Skip to main content

Limits

What OpenConsole refuses or cuts, so a script or an agent can handle it. Every refusal below is audited with outcome rate_limited (or denied / failed).

Requests​

LimitApplies toOver it
10 requests a second, bursts of 30each API key, every call429 rate_limited with Retry-After: 1; wait a second. Audited as api_key.use / rate_limited, at most once a minute per key.
5 at once, then one every 6 secondsDeployment restart and scale, StatefulSet scale, YAML apply (not the dry run); per user, keys and web together429 rate_limited with Retry-After: 6.
10 at once, then one a secondLog reads and log streams; per user429 rate_limited (a stream gets an error frame with retry: true).
MAX_EXEC_SESSIONS_PER_USER open shells (default 3)Shells and one-command exec; per user, keys and web together429 too_many_sessions; a shell from a browser is refused with too many open shells and close code 4429.
MAX_REPLICAS (default 100)Scale400 invalid_input, detail replicas must be between 0 and 100.

The status is followed by the body's error code; the body is described in Errors. Both variables are on Environment variables.

Reading a log: GET …/pods/{name}/logs​

ParameterDefaultLimit
tailLines5001 to 10 000; outside → 400 invalid_input, detail tailLines must be between 1 and 10000
sinceSecondsnonea positive number; otherwise 400 invalid_input
previous, timestamps, containerfalse, false, the main container

The answer is plain text of at most 5 MiB. A longer log is cut at a line end and the response carries X-OpenConsole-Truncated: true: ask for fewer lines or a shorter sinceSeconds. Errors: 404 pod_not_found; 400 container_not_found, not_ready (the container hasn't started) or previous_unavailable; 403 cluster_forbidden when OpenConsole's own account on the cluster may not read the pod or its log (its ClusterRole).

Running a command: POST …/pods/{name}/exec​

commandAn argv array, not run in a shell: ["sh", "-c", "ps aux | head"] for pipes. No stdin.
timeoutSeconds1 to 120, default 30.
Outputstdout and stderr, 1 MiB each; past that the rest is dropped and truncated is true, while the command runs to its end.
TimeoutWhat was printed so far, exitCode: -1, truncated: true. Kubernetes doesn't stop the process; it may go on running in the container.
Exit codeA non-zero exit code is a 200 with that exitCode.
TextOutput is text: bytes that aren't UTF-8 become �; pipe through base64 for binary.
Errors404 pod_not_found; 400 container_not_found or not_ready (the container isn't running); 502 exec_failed when the exec itself fails.

A command run with a key is always recorded; the audit event keeps the program (command[0]), the exit code, bytes, duration and the recording id, never the arguments. The recording reads like a shell's: its first line is $ <command>, then the output. Both endpoints on this page (log and command) work with a web session too, not only with a key.

WebSockets​

Live logs, shells and the event stream are WebSockets. A key is never accepted in a URL, so:

  1. POST /api/ws-ticket with the key and {"path": "/ws/namespaces/<ns>/pods/<pod>/logs"} (the socket's path); the answer is {"ticket": "…"}.
  2. Open the socket at that path with ?ticket=<ticket> within 30 seconds. A ticket works once, for that path only. ?token= is not accepted.

The shell socket (/ws/namespaces/<ns>/pods/<pod>/exec) also takes container and command next to ticket, for example command=/bin/sh; without command it tries bash, then ash, then sh.

curl -s -H "Authorization: Bearer $OPENCONSOLE_API_KEY" -H "Content-Type: application/json" \
-d '{"path": "/ws/namespaces/crashlab/pods/healthy-66fb4fb769-4cjqs/logs"}' "$OC_URL/api/ws-ticket"
{"ticket":"89240df50d12c7a283c0607e0b38a0c95dff80d1d482714b"}

For an agent the plain HTTP endpoints above are simpler: a log snapshot and one command need no socket.

A shell closes after EXEC_IDLE_TIMEOUT without input (default 5 minutes).