Allow to properly shutdown the PIP service

[otbutz]

I tested the following using the staging branch:

npm start &
PIP_PID=$!
sleep 10
kill -TERM $PIP_PID

The result is that the npm process exits but node and the other PIP processes(empire, country, etc) keep running. (see npm/npm#4603)

This also affects the docker image which is started with npm start aswell.

My current workaround for this behaviour is to invoke node directly:

node --max_old_space_size=4096 index.js &
PIP_PID=$!
sleep 10
kill -TERM $PIP_PID

A separate stop script might be a cleaner solution though.

[orangejulius]

Hi @otbutz,
We are all too familiar with npm/npm#4603, which is a pretty disappointing decision from NPM. I just opened #63, which should fix this. Can you test it out?

[otbutz]

Had a look at your pull request. It should work for the Docker image as the node process will receive all incoming signals but it won't fix the npm start usecase.

That's why i'd suggest an approach like this one: https://stackoverflow.com/a/39128820

The stop endpoint should only work for local connections.

Edit: killing by process title should work too but requires additional commandline tools

[orangejulius]

Ugh. That's clever, but a lot of extra complexity to add to all our services. We are not going to add that. I'd suggest you avoid npm start when backgrounding processes then. I now recall that we also did this in Mapzen Search production.

[otbutz]

The ability to create a pid file would also solve the problem and should be pretty straightforward to implement.

https://www.npmjs.com/package/npid

[orangejulius]

There are a lot of problesm with PID files. I've worked with them at scale for many years and a lot can go wrong.

As far as I know, the only really solid ways to run a service on a linux machine are to use a systemd service, which is out of scope for these Pelias repos, or to use a Docker container, which can always terminate the right processes when it shuts down.

We'll add start scripts to circumvent npm start in all our repos over time (feel free to submit PRs for ones you want right away). I'd suggest you use those for this case. We'll ensure the correct start commands are in those scripts going forward, so you should be able to count on them.

[orangejulius]

I closed this issue because is merged #63, but if there are other solutions I'm missing, let me know :)