Logs and shell access
Reading a workload's logs, getting a shell inside it, and what the platform deliberately will not let you do.
When a workload is not doing what you expect, there are three ways in, and they are all recorded in your project’s audit log.
Logs
Every workload page has a Logs view. It streams from the running pods — if a workload has several replicas you are reading them together, so a log line that appears only intermittently is usually one replica misbehaving rather than a race.
Logs are what the container writes to stdout and stderr. Nothing is collected from inside the filesystem, so an application that writes its log to a file will look silent here. If you control the image, log to stdout.
Shell, from the console
The workload page also has a Terminal. It opens an interactive shell in the running container, in the browser, with no setup.
This is the quickest way to answer “is the config file actually there” or “can this container reach that host”. It is a shell in a running container: anything you change is gone when the pod restarts, which for a workload that is crash-looping may be very soon.
Shell, over SSH
The same access is available from your own terminal, which is better for anything longer than a few commands.
1. Add your public key. Profile → SSH keys, and paste the contents of
~/.ssh/id_ed25519.pub. OpenSSH-format ssh-ed25519, ssh-rsa and
ecdsa-sha2-nistp{256,384,521} keys are accepted. The console shows the
fingerprint it will match against — check it against your own copy:
ssh-keygen -lf ~/.ssh/id_ed25519.pub
2. Connect, addressing the workload as project-slug/workload-name. The
hostname and port come from your operator:
ssh -p 2222 my-project/web@gateway.example.com
The project slug is in the project’s URL; the workload name is the one in its
row. If your SSH client refuses a / in the username, : works identically —
my-project:web@….
3. Check the host key on first connect. Your operator publishes the
gateway’s fingerprint; verify it matches before answering yes. This is the
one step worth not skipping — it is what stops anyone between you and the
gateway impersonating it.
What SSH here will not do
It is a shell, not a tunnel. Port forwarding (-L and -R), agent
forwarding and X11 are refused, deliberately. If you were hoping to forward
a database port to your laptop, that is not available today.
The username always has to resolve to a workload you have access to. There is no shell on the cluster itself, and no way to reach a workload in a project you are not a member of.
It is all audited
Console terminal and SSH are the same operation over different transports, and both land in the project’s audit log with who connected and to which workload. That is deliberate: shell access to a running workload is the most powerful thing most users can do, and it should be as visible as deleting something.