Issues with using docker with FTL introduced in v0.215.0

[alec-brooks]

It seems that v0.215.0 has introduced a bug where it is no longer possible to run commands such as ftl serve locally.

I am attempting to run these commands in the ftl repo using the ftl script located at scripts/ftl.

Below is some of my terminal output that demonstrates this bug having been introduced in v0.215.0.

ftl🐚 ftl ((v0.214.0)) $ which ftl             
/Users/abrooks/Development/ftl/scripts/ftl
ftl🐚 ftl ((v0.214.0)) $ ftl serve
trace: Loading config from /Users/abrooks/Development/ftl/ftl-project.toml
trace: /xyz.block.ftl.v1.ControllerService/Ping (unary)
trace:localhost: Trying authenticator
trace:localhost: No authenticator found for http://localhost:8892/xyz.block.ftl.v1.ControllerService/Ping in map[ftl.tbddev.org:ftl-authenticator-tbddev]
debug: Unary RPC failed: unavailable: dial tcp [::1]:8892: connect: connection refused: /xyz.block.ftl.v1.ControllerService/Ping
info: Starting FTL with 1 controller(s)
trace: exec: cd . && docker ps -a --filter name=^/ftl-db-1\$ --format \{\{.Names}}
trace: exec: cd . && docker start ftl-db-1
trace: exec: cd . && docker port ftl-db-1 5432/tcp
debug: Reusing existing docker container "ftl-db-1" on port "54320" for postgres db
debug: Waiting for ftl-db-1 to be healthy
trace: exec: cd . && docker inspect --format \{\{.State.Health.Status}} ftl-db-1
debug: Postgres DSN: postgres://postgres:secret@localhost:54320/ftl?sslmode=disable
^Cdebug: FTL terminating with signal interrupt
ftl🐚 ftl ((v0.214.0)) $ git checkout tags/v0.215.0
Previous HEAD position was c3c7077f feat: use secrets instead of env vars for DSNs (#1483)
HEAD is now at 9860ee37 feat: use docker api instead of exec (fixes #1169) (#1488)
ftl🐚 ftl ((v0.215.0)) $ ftl serve
trace: Loading config from /Users/abrooks/Development/ftl/ftl-project.toml
trace: /xyz.block.ftl.v1.ControllerService/Ping (unary)
trace:localhost: Trying authenticator
trace:localhost: No authenticator found for http://localhost:8892/xyz.block.ftl.v1.ControllerService/Ping in map[ftl.tbddev.org:ftl-authenticator-tbddev]
debug: Unary RPC failed: unavailable: dial tcp [::1]:8892: connect: connection refused: /xyz.block.ftl.v1.ControllerService/Ping
info: Starting FTL with 1 controller(s)
ftl: error: failed to list containers: Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?
ftl🐚 ftl ((v0.215.0)) $ docker ps
CONTAINER ID   IMAGE             COMMAND                  CREATED       STATUS                  PORTS                     NAMES
b4ba4dff08de   postgres:latest   "docker-entrypoint.s…"   7 weeks ago   Up 20 hours (healthy)   0.0.0.0:54320->5432/tcp   ftl-db-1

It is worth noting that I am running on an M1 MBP running macOS 14.4.1. Using Docker Desktop for macOS.
Docker desktop stats below:
Version: 4.29.0 (145265)
Engine: 26.0.0
Compose: v2.26.1-desktop.1
Credential Helper: v0.8.1
Kubernetes: v1.29.2

[alecthomas]

@gak not urgent, as Alec has a workaround for now, but we should try and figure out why client.FromEnv isn't working in this scenario, and I guess why the docker CLI is working, and try and emulate whatever it's doing.

[gak]

The fix was a setting in Docker Desktop allowing the socket to be available:

Settings -> Advanced (it was not obvious for me--had to scroll the left pane down) -> "Allow the default Docker socket to be used (requires password)" -> Needs to be enabled