synops-backup: PG-dump + CAS-filiste + metadata-snapshot (oppgave 28.6)

Nytt Rust CLI-verktøy som erstatter scripts/backup-pg.sh: - pg_dump -Fc via docker exec (konsistent snapshot) - CAS-manifest: liste over alle filer med hash og størrelse - Metadata-snapshot (JSON) med tidsstempel, modus, statistikk - --full / --incremental / --payload-json for jobbkø - Rotasjon av gamle dumper (30 dager, kun ved --full) Output: strukturert JSON med backup-sti og detaljer.
2026-03-18 20:45:02 +00:00 · 2026-03-18 20:45:02 +00:00 · f1e6355037
commit f1e6355037
parent 0f7fdb75e0
6 changed files with 3006 additions and 7 deletions
--- a/docs/infra/backup.md
+++ b/docs/infra/backup.md
@ -17,22 +17,29 @@ PostgreSQL (autoritativ kilde)
  └──→ Sanntid via PG LISTEN/NOTIFY → WebSocket
 ```

-## 1. PG-dump (daglig)
+## 1. Backup via `synops-backup`

-**Script:** `scripts/backup-pg.sh`
+**Verktøy:** `tools/synops-backup/` (Rust CLI, erstatter `scripts/backup-pg.sh`)
 **Cron:** `/etc/cron.d/synops-backup` — `0 3 * * *`
-**Logg:** `/srv/synops/logs/backup-pg.log`
 **Dumper:** `/srv/synops/backup/pg/`
+**Manifester:** `/srv/synops/backup/manifests/`

 Prosess:
 1. Sjekker at PG-containeren kjører
 2. `pg_dump -Fc` (custom format, komprimert) — konsistent snapshot uten nedetid
 3. Verifiserer at dump-filen ikke er tom
-4. Sletter dumper eldre enn 30 dager
+4. Bygger CAS-manifest (liste over alle filer med hash og størrelse)
+5. Skriver metadata-snapshot (JSON) med tidsstempel, modus og statistikk
+6. Roterer dumper eldre enn 30 dager (kun ved `--full`)
+
+Moduser:
+- `--full` — Full PG-dump + komplett CAS-filiste + metadata
+- `--incremental` — PG-dump + kun nye CAS-filer siden forrige manifest
+- `--payload-json '{"mode":"full"}'` — Jobbkø-dispatch

 Manuell kjøring:
 ```bash
-/home/vegard/synops/scripts/backup-pg.sh
+synops-backup --full
 ```

 Verifiser dump:
--- a/tasks.md
+++ b/tasks.md
@ -380,8 +380,7 @@ modell som brukes til hva.

 - [x] 28.4 `synops-notify`: send varsel via epost (synops-mail), WebSocket-push, eller begge. Input: `--to <node_id> --message <tekst> [--channel email|ws|both]`. Brukes av orkestreringer og vaktmesteren.
 - [x] 28.5 `synops-validate`: sjekk at en node matcher forventet skjema for sin node_kind. Input: `--node-id <uuid>`. Output: liste av avvik. Brukes av valideringsfasen og som pre-commit sjekk.
- [~] 28.6 `synops-backup`: PG-dump + CAS-filiste + metadata-snapshot. Input: `[--full | --incremental]`. Output: backup-sti. Erstatter cron-scriptet fra 12.2.
-  > Påbegynt: 2026-03-18T20:40
+- [x] 28.6 `synops-backup`: PG-dump + CAS-filiste + metadata-snapshot. Input: `[--full | --incremental]`. Output: backup-sti. Erstatter cron-scriptet fra 12.2.
 - [ ] 28.7 `synops-health`: sjekk status for alle tjenester (PG, Caddy, vaktmesteren, LiteLLM, Whisper, LiveKit). Output: JSON med status per tjeneste. Brukes av admin-dashboard og overvåking.

 ## Fase 29: Universell input — alle modaliteter blir noder
--- a/tools/README.md
+++ b/tools/README.md
@ -26,6 +26,7 @@ eller maskinrommet-API. Ligger i PATH via symlink eller direkte kall.
 | `synops-mail` | Send epost via msmtp (vaktmester@synops.no) | Ferdig (venter SMTP-credentials) |
 | `synops-notify` | Send varsel via epost, WebSocket-push, eller begge | Ferdig |
 | `synops-validate` | Valider at en node matcher forventet skjema for sin node_kind | Ferdig |
+| `synops-backup` | PG-dump + CAS-filiste + metadata-snapshot (`--full` / `--incremental`) | Ferdig |

 ## Delt bibliotek

--- a/tools/synops-backup/Cargo.lock
+++ b/tools/synops-backup/Cargo.lock
--- a/tools/synops-backup/Cargo.toml
+++ b/tools/synops-backup/Cargo.toml
@ -0,0 +1,17 @@
+[package]
+name = "synops-backup"
+version = "0.1.0"
+edition = "2024"
+
+[[bin]]
+name = "synops-backup"
+path = "src/main.rs"
+
+[dependencies]
+clap = { version = "4", features = ["derive", "env"] }
+synops-common = { path = "../synops-common" }
+chrono = { version = "0.4", features = ["serde"] }
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+tokio = { version = "1", features = ["full"] }
+tracing = "0.1"
--- a/tools/synops-backup/src/main.rs
+++ b/tools/synops-backup/src/main.rs
@ -0,0 +1,528 @@
+// synops-backup — PG-dump + CAS-filiste + metadata-snapshot.
+//
+// Erstatter scripts/backup-pg.sh (cron-script fra oppgave 12.2).
+// Gjør tre ting:
+//   1. pg_dump -Fc via docker exec (konsistent snapshot)
+//   2. Lister alle CAS-filer med størrelse
+//   3. Skriver metadata-snapshot (JSON) med tidsstempel, modus, statistikk
+//
+// Bruk:
+//   synops-backup --full              # Full backup
+//   synops-backup --incremental       # Inkrementell (kun endringer siden forrige)
+//   synops-backup --payload-json '{"mode":"full"}'  # Jobbkø-dispatch
+//
+// Output: backup-sti til stdout (JSON).
+// Feil: stderr + exit code != 0.
+//
+// Ref: docs/infra/backup.md, docs/retninger/unix_filosofi.md
+
+use chrono::Utc;
+use clap::Parser;
+use serde::Serialize;
+use std::path::{Path, PathBuf};
+
+/// Backup-modus.
+#[derive(Debug, Clone, Copy, PartialEq)]
+enum Mode {
+    Full,
+    Incremental,
+}
+
+/// PG-dump + CAS-filiste + metadata-snapshot.
+#[derive(Parser)]
+#[command(name = "synops-backup", about = "Backup: PG-dump, CAS-filiste og metadata-snapshot")]
+struct Cli {
+    /// Full backup (PG-dump + CAS-liste + metadata)
+    #[arg(long, conflicts_with = "incremental")]
+    full: bool,
+
+    /// Inkrementell backup (kun CAS-filer nyere enn forrige backup)
+    #[arg(long, conflicts_with = "full")]
+    incremental: bool,
+
+    /// Backup-katalog (default: /srv/synops/backup)
+    #[arg(long, default_value = "/srv/synops/backup")]
+    backup_dir: String,
+
+    /// PG Docker-container (default: sidelinja-postgres-1)
+    #[arg(long, default_value = "sidelinja-postgres-1")]
+    pg_container: String,
+
+    /// Database-bruker (default: sidelinja)
+    #[arg(long, default_value = "sidelinja")]
+    db_user: String,
+
+    /// Databasenavn (default: sidelinja)
+    #[arg(long, default_value = "sidelinja")]
+    db_name: String,
+
+    /// Antall dager å beholde gamle backuper (default: 30)
+    #[arg(long, default_value = "30")]
+    retain_days: u32,
+
+    /// Payload fra jobbkø (JSON). Overstyrer andre argumenter.
+    #[arg(long)]
+    payload_json: Option<String>,
+}
+
+/// Resultat fra backup.
+#[derive(Serialize)]
+struct BackupResult {
+    ok: bool,
+    mode: String,
+    timestamp: String,
+    backup_dir: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pg_dump: Option<PgDumpResult>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    cas_manifest: Option<CasManifestResult>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    metadata_file: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    rotation: Option<RotationResult>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    error: Option<String>,
+}
+
+#[derive(Serialize)]
+struct PgDumpResult {
+    path: String,
+    size_bytes: u64,
+    size_human: String,
+}
+
+#[derive(Serialize)]
+struct CasManifestResult {
+    path: String,
+    file_count: u64,
+    total_size_bytes: u64,
+    total_size_human: String,
+}
+
+#[derive(Serialize)]
+struct RotationResult {
+    deleted_count: u32,
+    retain_days: u32,
+}
+
+/// Metadata-snapshot som skrives til backup-katalogen.
+#[derive(Serialize)]
+struct MetadataSnapshot {
+    timestamp: String,
+    mode: String,
+    pg_dump_size_bytes: Option<u64>,
+    cas_file_count: Option<u64>,
+    cas_total_size_bytes: Option<u64>,
+    hostname: String,
+    db_name: String,
+}
+
+#[tokio::main]
+async fn main() {
+    synops_common::logging::init("synops_backup");
+
+    let cli = Cli::parse();
+
+    // Resolve mode fra args eller payload-json
+    let (mode, backup_dir, pg_container, db_user, db_name, retain_days) =
+        if let Some(ref json_str) = cli.payload_json {
+            let payload: serde_json::Value = serde_json::from_str(json_str).unwrap_or_else(|e| {
+                eprintln!("Ugyldig --payload-json: {e}");
+                std::process::exit(1);
+            });
+            let mode = match payload["mode"].as_str().unwrap_or("full") {
+                "incremental" => Mode::Incremental,
+                _ => Mode::Full,
+            };
+            let backup_dir = payload["backup_dir"]
+                .as_str()
+                .unwrap_or("/srv/synops/backup")
+                .to_string();
+            let pg_container = payload["pg_container"]
+                .as_str()
+                .unwrap_or("sidelinja-postgres-1")
+                .to_string();
+            let db_user = payload["db_user"]
+                .as_str()
+                .unwrap_or("sidelinja")
+                .to_string();
+            let db_name = payload["db_name"]
+                .as_str()
+                .unwrap_or("sidelinja")
+                .to_string();
+            let retain_days = payload["retain_days"].as_u64().unwrap_or(30) as u32;
+            (mode, backup_dir, pg_container, db_user, db_name, retain_days)
+        } else {
+            let mode = if cli.incremental {
+                Mode::Incremental
+            } else {
+                Mode::Full // default
+            };
+            (
+                mode,
+                cli.backup_dir,
+                cli.pg_container,
+                cli.db_user,
+                cli.db_name,
+                cli.retain_days,
+            )
+        };
+
+    let timestamp = Utc::now().format("%Y%m%d_%H%M%S").to_string();
+    let timestamp_iso = Utc::now().to_rfc3339();
+
+    // Opprett backup-kataloger
+    let pg_dir = PathBuf::from(&backup_dir).join("pg");
+    let manifest_dir = PathBuf::from(&backup_dir).join("manifests");
+    for dir in [&pg_dir, &manifest_dir] {
+        if let Err(e) = tokio::fs::create_dir_all(dir).await {
+            let result = BackupResult {
+                ok: false,
+                mode: mode_str(mode).to_string(),
+                timestamp: timestamp_iso.clone(),
+                backup_dir: backup_dir.clone(),
+                pg_dump: None,
+                cas_manifest: None,
+                metadata_file: None,
+                rotation: None,
+                error: Some(format!("Kunne ikke opprette katalog {}: {e}", dir.display())),
+            };
+            println!("{}", serde_json::to_string_pretty(&result).unwrap());
+            std::process::exit(1);
+        }
+    }
+
+    let mut all_ok = true;
+    let mut pg_dump_result = None;
+    let mut cas_manifest_result = None;
+    let mut rotation_result = None;
+    let mut error_msg = None;
+
+    // 1. PG-dump
+    tracing::info!(mode = mode_str(mode), "Starter backup");
+    match run_pg_dump(&pg_container, &db_user, &db_name, &pg_dir, &timestamp).await {
+        Ok(r) => {
+            tracing::info!(path = %r.path, size = r.size_human, "PG-dump ferdig");
+            pg_dump_result = Some(r);
+        }
+        Err(e) => {
+            tracing::error!(error = %e, "PG-dump feilet");
+            all_ok = false;
+            error_msg = Some(e);
+        }
+    }
+
+    // 2. CAS-filiste
+    let cas_root = synops_common::cas::root();
+    match build_cas_manifest(&cas_root, &manifest_dir, &timestamp, mode).await {
+        Ok(r) => {
+            tracing::info!(
+                files = r.file_count,
+                size = r.total_size_human,
+                "CAS-manifest ferdig"
+            );
+            cas_manifest_result = Some(r);
+        }
+        Err(e) => {
+            tracing::error!(error = %e, "CAS-manifest feilet");
+            all_ok = false;
+            if let Some(ref mut existing) = error_msg {
+                existing.push_str(&format!("; CAS: {e}"));
+            } else {
+                error_msg = Some(e);
+            }
+        }
+    }
+
+    // 3. Metadata-snapshot
+    let hostname = std::env::var("HOSTNAME")
+        .or_else(|_| {
+            std::process::Command::new("hostname")
+                .output()
+                .map(|o| String::from_utf8_lossy(&o.stdout).trim().to_string())
+        })
+        .unwrap_or_else(|_| "unknown".to_string());
+
+    let metadata = MetadataSnapshot {
+        timestamp: timestamp_iso.clone(),
+        mode: mode_str(mode).to_string(),
+        pg_dump_size_bytes: pg_dump_result.as_ref().map(|r| r.size_bytes),
+        cas_file_count: cas_manifest_result.as_ref().map(|r| r.file_count),
+        cas_total_size_bytes: cas_manifest_result.as_ref().map(|r| r.total_size_bytes),
+        hostname,
+        db_name: db_name.clone(),
+    };
+
+    let metadata_path = manifest_dir.join(format!("backup_{timestamp}.json"));
+    let metadata_json = serde_json::to_string_pretty(&metadata).unwrap();
+    if let Err(e) = tokio::fs::write(&metadata_path, &metadata_json).await {
+        tracing::error!(error = %e, "Kunne ikke skrive metadata-snapshot");
+        all_ok = false;
+    }
+
+    // 4. Rotasjon (kun ved full backup)
+    if mode == Mode::Full && all_ok {
+        match rotate_old_backups(&pg_dir, retain_days).await {
+            Ok(r) => {
+                if r.deleted_count > 0 {
+                    tracing::info!(
+                        deleted = r.deleted_count,
+                        retain_days = r.retain_days,
+                        "Rotasjon ferdig"
+                    );
+                }
+                rotation_result = Some(r);
+            }
+            Err(e) => {
+                tracing::warn!(error = %e, "Rotasjon feilet (backup er OK)");
+                // Rotasjonsfeil er ikke kritisk — backupen er allerede tatt
+            }
+        }
+    }
+
+    // Output
+    let result = BackupResult {
+        ok: all_ok,
+        mode: mode_str(mode).to_string(),
+        timestamp: timestamp_iso,
+        backup_dir,
+        pg_dump: pg_dump_result,
+        cas_manifest: cas_manifest_result,
+        metadata_file: Some(metadata_path.to_string_lossy().to_string()),
+        rotation: rotation_result,
+        error: error_msg,
+    };
+
+    println!("{}", serde_json::to_string_pretty(&result).unwrap());
+
+    if !all_ok {
+        std::process::exit(1);
+    }
+}
+
+fn mode_str(mode: Mode) -> &'static str {
+    match mode {
+        Mode::Full => "full",
+        Mode::Incremental => "incremental",
+    }
+}
+
+/// Kjør pg_dump via docker exec. Returnerer dump-filsti og størrelse.
+async fn run_pg_dump(
+    container: &str,
+    db_user: &str,
+    db_name: &str,
+    pg_dir: &Path,
+    timestamp: &str,
+) -> Result<PgDumpResult, String> {
+    // Sjekk at containeren kjører
+    let inspect = tokio::process::Command::new("docker")
+        .args(["inspect", container, "--format", "{{.State.Running}}"])
+        .output()
+        .await
+        .map_err(|e| format!("Kunne ikke kjøre docker inspect: {e}"))?;
+
+    let running = String::from_utf8_lossy(&inspect.stdout).trim().to_string();
+    if running != "true" {
+        return Err(format!("PostgreSQL-container {container} kjører ikke"));
+    }
+
+    let dump_file = pg_dir.join(format!("{db_name}_{timestamp}.dump"));
+
+    // pg_dump via docker exec, output til fil
+    let output = tokio::process::Command::new("docker")
+        .args([
+            "exec", container, "pg_dump", "-U", db_user, "-Fc", db_name,
+        ])
+        .stdout(std::process::Stdio::piped())
+        .stderr(std::process::Stdio::piped())
+        .output()
+        .await
+        .map_err(|e| format!("Kunne ikke kjøre pg_dump: {e}"))?;
+
+    if !output.status.success() {
+        let stderr = String::from_utf8_lossy(&output.stderr);
+        return Err(format!("pg_dump feilet: {stderr}"));
+    }
+
+    // Skriv dump til fil
+    tokio::fs::write(&dump_file, &output.stdout)
+        .await
+        .map_err(|e| format!("Kunne ikke skrive dump-fil: {e}"))?;
+
+    let size_bytes = output.stdout.len() as u64;
+    if size_bytes < 100 {
+        // Slett tom/korrupt dump
+        let _ = tokio::fs::remove_file(&dump_file).await;
+        return Err(format!(
+            "Dump-filen er for liten ({size_bytes} bytes), noe gikk galt"
+        ));
+    }
+
+    Ok(PgDumpResult {
+        path: dump_file.to_string_lossy().to_string(),
+        size_bytes,
+        size_human: human_size(size_bytes),
+    })
+}
+
+/// Bygg CAS-manifest: liste over alle filer med hash og størrelse.
+///
+/// Full modus: lister alle filer.
+/// Inkrementell: lister kun filer nyere enn forrige manifest.
+async fn build_cas_manifest(
+    cas_root: &str,
+    manifest_dir: &Path,
+    timestamp: &str,
+    mode: Mode,
+) -> Result<CasManifestResult, String> {
+    let cas_path = Path::new(cas_root);
+    if !cas_path.exists() {
+        return Err(format!("CAS-katalog {cas_root} finnes ikke"));
+    }
+
+    // For inkrementell: finn tidsstempel for forrige manifest
+    let cutoff = if mode == Mode::Incremental {
+        find_last_manifest_time(manifest_dir).await
+    } else {
+        None
+    };
+
+    // Bruk find for å liste CAS-filer effektivt
+    let mut args = vec![cas_root.to_string(), "-type".to_string(), "f".to_string()];
+    if let Some(ref cutoff_time) = cutoff {
+        // Kun filer nyere enn cutoff
+        args.extend(["-newer".to_string(), cutoff_time.clone()]);
+    }
+
+    let output = tokio::process::Command::new("find")
+        .args(&args)
+        .stdout(std::process::Stdio::piped())
+        .stderr(std::process::Stdio::piped())
+        .output()
+        .await
+        .map_err(|e| format!("Kunne ikke liste CAS-filer: {e}"))?;
+
+    let file_list = String::from_utf8_lossy(&output.stdout);
+    let mut file_count: u64 = 0;
+    let mut total_size: u64 = 0;
+    let mut manifest_lines = Vec::new();
+
+    for line in file_list.lines() {
+        let line = line.trim();
+        if line.is_empty() {
+            continue;
+        }
+        let path = Path::new(line);
+        let size = match tokio::fs::metadata(path).await {
+            Ok(m) => m.len(),
+            Err(_) => continue,
+        };
+
+        // Extraher hash fra filnavn (siste path-komponent)
+        let hash = path
+            .file_name()
+            .map(|n| n.to_string_lossy().to_string())
+            .unwrap_or_default();
+
+        manifest_lines.push(format!("{hash}\t{size}"));
+        file_count += 1;
+        total_size += size;
+    }
+
+    let manifest_name = if mode == Mode::Incremental {
+        format!("cas_incremental_{timestamp}.tsv")
+    } else {
+        format!("cas_full_{timestamp}.tsv")
+    };
+    let manifest_path = manifest_dir.join(&manifest_name);
+
+    // Skriv manifest (TSV: hash\tsize)
+    let mode_label = mode_str(mode);
+    let header = format!("# CAS manifest — {mode_label} — {timestamp}\n# hash\tsize_bytes\n");
+    let content = format!("{}{}\n", header, manifest_lines.join("\n"));
+    tokio::fs::write(&manifest_path, &content)
+        .await
+        .map_err(|e| format!("Kunne ikke skrive CAS-manifest: {e}"))?;
+
+    Ok(CasManifestResult {
+        path: manifest_path.to_string_lossy().to_string(),
+        file_count,
+        total_size_bytes: total_size,
+        total_size_human: human_size(total_size),
+    })
+}
+
+/// Finn tidspunkt for forrige manifest (brukes som cutoff for inkrementell).
+async fn find_last_manifest_time(manifest_dir: &Path) -> Option<String> {
+    let mut entries = match tokio::fs::read_dir(manifest_dir).await {
+        Ok(e) => e,
+        Err(_) => return None,
+    };
+
+    let mut latest_path: Option<PathBuf> = None;
+    let mut latest_modified = std::time::SystemTime::UNIX_EPOCH;
+
+    while let Ok(Some(entry)) = entries.next_entry().await {
+        let name = entry.file_name().to_string_lossy().to_string();
+        if name.starts_with("cas_") && name.ends_with(".tsv") {
+            if let Ok(meta) = entry.metadata().await {
+                if let Ok(modified) = meta.modified() {
+                    if modified > latest_modified {
+                        latest_modified = modified;
+                        latest_path = Some(entry.path());
+                    }
+                }
+            }
+        }
+    }
+
+    latest_path.map(|p| p.to_string_lossy().to_string())
+}
+
+/// Slett PG-dumper eldre enn retain_days.
+async fn rotate_old_backups(pg_dir: &Path, retain_days: u32) -> Result<RotationResult, String> {
+    let cutoff = std::time::SystemTime::now()
+        - std::time::Duration::from_secs(retain_days as u64 * 86400);
+
+    let mut entries = tokio::fs::read_dir(pg_dir)
+        .await
+        .map_err(|e| format!("Kunne ikke lese backup-katalog: {e}"))?;
+
+    let mut deleted_count = 0u32;
+
+    while let Ok(Some(entry)) = entries.next_entry().await {
+        let name = entry.file_name().to_string_lossy().to_string();
+        if !name.ends_with(".dump") {
+            continue;
+        }
+        if let Ok(meta) = entry.metadata().await {
+            if let Ok(modified) = meta.modified() {
+                if modified < cutoff {
+                    if tokio::fs::remove_file(entry.path()).await.is_ok() {
+                        tracing::info!(file = %name, "Slettet gammel dump");
+                        deleted_count += 1;
+                    }
+                }
+            }
+        }
+    }
+
+    Ok(RotationResult {
+        deleted_count,
+        retain_days,
+    })
+}
+
+/// Formater bytes til lesbar størrelse.
+fn human_size(bytes: u64) -> String {
+    const UNITS: &[&str] = &["B", "KiB", "MiB", "GiB", "TiB"];
+    let mut size = bytes as f64;
+    for unit in UNITS {
+        if size < 1024.0 {
+            return format!("{size:.1} {unit}");
+        }
+        size /= 1024.0;
+    }
+    format!("{size:.1} PiB")
+}