Troubleshooting

Troubleshooting

Also see Known Issues

If your app isn’t working and you’re seeing either 504s or an “unhealthy” message, you’re in the right place. The first places to check for clues are gigalixir logs and gigalixir ps. For example, check if the lastState shows OOMKilled, which indicates your app is using more memory than it has provisioned. If nothing pops out at you there, keep reading.

A 504 means that our load balancer isn’t able to reach your app. This is usually because the app isn’t running. An app that isn’t running is usually failing health checks and we constantly restart apps that fail health checks in hopes that it will become healthy.

If you’ve just deployed, and you’re not seeing 504s, but you’re still seeing the old version of your app instead of the new version, it’s the same problem. This happens when the new version does not pass health checks. When the new version doesn’t pass health checks, we don’t route traffic to it and we don’t terminate the old version.

Our health checks simply check that your app is listening on port $PORT. If you’re running a non-HTTP Elixir app, but need to just get health checks to pass, take a look at https://github.com/jesseshieh/elixir-tcp-accept-and-close

If you’re using Mix, see troubleshooting mix.

If you’re using Distillery, see troubleshooting distillery.

If you’re using Elixir Releases, see troubleshooting Elixir releases.

Mix

Let’s verify that your app works locally.

Run the following commands

mix deps.get
mix compile
SECRET_KEY_BASE="$(mix phx.gen.secret)" MIX_ENV=prod DATABASE_URL="postgresql://user:pass@localhost:5432/foo" PORT=4000 mix phx.server
curl localhost:4000

If it doesn’t work, the first thing to check is your prod.exs file. Often, it is missing an http configuration or there is a typo in the FooWeb.Endpoint module name.

If everything works locally, you might be running a different version of Elixir in production. See How do I specify my Elixir, Erlang, Node, NPM, etc versions?.

Another possibility is that your app is running out of memory and can’t start up properly. To fix this, try scaling up. See How to Scale an App.

If the above commands still do not succeed and your app is open source, then please contact us for help. If not open source, contact us anyway and we’ll do our best to help you.

Distillery

If you’re having trouble getting things working, you can verify a few things locally.

First, try generating and running a Distillery release locally by running

mix deps.get
mix compile
export SECRET_KEY_BASE="$(mix phx.gen.secret)"
export DATABASE_URL="postgresql://user:pass@localhost:5432/foo"
MIX_ENV=prod mix distillery.release --env=prod
# if you are a running distillery below 2.1, then run this instead: MIX_ENV=prod mix release --env=prod
APP_NAME=gigalixir_getting_started
MY_HOSTNAME=example.com MY_COOKIE=secret REPLACE_OS_VARS=true MY_NODE_NAME=foo@127.0.0.1 PORT=4000 _build/prod/rel/$APP_NAME/bin/$APP_NAME foreground
curl localhost:4000

Don’t forget to replace gigalixir_getting_started with your own app name. Also, change/add the environment variables as needed.

You can safely ignore Kubernetes errors like [libcluster:k8s_example] errors because you probably aren’t running inside Kubernetes.

If they don’t work, the first place to check is prod.exs. Make sure you have server: true somewhere and there are no typos.

In case static assets don’t show up, you can try the following and then re-run the commands above.

cd assets
npm install
npm run deploy
cd ..
mix phx.digest

If your problem is with one of the buildpacks, try running the full build using Docker and Herokuish by running

APP_ROOT=$(pwd)
rm -rf /tmp/gigalixir/cache
rm -rf _build
mkdir -p /tmp/gigalixir/cache
docker run -it --rm -v $APP_ROOT:/tmp/app -v /tmp/gigalixir/cache/:/tmp/cache us.gcr.io/gigalixir-152404/herokuish

Or to inspect closer, run

docker run -it --rm -v $APP_ROOT:/tmp/app -v /tmp/gigalixir/cache/:/tmp/cache --entrypoint=/bin/bash us.gcr.io/gigalixir-152404/herokuish

# and then inside the container run
build-slug

# inspect /app folder
# check /tmp/cache

If everything works locally, you might be running a different version of Elixir in production. See How do I specify my Elixir, Erlang, Node, NPM, etc versions?.

Another possibility is that your app is running out of memory and can’t start up properly. To fix this, try scaling up. See How to Scale an App.

If the above commands still do not succeed and your app is open source, then please contact us for help. If not open source, contact us anyway and we’ll do our best to help you.

Elixir Releases

If you’re having trouble getting things working, you can verify a few things locally.

First, try generating and running a release locally by running

mix deps.get
export SECRET_KEY_BASE="$(mix phx.gen.secret)"
export DATABASE_URL="postgresql://user:pass@localhost:5432/foo"
MIX_ENV=prod mix release
APP_NAME=gigalixir_getting_started
PORT=4000 _build/prod/rel/$APP_NAME/bin/$APP_NAME start
curl localhost:4000

Don’t forget to replace gigalixir_getting_started with your own app name. Also, change/add the environment variables as needed.

You can safely ignore Kubernetes errors like [libcluster:k8s_example] errors because you probably aren’t running inside Kubernetes.

If they don’t work, the first place to check is prod.exs. Make sure you have server: true somewhere and there are no typos.

In case static assets don’t show up, you can try the following and then re-run the commands above.

cd assets
npm install
npm run deploy
cd ..
mix phx.digest

If your problem is with one of the buildpacks, try running the full build using Docker and Herokuish by running

APP_ROOT=$(pwd)
rm -rf /tmp/gigalixir/cache
rm -rf _build
mkdir -p /tmp/gigalixir/cache
docker run -it --rm -v $APP_ROOT:/tmp/app -v /tmp/gigalixir/cache/:/tmp/cache us.gcr.io/gigalixir-152404/herokuish

Or to inspect closer, run

docker run -it --rm -v $APP_ROOT:/tmp/app -v /tmp/gigalixir/cache/:/tmp/cache --entrypoint=/bin/bash us.gcr.io/gigalixir-152404/herokuish

# and then inside the container run
build-slug

# inspect /app folder
# check /tmp/cache

If everything works locally, you might be running a different version of Elixir in production. See How do I specify my Elixir, Erlang, Node, NPM, etc versions?.

Another possibility is that your app is running out of memory and can’t start up properly. To fix this, try scaling up. See How to Scale an App.

If the above commands still do not succeed and your app is open source, then please contact us for help. If not open source, contact us anyway and we’ll do our best to help you.

Known Issues

  • git push hangs and then times out

    • Try running git config --local http.version HTTP/1.1. We’ve seen this issue happen with many customers and we’ve been able to narrow it down to an HTTP/2 issue of some kind with some versions of curl or git, but haven’t been able to reproduce it. Many customers report that switching to HTTP/1.1 seems to fix the issue. For more information, try setting GIT_TRACE=1 GIT_CURL_VERBOSE=1 when pushing. If you can also send us the output, that would be helpful. Often what we’ll see in the output is something like 17 bytes stray data read before trying h2 connection.

  • (FunctionClauseError) no function clause matching in List.first/1 when running gigalixir ps:migrate

    • If you have a releases config in your mix.exs, make sure it is named the same as your app a few lines above. This is something we need to figure out how to fix, but in the meantime, we need the release name to match the app name. Let us know if you encounter this issue so we can bump the priority!

  • Warning: Multiple default buildpacks reported the ability to handle this app. The first buildpack in the list below will be used.

    • This warning is safe to ignore. It is a temporary warning due to a workaround.

  • curl: (56) GnuTLS recv error (-110): The TLS connection was non-properly terminated.

    • Currently, the load balancer for domains under gigalixirapp.com has a request timeout of 30 seconds. If your request takes longer than 30 seconds to respond, the load balancer cuts the connection. Often, the cryptic error message you will see when using curl is the above. The load balancer for custom domains does not have this problem.

  • php apps don’t work well with the default stack, gigalixir-18. If you are deploying php, please downgrade your stack to giglaixir-16 with something like gigalixir stack:set -s gigalixir-16. Th e reason is because gigalixir-18 is based on heroku-18 which does not have libreadline.so preinstalled for some reason where giglaixir-16, based on heroku-16, does.

  • Did not find exactly 1 release

    • This can happen for a few different reasons, but usually clearing your build cache or retrying will resolve it. In some cases, if you added release=true to your elixir_buildpack.config file, it caches the release and is never deleted even when you bump the app version in your mix.exs. This results in two release folders and gigalixir does not know which release you intend to deploy and errors out. Clearing the cache resolves this issue. In some cases, if two deploys are running concurrently, you can end up with two release tarballs at the same time. This is a known issue we intend to fix, but usually re-running the deploy will work fine since it is a race condition.

Common Errors

A good first thing to try when you get a git push error is cleaning your build cache.

  • My deploy succeeded, but nothing happened.

    • When git push gigalixir master succeeds, it means your code was compiled and built without any problems, but there can still be problems during runtime. Other platforms will just let your app fail, but gigalixir performs tcp health checks on port 4000 on your new release before terminating the old release. So if your new release is failing health checks, it can appear as if nothing is happening because in a sense, nothing is. Check gigalixir logs for any startup errors.

  • My app takes a long time to startup.

    • Most likely, this is because your CPU reservation isn’t enough and there isn’t any extra CPU available on the machine to give you. Try scaling up your instance sizes. See How to Scale an App.

  • failed to connect: ** (Postgrex.Error) FATAL 53300 (too_many_connections): too many connections for database

    • If you have a free tier database, the number of connections is limited. Try lowering the pool_size in your prod.exs to 2, or if you’re using prod.secret.exs setting the POOL_SIZE environment variable using gigalixir config:set POOL_SIZE=2.

  • ~/.netrc access too permissive: access permissions must restrict access to only the owner

    • run chmod og-rwx ~/.netrc

  • git push gigalixir master asks for my password

    • First try running gigalixir login and try again. If that doesn’t work, try resetting your git remote by running gigalixir git:remote $APP and trying again.

  • remote: cp: cannot overwrite directory ‘/tmp/cache/node_modules/phoenix_html’ with non-directory

  • conn.remote_ip has 127.0.0.1 instead of the real client ip

  • (File.Error) could not read file “foo/bar”: no such file or directory

    • Often, this means that Distillery did not package the foo directory into your release tarball. Try using Distillery Overlays to add the foo directory. For example, adjusting your rel/config.exs to something like this

      release :gigalixir_getting_started do
  set version: current_version(:gigalixir_getting_started)
  set applications: [
    :runtime_tools
  ]
  set overlays: [
    {:copy, "foo", "foo"}
  ]
end

      For more, see https://github.com/bitwalker/distillery/blob/master/docs/Overlays.md

  • cd: /tmp/build/./assets: No such file or directory

    • This means the Phoenix static buildpack could not find your assets folder. Either specify where it is or remove the buildpack. To specify, configure the buildpack following https://github.com/gjaldon/heroku-buildpack-phoenix-static. To remove, create a .buildpacks file with the buildpacks you need. For example, just https://github.com/HashNuke/heroku-buildpack-elixir

  • SMTP/Email Network Failures e.g. {:network_failure, ‘smtp.mailgun.org’, {:error, :timeout}}

  • unknown command: MIX_ENV=prod mix phx.server

    • If you are you are using a custom Procfile with an environment variable at the front of the command, you’ll get this error. Try adding env to the front of the command. See https://github.com/ddollar/foreman/issues/265. We use the most command Ruby Foreman which behaves differently from Heroku’s for this situation.

  • init terminating in do_boot ({cannot get bootfile,no_dot_erlang.boot})

  • Could not invoke task “release”: –env : Unknown option

    • This happens when you upgrade to elixir 1.9, but are still using distillery older than 2.1. Upgrade distillery to fix this issue, but be sure to also change your rel/config.exs file. Mix.Releases.Config needs to be renamed to Distillery.Releases.Config

  • sh: 1: mix: not found

    • If you have an old Phoenix project where a package.json file exists in the project root folder, the herokuish buildpack might mistakenly recognize it as a Node.js project, and thus fail to build it properly. You may need to manually add a .buildpacks file in your root folder, as documented in the “Specify Buildpacks” sections above.

Support/Help

Feel free to email help@gigalixir.com for any questions or issues, we generally respond quickly.