Files
SystemsMonitor/PegasusMonitor/ServerMonitor.swift
T
2026-10-21 14:54:07 +01:00

352 lines
17 KiB
Swift

import Foundation
import SwiftUI
import UserNotifications
enum HealthStatus: String, Codable {
case ok, warn, fail
var title: String { switch self { case .ok: "Healthy"; case .warn: "Warning"; case .fail: "Critical" } }
var symbol: String { switch self { case .ok: "checkmark.circle.fill"; case .warn: "exclamationmark.triangle.fill"; case .fail: "xmark.circle.fill" } }
var color: Color { switch self { case .ok: .green; case .warn: .orange; case .fail: .red } }
}
struct HealthCounts: Codable { let ok: Int; let warn: Int; let fail: Int }
struct HealthGroup: Codable, Identifiable { let name: String; let status: HealthStatus; let missing: [String]; var id: String { name } }
struct HealthCheck: Codable, Identifiable { let name: String; let status: HealthStatus; let detail: String; let section: String; var id: String { "\(section)/\(name)" } }
struct HealthReport: Codable {
let schema: Int
let host: String?
let generatedAt: String?
let uptimeSeconds: Int?
var uptimeLabel: String? {
guard let uptimeSeconds, uptimeSeconds >= 0 else { return nil }
let days = uptimeSeconds / 86400
let hours = (uptimeSeconds % 86400) / 3600
let minutes = (uptimeSeconds % 3600) / 60
if days > 0 { return "\(days)d \(hours)h \(minutes)m" }
if hours > 0 { return "\(hours)h \(minutes)m" }
return "\(minutes)m"
}
let overall: HealthStatus
let counts: HealthCounts
let groups: [HealthGroup]
let checks: [HealthCheck]
let diagnostics: String?
let services: [DiscoveredService]?
}
// Service discovery is optional so legacy Pegasus health JSON stays compatible.
struct DiscoveredService: Codable, Identifiable, Equatable {
let name: String
let state: String
let detail: String
var id: String { name }
var isInstalled: Bool { state != "not_installed" }
var isHealthy: Bool { state == "running" }
var symbol: String {
switch state {
case "running": "checkmark.circle.fill"
case "failed", "required_missing": "xmark.circle.fill"
case "stopped": "exclamationmark.triangle.fill"
case "disabled": "pause.circle"
case "not_installed": "minus.circle"
default: "questionmark.circle"
}
}
var color: Color {
switch state {
case "running": .green
case "failed", "required_missing": .red
case "stopped": .orange
default: .secondary
}
}
var title: String {
switch state {
case "running": "Running"
case "failed": "Failed"
case "stopped": "Stopped"
case "disabled": "Disabled"
case "not_installed": "Not installed"
case "required_missing": "Required — missing"
default: "Unknown"
}
}
}
struct ServerProfile: Identifiable, Codable, Equatable {
var id: UUID = UUID()
var name: String = "New Server"
var host: String = ""
var user: String = ""
var port: String = "22"
var keyFile: String = ""
var remoteCommand: String = "/usr/local/bin/pegasus-monitor-health-json"
var enabled: Bool = true
var requiredServices: [String] = []
var discoverServices: Bool = true
// Preserve profiles saved by 1.1, which did not include service preferences.
enum CodingKeys: String, CodingKey {
case id, name, host, user, port, keyFile, remoteCommand, enabled, requiredServices, discoverServices
}
init(id: UUID = UUID(), name: String = "New Server", host: String = "", user: String = "", port: String = "22",
keyFile: String = "", remoteCommand: String = "/usr/local/bin/pegasus-monitor-health-json",
enabled: Bool = true, requiredServices: [String] = [], discoverServices: Bool = true) {
self.id = id; self.name = name; self.host = host; self.user = user; self.port = port
self.keyFile = keyFile; self.remoteCommand = remoteCommand; self.enabled = enabled
self.requiredServices = requiredServices; self.discoverServices = discoverServices
}
init(from decoder: Decoder) throws {
let c = try decoder.container(keyedBy: CodingKeys.self)
id = try c.decode(UUID.self, forKey: .id)
name = try c.decode(String.self, forKey: .name)
host = try c.decode(String.self, forKey: .host)
user = try c.decode(String.self, forKey: .user)
port = try c.decode(String.self, forKey: .port)
keyFile = try c.decode(String.self, forKey: .keyFile)
remoteCommand = try c.decode(String.self, forKey: .remoteCommand)
enabled = try c.decode(Bool.self, forKey: .enabled)
requiredServices = try c.decodeIfPresent([String].self, forKey: .requiredServices) ?? []
discoverServices = try c.decodeIfPresent(Bool.self, forKey: .discoverServices) ?? true
}
static func legacy() -> ServerProfile {
let defaults = UserDefaults.standard
return ServerProfile(name: "Pegasus",
host: defaults.string(forKey: "server.host") ?? "192.168.1.2",
user: defaults.string(forKey: "server.user") ?? "mag",
port: defaults.string(forKey: "server.port") ?? "22",
keyFile: defaults.string(forKey: "server.keyFile") ?? "",
remoteCommand: "sudo -n /usr/local/bin/pegasus-health-json")
}
}
struct ServerResult {
var report: HealthReport?
var error: String?
var lastChecked: Date?
var refreshing = false
var status: HealthStatus { error != nil ? .fail : (report?.overall ?? .warn) }
}
@MainActor
final class ServerMonitor: ObservableObject {
@Published private(set) var servers: [ServerProfile] = []
@Published private(set) var results: [UUID: ServerResult] = [:]
@Published var selectedID: UUID?
@Published var intervalMinutes: Int { didSet { UserDefaults.standard.set(intervalMinutes, forKey: "monitor.intervalMinutes"); configureTimer() } }
@Published var notifications: Bool { didSet { UserDefaults.standard.set(notifications, forKey: "monitor.notifications") } }
private var timer: Timer?
private var visited = Set<UUID>()
private let profilesKey = "monitor.serverProfiles.v1"
private let selectedKey = "monitor.selectedServer.v1"
init() {
let defaults = UserDefaults.standard
intervalMinutes = max(1, defaults.object(forKey: "monitor.intervalMinutes") as? Int ?? 5)
notifications = defaults.object(forKey: "monitor.notifications") as? Bool ?? true
if let bytes = defaults.data(forKey: profilesKey), let loaded = try? JSONDecoder().decode([ServerProfile].self, from: bytes) {
servers = loaded
} else {
servers = [.legacy()]
}
if let stored = defaults.string(forKey: selectedKey), let id = UUID(uuidString: stored), servers.contains(where: { $0.id == id }) {
selectedID = id
} else {
selectedID = servers.first?.id
}
configureTimer()
Task { await refreshAll() }
}
var selected: ServerProfile? { servers.first(where: { $0.id == selectedID }) }
var selectedResult: ServerResult? { selectedID.flatMap { results[$0] } }
var report: HealthReport? { selectedResult?.report }
var selectedServices: [DiscoveredService] {
guard let profile = selected else { return [] }
return discoveredServices(for: profile.id).filter { profile.requiredServices.contains($0.name) || (profile.discoverServices && $0.state != "not_installed") }
}
func discoveredServices(for id: UUID) -> [DiscoveredService] {
guard let profile = servers.first(where: { $0.id == id }) else { return [] }
let discovered = results[id]?.report?.services ?? []
var byName = Dictionary(uniqueKeysWithValues: discovered.map { ($0.name, $0) })
for name in profile.requiredServices where byName[name] == nil {
byName[name] = DiscoveredService(name: name, state: "not_installed", detail: "Not reported by agent")
}
return byName.values.sorted { $0.name.localizedCaseInsensitiveCompare($1.name) == .orderedAscending }
}
func serviceStatus(_ item: DiscoveredService, required: Bool) -> HealthStatus? {
if required && !item.isHealthy { return .fail }
if item.state == "failed" { return .fail }
if item.state == "stopped" { return .warn }
return nil
}
func applyServicePolicy(_ report: HealthReport, for profile: ServerProfile) -> HealthReport {
let services = report.services ?? []
// If an older agent has no service discovery, do not invent a healthy state.
// Explicit required services with no reported data are a failure.
let known = Dictionary(uniqueKeysWithValues: services.map { ($0.name, $0) })
var extra: [HealthCheck] = []
for required in Set(profile.requiredServices) {
let item = known[required]
if item?.isHealthy != true {
let description = item?.title ?? "Not reported by monitoring agent"
extra.append(HealthCheck(name: "Required: \(required)", status: .fail, detail: description, section: "Required services"))
}
}
let existing = Set(report.checks.map { $0.id })
extra.removeAll { existing.contains($0.id) }
guard !extra.isEmpty else { return report }
let checks = report.checks + extra
let counts = HealthCounts(ok: report.counts.ok, warn: report.counts.warn, fail: report.counts.fail + extra.count)
let groups = report.groups + [HealthGroup(name: "Required services", status: .fail, missing: extra.map(\.name))]
return HealthReport(schema: report.schema, host: report.host, generatedAt: report.generatedAt, uptimeSeconds: report.uptimeSeconds, overall: .fail,
counts: counts, groups: groups, checks: checks, diagnostics: report.diagnostics, services: report.services)
}
var status: HealthStatus {
let statuses = servers.filter(\.enabled).map { results[$0.id]?.status ?? .warn }
if statuses.contains(.fail) { return .fail }
if statuses.contains(.warn) { return .warn }
return statuses.isEmpty ? .warn : .ok
}
var aggregate: HealthCounts {
let active = servers.filter(\.enabled)
return HealthCounts(ok: active.reduce(0) { $0 + (results[$1.id]?.report?.counts.ok ?? 0) },
warn: active.reduce(0) { $0 + (results[$1.id]?.report?.counts.warn ?? 0) },
fail: active.reduce(0) { $0 + (results[$1.id]?.report?.counts.fail ?? (results[$1.id]?.error == nil ? 0 : 1)) })
}
var offlineCount: Int { servers.filter { $0.enabled && results[$0.id]?.error != nil }.count }
func select(_ id: UUID) {
guard servers.contains(where: { $0.id == id }) else { return }
selectedID = id
UserDefaults.standard.set(id.uuidString, forKey: selectedKey)
}
func save(_ profile: ServerProfile) {
if let index = servers.firstIndex(where: { $0.id == profile.id }) { servers[index] = profile }
else { servers.append(profile) }
results.removeValue(forKey: profile.id)
persist()
select(profile.id)
Task { await refresh(id: profile.id) }
}
func remove(_ id: UUID) {
servers.removeAll { $0.id == id }
results.removeValue(forKey: id)
visited.remove(id)
if selectedID == id { selectedID = servers.first?.id }
persist()
}
func move(from source: IndexSet, to destination: Int) {
servers.move(fromOffsets: source, toOffset: destination)
persist()
}
private func persist() {
if let bytes = try? JSONEncoder().encode(servers) { UserDefaults.standard.set(bytes, forKey: profilesKey) }
UserDefaults.standard.set(selectedID?.uuidString, forKey: selectedKey)
}
func configureTimer() {
timer?.invalidate()
timer = Timer.scheduledTimer(withTimeInterval: TimeInterval(max(1, intervalMinutes) * 60), repeats: true) { [weak self] _ in
Task { @MainActor [weak self] in await self?.refreshAll() }
}
}
func refreshAll() async {
await withTaskGroup(of: Void.self) { group in
for server in servers where server.enabled {
group.addTask { await self.refresh(id: server.id) }
}
}
}
func refreshSelected() async {
if let id = selectedID { await refresh(id: id) }
}
func refresh(id: UUID) async {
guard let server = servers.first(where: { $0.id == id }), server.enabled else { return }
if results[id]?.refreshing == true { return }
var state = results[id] ?? ServerResult()
state.refreshing = true
results[id] = state
let previous = state.status
do {
let data = try await SSHRunner.run(host: server.host, user: server.user, port: server.port,
keyFile: server.keyFile, remoteCommand: server.remoteCommand)
let decoded = try JSONDecoder().decode(HealthReport.self, from: data)
let report = applyServicePolicy(decoded, for: server)
guard report.schema == 1 else { throw SSHError.invalidResponse }
state.report = report
state.error = nil
if notifications && visited.contains(id) && previous == .ok && report.overall != .ok {
notify(title: "\(server.name): \(report.overall.title)", message: "\(report.counts.fail) failures, \(report.counts.warn) warnings")
}
} catch {
state.error = error.localizedDescription
if notifications && visited.contains(id) && previous != .fail {
notify(title: "\(server.name) unreachable", message: error.localizedDescription)
}
}
state.lastChecked = Date()
state.refreshing = false
results[id] = state
visited.insert(id)
}
private func notify(title: String, message: String) {
let center = UNUserNotificationCenter.current()
center.requestAuthorization(options: [.alert, .sound]) { allowed, _ in
guard allowed else { return }
let content = UNMutableNotificationContent()
content.title = title
content.body = message
center.add(UNNotificationRequest(identifier: UUID().uuidString, content: content, trigger: nil))
}
}
}
enum SSHError: LocalizedError {
case invalidSettings, invalidResponse, process(String)
var errorDescription: String? {
switch self {
case .invalidSettings: "Invalid SSH settings (host, user, port or command)."
case .invalidResponse: "Invalid health JSON. Install the remote health adapter."
case .process(let text): text
}
}
}
enum SSHRunner {
static func run(host: String, user: String, port: String, keyFile: String, remoteCommand: String) async throws -> Data {
guard !host.isEmpty, !user.isEmpty, let number = Int(port), (1...65535).contains(number),
host.range(of: #"^[A-Za-z0-9._:-]+$"#, options: .regularExpression) != nil,
user.range(of: #"^[A-Za-z0-9_-]+$"#, options: .regularExpression) != nil,
!remoteCommand.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty else { throw SSHError.invalidSettings }
return try await Task.detached(priority: .utility) {
let process = Process()
process.executableURL = URL(fileURLWithPath: "/usr/bin/ssh")
var args = ["-T", "-o", "BatchMode=yes", "-o", "StrictHostKeyChecking=yes", "-o", "ConnectTimeout=8", "-o", "ServerAliveInterval=5", "-o", "ServerAliveCountMax=2", "-p", String(number)]
if !keyFile.isEmpty { args += ["-i", (keyFile as NSString).expandingTildeInPath] }
args += ["\(user)@\(host)", remoteCommand]
process.arguments = args
let output = Pipe()
let errors = Pipe()
process.standardOutput = output
process.standardError = errors
try process.run()
let stdout = Task.detached { output.fileHandleForReading.readDataToEndOfFile() }
let stderr = Task.detached { errors.fileHandleForReading.readDataToEndOfFile() }
let timeout = Task.detached {
try? await Task.sleep(for: .seconds(90))
if process.isRunning { process.terminate() }
}
let bytes = await stdout.value
let errorBytes = await stderr.value
process.waitUntilExit()
timeout.cancel()
if bytes.isEmpty {
let message = String(data: errorBytes, encoding: .utf8) ?? "SSH returned no data"
throw SSHError.process(message.isEmpty ? "SSH returned no data" : message)
}
return bytes
}.value
}
}