runnerd 0.0.8

//! The control routes: pause (of one repository or all, forced or not) and resume, reseed, cache clean, the config
//! file (show, check, replace and reload) and the doctor, with the sizing advice. Every handler takes the
//! authenticated principal. The engine carries out what changes scheduling through `Control`; this module only
//! writes the config file, atomically, after it validated it.

use std::io::{self, Write};
use std::os::unix::fs::{OpenOptionsExt, PermissionsExt};
use std::path::{Path, PathBuf};

use axum::Json;
use axum::extract::{Path as UrlPath, Query, State};
use axum::http::StatusCode;
use axum::response::{IntoResponse, Response};
use runnerd_core::api::{Busy, ConfigCheck, ConfigFile, Doctor, PauseQuery, ReplaceQuery};
use runnerd_core::config::{Config, revision};
use runnerd_core::scheduler::engine::{Commit, ControlError, RunningJobs, RunningSlots, Target};
use runnerd_core::sizing::Window;
use tracing::info;

use super::{AppState, Authenticated, ConfigText, MemoryMax, SyncDir, error};

pub(super) fn refused(control_error: &ControlError) -> Response {
    let status = match control_error {
        ControlError::NotFound(_) => StatusCode::NOT_FOUND,
        ControlError::Busy { message, jobs } => {
            return (
                StatusCode::CONFLICT,
                Json(Busy {
                    message: message.clone(),
                    jobs: jobs.clone(),
                }),
            )
                .into_response();
        }
        ControlError::Conflict(_) => StatusCode::CONFLICT,
        ControlError::Invalid(_) => StatusCode::UNPROCESSABLE_ENTITY,
        ControlError::Failed(_) | ControlError::NotRecorded(_) => StatusCode::INTERNAL_SERVER_ERROR,
        ControlError::Stopped => StatusCode::SERVICE_UNAVAILABLE,
    };
    error(status, &control_error.to_string())
}

pub(super) async fn pause(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    UrlPath((owner, name)): UrlPath<(String, String)>,
    Query(query): Query<PauseQuery>,
) -> Response {
    let repository = format!("{owner}/{name}");
    info!(
        "{repository}: pause (force: {}) requested by {principal:?}",
        query.force
    );
    pause_target(&state, Target::Repository(repository), &query).await
}

pub(super) async fn pause_all(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    Query(query): Query<PauseQuery>,
) -> Response {
    info!(
        "pause of every repository (force: {}) requested by {principal:?}",
        query.force
    );
    pause_target(&state, Target::All, &query).await
}

async fn pause_target(state: &AppState, target: Target, query: &PauseQuery) -> Response {
    let jobs = if query.force {
        RunningJobs::Kill
    } else {
        RunningJobs::Keep
    };
    match state.handle.control.pause(target, jobs).await {
        Ok(paused) => Json(paused).into_response(),
        Err(control_error) => refused(&control_error),
    }
}

pub(super) async fn resume(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    UrlPath((owner, name)): UrlPath<(String, String)>,
) -> Response {
    let repository = format!("{owner}/{name}");
    let target = Target::Repository(repository.clone());
    match state.handle.control.resume(target).await {
        Ok(()) => {
            info!("{repository} resumed by {principal:?}");
            StatusCode::NO_CONTENT.into_response()
        }
        Err(control_error) => refused(&control_error),
    }
}

pub(super) async fn resume_all(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
) -> Response {
    match state.handle.control.resume(Target::All).await {
        Ok(()) => {
            info!("every repository resumed by {principal:?}");
            StatusCode::NO_CONTENT.into_response()
        }
        Err(control_error) => refused(&control_error),
    }
}

pub(super) async fn reseed(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    UrlPath(profile): UrlPath<String>,
) -> Response {
    info!("{profile}: reseed requested by {principal:?}");
    match state.handle.control.reseed(&profile).await {
        Ok(()) => StatusCode::ACCEPTED.into_response(),
        Err(control_error) => refused(&control_error),
    }
}

pub(super) async fn clean_cache(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    UrlPath(name): UrlPath<String>,
) -> Response {
    info!("{name}: clean requested by {principal:?}");
    match state.handle.control.clean_cache(&name).await {
        Ok(()) => StatusCode::NO_CONTENT.into_response(),
        Err(control_error) => refused(&control_error),
    }
}

pub(super) async fn config(_: Authenticated, State(state): State<AppState>) -> Json<ConfigFile> {
    let file = state.config.lock().await;
    Json(ConfigFile {
        path: file.path.display().to_string(),
        text: file.text.clone(),
        revision: Some(revision(&file.text)),
    })
}

/// Checks a config off the async threads: reading a slice's `MemoryMax` runs `systemctl`.
async fn check(
    memory_max: MemoryMax,
    text: String,
) -> Result<(Option<Config>, ConfigCheck), tokio::task::JoinError> {
    tokio::task::spawn_blocking(move || Config::check(&text, memory_max)).await
}

fn failed(join_error: &tokio::task::JoinError) -> Response {
    error(StatusCode::INTERNAL_SERVER_ERROR, &join_error.to_string())
}

pub(super) async fn check_config(
    _: Authenticated,
    State(state): State<AppState>,
    text: String,
) -> Response {
    match check(state.memory_max, text).await {
        Ok((_, checked)) => Json(checked).into_response(),
        Err(join_error) => failed(&join_error),
    }
}

/// Validates the new config; the engine writes it in place of the file and runs with it. A config that is invalid,
/// or whose file cannot be written, leaves the file and the engine as they were.
pub(super) async fn replace_config(
    Authenticated(principal): Authenticated,
    State(state): State<AppState>,
    Query(query): Query<ReplaceQuery>,
    text: String,
) -> Response {
    // One replacement at a time, so the file and the engine end up with the same config.
    let mut file = state.config.lock().await;
    // A text made from a config replaced since, as by auto sizing, or edited on disk since, would undo that change.
    let on_disk = match query.revision {
        None => None,
        Some(read) if read != revision(&file.text) => {
            return error(
                StatusCode::CONFLICT,
                &format!(
                    "the config changed since revision {read} was read (now {}); read it again and redo the change",
                    revision(&file.text)
                ),
            );
        }
        Some(_) => {
            if std::fs::read_to_string(&file.path).ok().as_ref() != Some(&file.text) {
                return error(
                    StatusCode::CONFLICT,
                    &format!(
                        "{} changed on disk since runnerd read it; saving would overwrite that edit",
                        file.path.display()
                    ),
                );
            }
            // Checked again when the engine writes it.
            Some(file.text.clone())
        }
    };
    match replace(&state, &mut file, text, RunningSlots::Drain, on_disk).await {
        Ok(checked) => {
            info!("{} replaced by {principal:?}", file.path.display());
            Json(checked).into_response()
        }
        Err(Refusal::Invalid(checked)) => {
            (StatusCode::UNPROCESSABLE_ENTITY, Json(checked)).into_response()
        }
        Err(Refusal::Control(control_error)) => refused(&control_error),
        Err(Refusal::Join(join_error)) => failed(&join_error),
    }
}

/// Why `replace` left the config as it was.
pub(super) enum Refusal {
    /// The new config does not validate.
    Invalid(ConfigCheck),
    /// The engine did not run it, or its file was not written.
    Control(ControlError),
    Join(tokio::task::JoinError),
}

impl std::fmt::Display for Refusal {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Refusal::Invalid(checked) => f.write_str(&checked.message),
            Refusal::Control(control_error) => control_error.fmt(f),
            Refusal::Join(join_error) => join_error.fmt(f),
        }
    }
}

/// Validates `text` against the slice's budget; the engine writes it in place of `file` and runs with it, between
/// its ticks, as `running` allows. With `on_disk`, the file is written only if it still holds that text then. What
/// `PUT /v1/config` and auto sizing do; the caller holds `file`'s lock.
pub(super) async fn replace(
    state: &AppState,
    file: &mut ConfigText,
    text: String,
    running: RunningSlots,
    on_disk: Option<String>,
) -> Result<ConfigCheck, Refusal> {
    let (config, mut checked) = check(state.memory_max, text.clone())
        .await
        .map_err(Refusal::Join)?;
    let Some(config) = config else {
        return Err(Refusal::Invalid(checked));
    };
    // Only a CPU budget needs the slice's CPUs: without one, a slice systemd cannot describe still runs.
    let cpus = if config.host.cpu_overcommit.is_some() {
        let (read, slice) = (state.cpus, config.host.slice.clone());
        match tokio::task::spawn_blocking(move || read(&slice)).await {
            Ok(Ok(cpus)) => Some(cpus),
            Ok(Err(io_error)) => {
                checked.valid = false;
                checked.message = format!(
                    "cpu_overcommit needs the slice {}'s CPUs, and they cannot be read: {io_error}",
                    config.host.slice
                );
                return Err(Refusal::Invalid(checked));
            }
            Err(join_error) => return Err(Refusal::Join(join_error)),
        }
    } else {
        None
    };
    let database = |config: &Config| config.history.database.clone();
    if Config::parse(&file.text).ok().as_ref().map(database) != Some(database(&config)) {
        checked
            .warnings
            .push("history.database takes effect when runnerd restarts".to_owned());
    }
    // The engine writes the file between its ticks and runs with the config before the next: nothing runs under a
    // config whose file was not written, and nothing changes if it was not.
    let (path, written, sync) = (file.path.clone(), text.clone(), state.sync_dir);
    let commit = Commit::new(move || {
        if let Some(expected) = on_disk {
            match std::fs::read_to_string(&path) {
                Ok(text) if text == expected => {}
                Ok(_) => {
                    return Err(format!(
                        "{} changed on disk since it was read; not written",
                        path.display()
                    ));
                }
                Err(io_error) => return Err(format!("cannot read {}: {io_error}", path.display())),
            }
        }
        commit(&path, &written, sync)
    });
    let warning = state
        .handle
        .control
        .replace_config(config, checked.memory_max, cpus, commit, running)
        .await
        .map_err(Refusal::Control)?;
    checked.warnings.extend(warning);
    checked.revision = Some(revision(&text));
    file.text = text;
    Ok(checked)
}

/// Writes `text` in place of `path`: `Err` when it was not written, a warning when it was but may not last a crash.
fn commit(path: &Path, text: &str, sync: SyncDir) -> Result<Option<String>, String> {
    match write_atomically(path, text, sync) {
        Ok(None) => Ok(None),
        // In place: the engine runs with it, as it would after a restart.
        Ok(Some(io_error)) => Ok(Some(format!(
            "{} is replaced, but flushing its directory to disk failed, so a crash may bring back the old file: {io_error}",
            path.display()
        ))),
        Err(io_error) => Err(format!("cannot write {}: {io_error}", path.display())),
    }
}

/// Flushes a directory's entries, such as a rename in it, to disk.
pub(crate) fn sync_dir(dir: &Path) -> io::Result<()> {
    std::fs::File::open(dir)?.sync_all()
}

/// Replaces the file `path` points to, through a symlink if it is one, with `text`: a temporary file in the same
/// directory with the same permissions, renamed over it, so a reader sees the old file or the new one. `Err` leaves
/// the old file; once renamed, the new file is in place, and `Ok` carries the error if flushing the rename to disk
/// with `sync_dir` failed, which a crash could undo.
fn write_atomically(path: &Path, text: &str, sync_dir: SyncDir) -> io::Result<Option<io::Error>> {
    let target = std::fs::canonicalize(path)?;
    let mode = std::fs::metadata(&target)?.permissions().mode();
    let dir = target.parent().unwrap_or(Path::new("/"));
    let name = target
        .file_name()
        .map(|name| name.to_string_lossy().into_owned())
        .unwrap_or_default();
    let temporary: PathBuf = dir.join(format!(".{name}.{}.tmp", std::process::id()));
    let written = (|| {
        let mut file = std::fs::OpenOptions::new()
            .write(true)
            .create_new(true)
            .mode(mode)
            .open(&temporary)?;
        // `mode` passes through the umask.
        file.set_permissions(std::fs::Permissions::from_mode(mode))?;
        file.write_all(text.as_bytes())?;
        file.sync_all()?;
        std::fs::rename(&temporary, &target)
    })();
    if let Err(io_error) = written {
        let _ = std::fs::remove_file(&temporary);
        return Err(io_error);
    }
    Ok(sync_dir(dir).err())
}

pub(super) async fn doctor(_: Authenticated, State(state): State<AppState>) -> Response {
    let sizing = super::history::sizing_now(&state, Window::new(None, super::history::now())).await;
    let text = state.config.lock().await.text.clone();
    let checked = tokio::task::spawn_blocking(move || {
        let config = Config::parse(&text).ok();
        let mut host = crate::doctor::Host::read(config.as_ref());
        host.sizing = Some(sizing);
        crate::doctor::checks(&host, config.as_ref())
    })
    .await;
    match checked {
        Ok(checks) => Json(Doctor { checks }).into_response(),
        Err(join_error) => failed(&join_error),
    }
}