pub struct ProbeArgs {
pub json: bool,
pub skip_mount: bool,
pub data_path: Option<PathBuf>,
pub libtorch_path: Option<PathBuf>,
pub docker: Option<String>,
}Expand description
Cluster readiness probe.
Default (single-host): probes the local box for GPU + libtorch arch
match + shared-data path + NCCL availability. Cluster context
(fdl @cluster probe / FDL_ENV=cluster): probes every host in
fdl.cluster.yml via SSH and aggregates the report.
Exit code: 0 when every checked component is green; 1 when any
issue was surfaced. fdl deploy and CI consume the --json shape.
Fields§
§json: boolEmit machine-readable JSON.
skip_mount: boolSkip the shared-data-mount visibility check. Useful for single-host setups without a shared filesystem configured.
data_path: Option<PathBuf>Override the shared-data path (default: cluster.yml’s
per-host data_path:, or the convention default
/flodl/data when unset).
libtorch_path: Option<PathBuf>Override the libtorch directory. Default walks up from cwd
for libtorch/.active. Use this when the libtorch install
lives outside the project tree (e.g. a separate virtiofs
share mounted at a known path on a worker node).
docker: Option<String>Treat NCCL as provided by a Docker image (compose service name
from fdl.yml, e.g. cuda). Suppresses host-level NCCL
discovery and reports “via Docker image <svc>” instead. In
cluster mode, this is auto-derived from each host’s docker:
field in fdl.cluster.yml.
Trait Implementations§
Source§impl FdlArgsTrait for ProbeArgs
impl FdlArgsTrait for ProbeArgs
Source§fn try_parse_from(args: &[String]) -> Result<Self, String>
fn try_parse_from(args: &[String]) -> Result<Self, String>
Source§fn render_help() -> String
fn render_help() -> String
--help to a string.Source§fn render_help_path(argv: &[String]) -> String
fn render_help_path(argv: &[String]) -> String
--help for a specific argv path (program name first, then
the tokens typed). The default ignores the path and returns
Self::render_help — correct for a single struct, whose help is
context-free. Read more