deployment

tidepool is one binary. The web UI is in it. It needs no database, no JVM and no other server. It listens on port 7070.

the container image

There is no published image yet. Build it from a checkout:

docker build -t tidepool .

The image is about 112 MB. It runs as uid 10001. Without arguments, it serves the retail example, in memory, as a demo with ingest and admin open. Publish its port on loopback only:

docker run -p 127.0.0.1:7070:7070 tidepool

Run it with your project and a volume for the data:

docker run -p 7070:7070 \
  -v ./my-project:/project -v tidepool-data:/data \
  -e TIDEPOOL_ADMIN_TOKEN=… -e TIDEPOOL_READ_TOKEN=… -e TIDEPOOL_INGEST_TOKEN=… \
  tidepool serve --config /project --data /data --listen 0.0.0.0:7070

The project directory must be writable by uid 10001. Without that, tidepool cannot write back the changes from the web UI.

On an address other than loopback, tidepool does not start without the ingest and admin tokens. Set the read token too, unless the dashboards are for everyone. See access.

a public web UI and private ingest

The web UI must be where its users are. The ingest endpoints must be only where the producers are. Give each its own listener:

# fragment: tidepool.yaml
listen:
  public: 0.0.0.0:7070
  ingest: 10.0.0.5:7071
  admin: 127.0.0.1:7072

Publish only the public port to the internet. Put the ingest address on the network of kayak and the other producers. Reach admin through SSH or a VPN. To let one producer write from the internet, give it a public credential.

Terminate TLS in front of tidepool, for example in a load balancer or a reverse proxy.

the binary

just build
./target/release/tidepool serve --config my-project

The command line reference lists every command and flag.

memory

tidepool keeps all data in memory. An orders row of the retail example takes about 28 bytes. 100 million orders take about 2.8 GB. Give the machine memory for the data, the caches of the query engine and the dictionaries. /metrics shows each of them.

memory_budget in tidepool.yaml is accepted, but tidepool does not enforce it yet.

the config in git

Mount the project directory from a git checkout. With config_writeback.git_commit: true, tidepool commits each change that it applies. The commit message summarizes the change and names its author: tidepool: … [by …].

To keep the changes out of history, set git_commit: false. tidepool then writes the files and does not commit them.

upgrades

tidepool is pre-1.0. The format of the config and of the data directory can change between minor versions. The release notes say when a data directory must start again empty.