Rlimits for file descriptors set to {&{8500 8500}}
Feb 23 18:02:35 hello-instance startup-script: INFO startup-script: ./hello-release: 31: exec: /app/hello_rc_exec.sh: not found
Feb 23 18:02:39 hello-instance startup-script: INFO startup-script: 2019/02/23 18:02:39 Listening on /tmp/cloudsql/hello:asia-east1:hello-db/.s.PGSQL.5432 for hello:asia-east1: hello-db
Feb 23 18:02:39 hello-instance startup-script: INFO startup-script: 2019/02/23 18:02:39 Ready for new connections
Feb 23 18:08:08 hello-instance ntpd[656]: kernel reports TIME_ERROR: 0x41: Clock Unsynchronized
hello_rc_exec.sh is generated after initialize Distillery. It is stored in _build/prod/rel/hello/bin/hello_rc_exec.sh
firewall rules
NAME NETWORK DIRECTION PRIORITY ALLOW DENY DISABLED
default-allow-http-8080 default INGRESS 1000 tcp:8080 False
...
I also run in ps aux | grep erl in the instance
hello_team#hello-instance:~$ ps aux | grep erl
hello_t+ 23166 0.0 0.0 12784 1032 pts/0 S+ 08:04 0:00 grep erl
Im not sure what information is needed to fix this
Please do ask for information and I will provide them.
Thank you

Answer 1

For posterity, here was the solution (worked out in this forum thread).
First, the poster had uploaded the hello file instead of hello.run to cloud storage. The tutorial intentionally specifies uploading hello.run because it is a full executable archive of the entire release, whereas hello is merely a wrapper script and is by itself not capable of executing the app. So this modification to the procedure needed to be reverted.
Second, the poster's app included the elixir_bcrypt library. This library includes a NIF whose platform-specific binary code is built in the deps directory (instead of the _build directory). The tutorial's procedure doesn't properly clean out binaries in deps prior to cross-compiling for deployment, and so the poster's macOS-built bcrypt library was leaking into the build. When deployed to compute engine on Debian, this crashed on initialization. The poster fixed this problem by deleting the deps directory and re-installing dependencies while cross-compiling.
It was also noted during the discussion that the tutorial promoted a poor practice of mounting the user's app in a volume when doing a Docker cross-compilation. Instead, it should simply copy the app into the image, perform the entire build there, and use docker cp to extract the built artifact. This practice would have prevented this issue. A work item was filed to modify the tutorial accordingly.

Answer 2

The solution is here.
Thank you for the help everyone!