Reject invalid deployment configuration early
This commit is contained in:
parent
2020dd4ed3
commit
0e351c3640
1 changed files with 543 additions and 55 deletions
|
|
@ -61,7 +61,7 @@ impl WorkerApplicationConfig {
|
|||
endpoint: target.endpoint.clone(),
|
||||
region: target.region.clone(),
|
||||
});
|
||||
Ok(Self {
|
||||
let config = Self {
|
||||
management: ManagementConfig::from_env(crate::COMPONENT_NAME)?,
|
||||
database_url: required_env("DATABASE_URL")?,
|
||||
database_connections: env_parse("REMORA_DATABASE_MAX_CONNECTIONS", "16")?,
|
||||
|
|
@ -124,7 +124,9 @@ impl WorkerApplicationConfig {
|
|||
"REMORA_HUGGING_FACE_ENDPOINT",
|
||||
"https://huggingface.co",
|
||||
),
|
||||
})
|
||||
};
|
||||
validate_worker_timing(&config.coordination, &config.worker)?;
|
||||
Ok(config)
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -324,6 +326,27 @@ fn parse_scheme_ports(input: &str) -> Result<Vec<(String, u16)>, RuntimeError> {
|
|||
.collect()
|
||||
}
|
||||
|
||||
fn validate_worker_timing(
|
||||
coordination: &StoreConfig,
|
||||
worker: &WorkerConfig,
|
||||
) -> Result<(), RuntimeError> {
|
||||
if coordination.global_limit == 0
|
||||
|| coordination.lease_ttl.is_zero()
|
||||
|| coordination.reserve_ttl.is_zero()
|
||||
|| coordination.progress_ttl.is_zero()
|
||||
|| worker.local_limit == 0
|
||||
|| worker.idle_delay.is_zero()
|
||||
|| worker.renew_interval.is_zero()
|
||||
|| worker.renew_interval >= coordination.lease_ttl
|
||||
|| worker.cancel_poll_interval.is_zero()
|
||||
{
|
||||
return Err(RuntimeError::message(
|
||||
"worker limits and timing must be positive, and renewal must precede lease expiry",
|
||||
));
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn wait_for_shutdown() -> Result<(), RuntimeError> {
|
||||
#[cfg(unix)]
|
||||
{
|
||||
|
|
@ -387,4 +410,19 @@ mod tests {
|
|||
assert!(parse_scheme_ports("https").is_err());
|
||||
assert!(parse_cidrs("private").is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn validates_worker_timing_before_external_connections() {
|
||||
let coordination = StoreConfig {
|
||||
global_limit: 1,
|
||||
lease_ttl: Duration::from_secs(30),
|
||||
reserve_ttl: Duration::from_secs(30),
|
||||
progress_ttl: Duration::from_secs(60),
|
||||
cancel_ttl: None,
|
||||
};
|
||||
let mut worker = WorkerConfig::default();
|
||||
assert!(validate_worker_timing(&coordination, &worker).is_ok());
|
||||
worker.renew_interval = coordination.lease_ttl;
|
||||
assert!(validate_worker_timing(&coordination, &worker).is_err());
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in a new issue