rustploy 0.1.11

//! Execução de uma limpeza (agendada ou "Executar agora"): roda os recursos
//! marcados em `DockerCleanupConfig` chamando as MESMAS funções `*_core` que
//! os botões manuais da aba Docker usam, agrega o resultado, publica
//! `Event::DockerCleanupCompleted` e persiste `last_run_at`/`next_run_at` +
//! um resumo, pra tela mostrar "última limpeza" sem depender do SSE.

use super::prune::{self as docker_prune, PruneStat};
use crate::features::maintenance::MaintenanceContext;
use crate::store::daemon_settings;
use chrono::Utc;
use shared::{DockerCleanupConfig, DockerCleanupLastRun, DockerCleanupResourceResult, Event};
use std::time::Duration;
use tokio::sync::OwnedSemaphorePermit;
use tracing::{info, warn};

/// Teto de uma limpeza inteira: ela segura a vaga única da fila de deploy, e
/// uma limpeza pendurada travaria todos os deploys.
const RUN_TIMEOUT: Duration = Duration::from_secs(900);

/// Roda a limpeza segurando a vaga única da fila (`permit`): enquanto ela
/// executa, nenhum deploy, pre-deploy nem job avulso compete por CPU/disco, e
/// ela nunca apaga cache no meio de um build. A vaga solta ao fim.
pub async fn run_holding_slot(
    state: &MaintenanceContext,
    config: DockerCleanupConfig,
    permit: OwnedSemaphorePermit,
) -> DockerCleanupConfig {
    let fallback = config.clone();
    let outcome = tokio::time::timeout(RUN_TIMEOUT, run(state, config)).await;
    drop(permit);
    match outcome {
        Ok(config) => {
            // Já com a vaga solta: cada aquecimento pega a própria vaga, e um
            // deploy que chegue no meio espera um build, não a fila inteira.
            if config.warm_after_cleanup {
                super::service_warm::warm_cold_services(state).await;
            }
            config
        }
        Err(_) => finish_timed_out(state, fallback).await,
    }
}

async fn finish_timed_out(
    state: &MaintenanceContext,
    mut config: DockerCleanupConfig,
) -> DockerCleanupConfig {
    let message = format!(
        "a limpeza passou de {}s e foi interrompida",
        RUN_TIMEOUT.as_secs()
    );
    warn!("maintenance: {message}");
    let now = Utc::now();
    config.last_run_at = Some(now);
    config.recompute_next_run(now);
    if let Err(e) = save_config(state, &config).await {
        warn!(error = %e, "maintenance: falha ao persistir config após timeout");
    }
    state.bus.publish(Event::DockerCleanupCompleted {
        at: now,
        results: vec![DockerCleanupResourceResult {
            resource: "cleanup".to_string(),
            count: 0,
            reclaimed_bytes: 0,
            error: Some(message),
        }],
    });
    config
}

/// Ordem importa: containers primeiro (libera imagens/redes em uso por eles),
/// depois imagens, depois volumes/redes que só ficam "sem uso" após isso.
pub async fn run(
    state: &MaintenanceContext,
    mut config: DockerCleanupConfig,
) -> DockerCleanupConfig {
    let mut results = Vec::new();

    if config.containers {
        results.push(record(
            "containers",
            docker_prune::prune_containers_core(state).await,
        ));
    }
    if config.images {
        results.push(record(
            "images",
            docker_prune::prune_images_core(
                state,
                config.images_all,
                config.images_older_than_hours,
            )
            .await,
        ));
    }
    if config.volumes {
        results.push(record(
            "volumes",
            docker_prune::prune_volumes_core(state, config.volumes_all).await,
        ));
    }
    if config.networks {
        results.push(record(
            "networks",
            docker_prune::prune_networks_core(state).await,
        ));
    }
    if config.build_cache {
        results.push(record(
            "build_cache",
            docker_prune::prune_build_cache_core(state, config.build_cache_older_than_hours).await,
        ));
    }

    let now = Utc::now();
    config.last_run_at = Some(now);
    config.recompute_next_run(now);

    if let Err(e) = save_config(state, &config).await {
        warn!(error = %e, "maintenance: falha ao persistir config após execução");
    }

    let last_run = DockerCleanupLastRun {
        at: now,
        results: results.clone(),
    };
    if let Ok(json) = serde_json::to_string(&last_run)
        && let Err(e) = daemon_settings::set(
            &state.db,
            daemon_settings::KEY_DOCKER_CLEANUP_LAST_RUN,
            &json,
        )
        .await
    {
        warn!(error = %e, "maintenance: falha ao persistir resumo da execução");
    }

    info!(count = results.len(), "maintenance: limpeza concluída");
    state
        .bus
        .publish(Event::DockerCleanupCompleted { at: now, results });

    config
}

fn record(resource: &str, result: Result<PruneStat, String>) -> DockerCleanupResourceResult {
    match result {
        Ok(s) => DockerCleanupResourceResult {
            resource: resource.to_string(),
            count: s.count,
            reclaimed_bytes: s.reclaimed_bytes,
            error: None,
        },
        Err(e) => {
            warn!(resource, error = %e, "maintenance: falha ao limpar recurso");
            DockerCleanupResourceResult {
                resource: resource.to_string(),
                count: 0,
                reclaimed_bytes: 0,
                error: Some(e),
            }
        }
    }
}

pub(crate) async fn save_config(
    state: &MaintenanceContext,
    config: &DockerCleanupConfig,
) -> anyhow::Result<()> {
    let json = serde_json::to_string(config)?;
    daemon_settings::set(&state.db, daemon_settings::KEY_DOCKER_CLEANUP_CONFIG, &json).await
}

pub(crate) async fn load_config(state: &MaintenanceContext) -> anyhow::Result<DockerCleanupConfig> {
    let raw = daemon_settings::get(&state.db, daemon_settings::KEY_DOCKER_CLEANUP_CONFIG).await?;
    Ok(match raw {
        Some(json) => serde_json::from_str(&json).unwrap_or_default(),
        None => DockerCleanupConfig::default(),
    })
}

pub(crate) async fn load_last_run(state: &MaintenanceContext) -> Option<DockerCleanupLastRun> {
    let raw = daemon_settings::get(&state.db, daemon_settings::KEY_DOCKER_CLEANUP_LAST_RUN)
        .await
        .ok()
        .flatten()?;
    serde_json::from_str(&raw).ok()
}