Time out deploy commands after 10 minutes

[slominskir-coding-agent]

Fixes #1.

What changed

• Timeout: SSHFacade waits at most 10 minutes (commandTimeout) for the deploy command. After that it closes the channel and the job records a SocketTimeoutException as its stack trace, with no exit code, and the stdout and stderr received so far.
• No transaction while the command runs: asyncExecuteRemoteCommand is now @TransactionAttribute(NOT_SUPPORTED). Before, it ran in a container transaction for the whole command. The jeffersonlab/wildfly:3.0.1 image keeps Wildfly's default transaction timeout of 300 seconds, so a command running longer than that would leave the transaction rolled back, and saving the job at the end would fail. deployJobFacade.edit(job) now saves the job in its own transaction.

The 10 minutes is a hard-coded constant, like the other SSH timeouts. If some deploys need longer, it could become an environment variable or a column on APP_ENV.

No deployment steps.

What happens on the remote host

The sshd library has no client API to signal the remote process, so the timeout only closes the channel:

• A command waiting for input, like the certificate prompt in the issue, gets end-of-input and exits.
• A command that is busy, like sleep, keeps running on the remote host until it ends on its own.

Checks

• gradle spotlessApply build passes (JDK 21, in the gradle:9-jdk21 container).
• I ran the real executeRemoteCommand (through reflection, with commandTimeout set to 3 seconds) against the project's Dockerfile-sshd container:

deploy_command (version 1.0.0 appended)	Time	Exit code	Stdout / stderr	Stack trace
touch /tmp/hello && echo	1.2 s	0	1.0.0	none
sh -c 'echo out; echo err >&2; exit 3' #	0.3 s	3	out / err	none
sh -c 'echo partial; sleep 30; echo done' #	3.3 s	none	partial	SocketTimeoutException
sh -c 'echo asking; read -r x; echo got:$x' #	3.2 s	none	asking	SocketTimeoutException

After the run, the sleep command was still running in the sshd container; the read command had exited.

Not checked: the transaction change and the /log page in the full Compose stack. Port 8081 on the VM is taken by another project's Keycloak.

🤖 Generated with Claude Code

[slominskir]

Some C++ app deploys build from source, and that can take a very long time, especially if Clang Tidy static analysis is repeated (generally done in CI). We'll likely need to revisit this hard-coded 10 minute timeout, but good enough for now.