rwhod: let users specify user-idle/machine-down timeouts

This commit is contained in:
2026-07-20 20:49:15 +09:00
parent 59621048f8
commit d67aaba58f
3 changed files with 129 additions and 37 deletions
+28 -6
View File
@@ -13,14 +13,24 @@ use roowho2_lib::{
/// `ruptime` gives a status line like uptime for each machine on the local network;
/// these are formed from packets broadcast by each host on the network once a minute.
///
/// Machines for which no status report has been received for 11 minutes are shown as being down.
/// Machines for which no status report has been received for a while (11 minutes by default,
/// see `--uptime-timeout`) are shown as being down.
#[derive(Debug, Parser)]
#[command(author = "Programvareverkstedet <projects@pvv.ntnu.no>", version)]
pub struct Args {
/// Users idle an hour or more are not counted unless the `-a` flag is given.
#[arg(long, short)]
#[arg(long, short, conflicts_with = "idle_timeout")]
all: bool,
/// Only count users idle for less than this many minutes.
/// Cannot be used together with `--all`.
#[arg(long, value_name = "MINUTES", default_value_t = DEFAULT_IDLE_TIMEOUT_MINUTES)]
idle_timeout: u32,
/// Consider a machine down if no status report has been received from it for this many minutes.
#[arg(long, value_name = "MINUTES", default_value_t = DEFAULT_UPTIME_TIMEOUT_MINUTES)]
uptime_timeout: u32,
/// Sort by load average.
#[arg(long, short, conflicts_with = "time", conflicts_with = "users")]
load: bool,
@@ -51,6 +61,12 @@ pub struct Args {
completions: Option<Shell>,
}
/// Users idle for longer than this are not counted by default.
const DEFAULT_IDLE_TIMEOUT_MINUTES: u32 = 60;
/// Machines that haven't reported in for longer than this are shown as down by default.
const DEFAULT_UPTIME_TIMEOUT_MINUTES: u32 = 11;
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let args = Args::parse();
@@ -65,12 +81,18 @@ async fn main() -> anyhow::Result<()> {
return Ok(());
}
let max_idle_seconds = if args.all {
None
} else {
Some(i64::from(args.idle_timeout) * 60)
};
let mut conn = zlink::tokio::unix::connect("/run/roowho2/roowho2.varlink")
.await
.expect("Failed to connect to rwhod server");
let mut reply = conn
.ruptime(args.all)
.ruptime(max_idle_seconds)
.await
.context("Failed to send rwho request")?
.map_err(|e| match e {
@@ -96,7 +118,7 @@ async fn main() -> anyhow::Result<()> {
// }
} else {
for entry in &reply {
let line = old_format_machine_entry(entry);
let line = old_format_machine_entry(entry, args.uptime_timeout);
println!("{}", line);
}
}
@@ -138,9 +160,9 @@ fn sort_entries(
});
}
fn old_format_machine_entry(entry: &WhodStatusUpdate) -> String {
fn old_format_machine_entry(entry: &WhodStatusUpdate, uptime_timeout_minutes: u32) -> String {
let time_since_last_ping = Utc::now() - entry.sendtime;
let is_up = time_since_last_ping <= Duration::minutes(11);
let is_up = time_since_last_ping <= Duration::minutes(i64::from(uptime_timeout_minutes));
let uptime = Utc::now() - entry.boot_time;
let days = uptime.num_days();
+16 -2
View File
@@ -19,9 +19,14 @@ use roowho2_lib::{
#[command(author = "Programvareverkstedet <projects@pvv.ntnu.no>", version)]
pub struct Args {
/// Print all machines responding even if no one is currently logged in
#[arg(long, short)]
#[arg(long, short, conflicts_with = "idle_timeout")]
all: bool,
/// Only show users idle for less than this many minutes.
/// Cannot be used together with `--all`.
#[arg(long, value_name = "MINUTES", default_value_t = DEFAULT_IDLE_TIMEOUT_MINUTES)]
idle_timeout: u32,
/// Print the output with the old formatting
#[arg(long, short)]
old: bool,
@@ -36,6 +41,9 @@ pub struct Args {
completions: Option<Shell>,
}
/// Users idle for longer than this are omitted from the output by default.
const DEFAULT_IDLE_TIMEOUT_MINUTES: u32 = 60;
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let args = Args::parse();
@@ -45,12 +53,18 @@ async fn main() -> anyhow::Result<()> {
return Ok(());
}
let max_idle_seconds = if args.all {
None
} else {
Some(i64::from(args.idle_timeout) * 60)
};
let mut conn = zlink::tokio::unix::connect("/run/roowho2/roowho2.varlink")
.await
.expect("Failed to connect to rwhod server");
let mut reply = conn
.rwho(args.all)
.rwho(max_idle_seconds)
.await
.context("Failed to send rwho request")?
.map_err(|e| match e {
+85 -29
View File
@@ -24,14 +24,20 @@ use crate::{
#[zlink::proxy("no.ntnu.pvv.roowho2.rwhod")]
pub trait VarlinkRwhodClientProxy {
/// `max_idle_seconds` is the maximum idle time (in seconds) for a user
/// to be included in the response. `None` means no limit (i.e. the old
/// `all: true`). The client is expected to always send an explicit
/// value, defaulting to whatever it considers reasonable if the user
/// didn't ask for a specific limit.
async fn rwho(
&mut self,
all: bool,
max_idle_seconds: Option<i64>,
) -> zlink::Result<Result<VarlinkRwhoResponse, VarlinkRwhodClientError>>;
/// See [`VarlinkRwhodClientProxy::rwho`] for the meaning of `max_idle_seconds`.
async fn ruptime(
&mut self,
all: bool,
max_idle_seconds: Option<i64>,
) -> zlink::Result<Result<VarlinkRuptimeResponse, VarlinkRwhodClientError>>;
}
@@ -40,14 +46,16 @@ pub trait VarlinkRwhodClientProxy {
pub enum VarlinkRwhodClientRequest {
#[serde(rename = "no.ntnu.pvv.roowho2.rwhod.Rwho")]
Rwho {
/// Retrieve all users, even those that have been idle for a long time.
all: bool,
/// Maximum idle time (in seconds) for a user to be included.
/// `None` means no limit (i.e. return all users).
max_idle_seconds: Option<i64>,
},
#[serde(rename = "no.ntnu.pvv.roowho2.rwhod.Ruptime")]
Ruptime {
/// Count all users, even those that have been idle for a long time.
all: bool,
/// Maximum idle time (in seconds) for a user to be counted.
/// `None` means no limit (i.e. return all users).
max_idle_seconds: Option<i64>,
},
}
@@ -165,8 +173,11 @@ impl VarlinkRoowhoo2ClientServer {
}
impl VarlinkRoowhoo2ClientServer {
async fn handle_rwho_request(&self, all: bool) -> VarlinkRwhoResponse {
tracing::debug!(all, "Handling Rwho request");
async fn handle_rwho_request(
&self,
max_idle_time: Option<chrono::TimeDelta>,
) -> VarlinkRwhoResponse {
tracing::debug!(?max_idle_time, "Handling Rwho request");
let store = self.whod_status_store.read().await;
let mut all_user_entries = Vec::with_capacity(store.len());
@@ -175,7 +186,7 @@ impl VarlinkRoowhoo2ClientServer {
status_update
.users
.iter()
.filter(|user| all || user.idle_time < chrono::Duration::hours(1))
.filter(|user| max_idle_time.is_none_or(|max| user.idle_time < max))
.cloned()
.map(|user| (status_update.hostname.clone(), user)),
);
@@ -184,18 +195,21 @@ impl VarlinkRoowhoo2ClientServer {
all_user_entries
}
async fn handle_ruptime_request(&self, all: bool) -> VarlinkRuptimeResponse {
tracing::debug!(all, "Handling Ruptime request");
async fn handle_ruptime_request(
&self,
max_idle_time: Option<chrono::TimeDelta>,
) -> VarlinkRuptimeResponse {
tracing::debug!(?max_idle_time, "Handling Ruptime request");
let store = self.whod_status_store.read().await;
store
.values()
.cloned()
.map(|mut status_update| {
if !all {
if let Some(max_idle_time) = max_idle_time {
status_update
.users
.retain(|user| user.idle_time < chrono::Duration::hours(1));
.retain(|user| user.idle_time < max_idle_time);
}
status_update
})
@@ -287,7 +301,7 @@ impl zlink::Service<UnixStream> for VarlinkRoowhoo2ClientServer {
Self::ReplyError<'service>,
> {
match call.method() {
VarlinkMethod::Rwhod(VarlinkRwhodClientRequest::Rwho { all }) => {
VarlinkMethod::Rwhod(VarlinkRwhodClientRequest::Rwho { max_idle_seconds }) => {
if !self.rwhod_enabled {
return (
MethodReply::Error(VarlinkReplyError::Rwhod(
@@ -297,19 +311,42 @@ impl zlink::Service<UnixStream> for VarlinkRoowhoo2ClientServer {
);
}
let result =
match timeout(Duration::from_secs(2), self.handle_rwho_request(*all)).await {
Ok(response) => response,
Err(_) => {
tracing::error!("Rwho request timed out after 2 seconds");
return (
MethodReply::Error(VarlinkReplyError::Rwhod(
VarlinkRwhodClientError::TimedOut,
)),
Default::default(),
);
}
};
let max_idle_time = match max_idle_seconds
.map(|secs| chrono::TimeDelta::try_seconds(secs).ok_or(()))
.transpose()
.map_err(|_| {
tracing::error!(
"Received out-of-range max_idle_seconds: {:?}",
max_idle_seconds
);
(
MethodReply::Error(VarlinkReplyError::Rwhod(
VarlinkRwhodClientError::InvalidRequest,
)),
Default::default(),
)
}) {
Ok(max_idle_time) => max_idle_time,
Err(err) => return err,
};
let result = match timeout(
Duration::from_secs(2),
self.handle_rwho_request(max_idle_time),
)
.await
{
Ok(response) => response,
Err(_) => {
tracing::error!("Rwho request timed out after 2 seconds");
return (
MethodReply::Error(VarlinkReplyError::Rwhod(
VarlinkRwhodClientError::TimedOut,
)),
Default::default(),
);
}
};
(
MethodReply::Single(Some(VarlinkReply::Rwhod(
@@ -318,7 +355,7 @@ impl zlink::Service<UnixStream> for VarlinkRoowhoo2ClientServer {
Default::default(),
)
}
VarlinkMethod::Rwhod(VarlinkRwhodClientRequest::Ruptime { all }) => {
VarlinkMethod::Rwhod(VarlinkRwhodClientRequest::Ruptime { max_idle_seconds }) => {
if !self.rwhod_enabled {
return (
MethodReply::Error(VarlinkReplyError::Rwhod(
@@ -328,9 +365,28 @@ impl zlink::Service<UnixStream> for VarlinkRoowhoo2ClientServer {
);
}
let max_idle_time = match max_idle_seconds
.map(|secs| chrono::TimeDelta::try_seconds(secs).ok_or(()))
.transpose()
.map_err(|_| {
tracing::error!(
"Received out-of-range max_idle_seconds: {:?}",
max_idle_seconds
);
(
MethodReply::Error(VarlinkReplyError::Rwhod(
VarlinkRwhodClientError::InvalidRequest,
)),
Default::default(),
)
}) {
Ok(max_idle_time) => max_idle_time,
Err(err) => return err,
};
let result = match timeout(
Duration::from_secs(2),
self.handle_ruptime_request(*all),
self.handle_ruptime_request(max_idle_time),
)
.await
{