Preserve verified health when provisioning completes

This commit is contained in:
2026-10-06 15:48:38 -04:00
parent 6999a33aef
commit 57d2bd11db
9 changed files with 256 additions and 41 deletions
+8 -6
View File
@@ -8,6 +8,7 @@ import type { ServerRow } from './servers.js';
import { parseGameEvent, type GameTelemetry } from '../../telemetry.js';
import { ONLINE_MAX_MS } from '../../server-freshness.js';
import { serverObservation } from './status.js';
import { recoverySettling } from '../provisioning/restarts.js';
const nativeFields = ['id','server','host_id','generation','library_id','image','bind_address','port','public_address','public_port','created_at','retired_at'];
const sql = `SELECT s.*, n.id AS native_instance_id,n.host_id,n.generation AS instance_generation,n.library_id AS game_library_id,
@@ -21,21 +22,21 @@ const sql = `SELECT s.*, n.id AS native_instance_id,n.host_id,n.generation AS in
(SELECT MAX(at) FROM (SELECT MAX(d.submitted_at) AS at FROM server_deployments d
WHERE d.server=s.name AND d.server_deleted_at IS NULL AND d.action IN ('create','start','restart','rebuild')
UNION ALL SELECT r.requested_at)) AS submission,
r.requested_at AS restart_at,r.healthy_at AS restart_healthy,t.received_at AS telemetry_at,t.payload AS telemetry_json
r.requested_at AS restart_at,r.healthy_at AS restart_healthy,r.operation_id AS restart_operation_id,t.received_at AS telemetry_at,t.payload AS telemetry_json
FROM servers s LEFT JOIN native_instances n ON n.server=s.name AND n.retired_at IS NULL
LEFT JOIN ec_hosts h ON h.id=n.host_id AND h.deleted_at IS NULL
LEFT JOIN restart_rollout r ON r.server=s.name LEFT JOIN server_telemetry t ON t.server=s.name`;
interface Row extends ServerRow {
instance_json: string | null; host_inventory: string | null; host_heartbeat: string | null; credential_active: number;
lifecycle_json: string | null; lifecycle_at: string | null; submission: string | null;
restart_at: string | null; restart_healthy: string | null; telemetry_at: string | null; telemetry_json: string | null;
restart_at: string | null; restart_healthy: string | null; restart_operation_id: string | null; telemetry_at: string | null; telemetry_json: string | null;
}
interface Record {
server: ServerRow; native: NativeInstance | null;
host: Pick<Host, 'connected' | 'connectionState' | 'inventory'> | null;
lifecycle: { id: string; action: string; state: string; submitted_at: string | null } | null;
lifecycleAt: string | null; submission: string | null; telemetry: GameTelemetry | null;
restartAt: string | null; restartHealthy: string | null;
restartAt: string | null; restartHealthy: string | null; restartOperationId: string | null;
}
/** A pass shares its reads; each asynchronous evidence boundary refreshes in one joined query. */
@@ -48,7 +49,7 @@ export class MonitoringSnapshot {
if (server === undefined) this.records.clear(); else this.records.delete(server);
for (const row of rows) {
const { instance_json,host_inventory,host_heartbeat,credential_active,lifecycle_json,lifecycle_at,submission,
restart_at,restart_healthy,telemetry_at,telemetry_json,...fields } = row;
restart_at,restart_healthy,restart_operation_id,telemetry_at,telemetry_json,...fields } = row;
const connected = !!credential_active && !!host_heartbeat && Date.now() - Date.parse(host_heartbeat) <= HEARTBEAT_MAX_AGE_MS;
let telemetry: GameTelemetry | null = null;
if (telemetry_json) try {
@@ -60,7 +61,7 @@ export class MonitoringSnapshot {
host: host_inventory ? { connected, connectionState: connected ? 'online' : 'offline',
inventory: parseHostInventory(row.host_id!, host_inventory) } : null,
lifecycle: lifecycle_json ? JSON.parse(lifecycle_json) : null, lifecycleAt: lifecycle_at, submission, telemetry,
restartAt: restart_at, restartHealthy: restart_healthy });
restartAt: restart_at, restartHealthy: restart_healthy, restartOperationId: restart_operation_id });
}
}
get(server: string) { return this.records.get(server); }
@@ -71,7 +72,8 @@ export class MonitoringSnapshot {
}
settling(server: string) {
const record = this.get(server);
return !!record?.restartAt && !record.restartHealthy && Date.now() - Date.parse(record.restartAt) < 15_000;
return recoverySettling(record?.restartAt ? { requested_at: record.restartAt,
healthy_at: record.restartHealthy, operation_id: record.restartOperationId } : undefined, record?.lifecycle);
}
telemetryEvidence(server: string) {
const record = this.get(server), report = record?.telemetry;
+7 -3
View File
@@ -1,7 +1,7 @@
import { db } from '../db/index.js';
import { parseGameEvent, type GameTelemetry } from '../../telemetry.js';
import { ONLINE_MAX_MS } from '../../server-freshness.js';
import { rconGetPlayerCount, rconStatusResult } from '../integrations/rcon.js';
import { rconGetPlayerCount, rconStatusResult, type RconStatusResult } from '../integrations/rcon.js';
import { latestGameSubmission } from './lifecycle-evidence.js';
export { latestGameSubmission } from './lifecycle-evidence.js';
@@ -59,9 +59,13 @@ export async function probePlayerEvidence(server: string): Promise<PlayerEvidenc
return { players: null, source: null, observedAt: new Date().toISOString() };
}
export type VerifiedGameStatus = { ok: true; players: number; map: string | null; observedAt: string } & (
{ source: 'rcon'; rcon: Extract<RconStatusResult, { ok: true }> }
| { source: 'telemetry'; receivedAt: string });
/** Host agent readiness remains a caller prerequisite. Old-process telemetry is excluded. */
export async function probeServerStatus(server: string, after?: string | null): Promise<
{ ok: true; players: number; map: string | null; source: 'rcon' | 'telemetry'; observedAt: string; receivedAt?: string } | { ok: false; reason: string; observedAt: string }
VerifiedGameStatus | { ok: false; reason: string; observedAt: string }
> {
const current = telemetryEvidence(server, after);
if (current) return { ok: true, ...current, source: 'telemetry', observedAt: new Date().toISOString() };
@@ -70,7 +74,7 @@ export async function probeServerStatus(server: string, after?: string | null):
const rcon = await rconStatusResult(server);
const latest = telemetryEvidence(server, after);
if (latest) return { ok: true, ...latest, source: 'telemetry', observedAt: new Date().toISOString() };
if (rcon.ok) return { ok: true, players: rcon.status.players, map: rcon.status.map, source: 'rcon', observedAt: new Date().toISOString() };
if (rcon.ok) return { ok: true, players: rcon.status.players, map: rcon.status.map, source: 'rcon', rcon, observedAt: new Date().toISOString() };
reason = rcon.reason;
} catch (err) { reason = (err as Error).message; }
const telemetry = telemetryEvidence(server, after);
+22
View File
@@ -0,0 +1,22 @@
import type { RconStatusResult } from '../integrations/rcon.js';
import type { ServerObservation } from '../../server-observation.js';
import type { ServerRow } from './servers.js';
/** Monitoring and provisioning publish the same complete RCON observation. */
export function successfulStatusFields(result: Extract<RconStatusResult, { ok: true }>, observedAt: string): Partial<ServerRow> {
return {
last_checked_at: observedAt, last_seen_at: observedAt,
last_players: result.status.players, last_map: result.status.map,
last_gametype: result.status.gametype, last_roster: JSON.stringify(result.status.roster),
last_reply_snippet: result.rawReply ?? null, last_error: null, rcon_failures: 0
};
}
/** Occupancy timers and match clearing follow the same source as the UI. */
export function activityFields(current: ServerRow, status: ServerObservation): Partial<ServerRow> {
if (status.health !== 'online' || status.last_players === null) return {};
return status.last_players > 0
? { active_since: current.active_since ?? status.last_seen_at, inactive_since: null }
: { active_since: null, inactive_since: current.inactive_since ?? status.last_seen_at,
game_started_at: null, game_ended_at: null };
}
+6 -6
View File
@@ -1,7 +1,7 @@
import { rconHealthError, selectServerObservation, type GameObservation } from '../../server-observation.js';
import { classifyHealth } from './health.js';
import { isProvisioning } from './provisioning.js';
import { restartSettling } from '../provisioning/restarts.js';
import { restartSettling, recoverySettling } from '../provisioning/restarts.js';
import { type ServerRow } from './servers.js';
import { telemetryReport, latestGameSubmission } from './status-evidence.js';
import { latestGameSubmissions } from './lifecycle-evidence.js';
@@ -13,11 +13,11 @@ interface Evidence { provisioning: boolean; settling: boolean; telemetry: GameTe
/** One set of evidence reads per snapshot, rather than four queries per server. */
export function observationReader() {
const handle = db();
const operations = new Map((handle.prepare(`SELECT server, action, state FROM server_deployments
const operations = new Map((handle.prepare(`SELECT id, server, action, state FROM server_deployments
WHERE rowid IN (SELECT MAX(rowid) FROM server_deployments WHERE server_deleted_at IS NULL GROUP BY server)`).all() as
{ server: string; action: string; state: string }[]).map(row => [row.server, row]));
const restarts = new Map((handle.prepare('SELECT server, requested_at, healthy_at FROM restart_rollout').all() as
{ server: string; requested_at: string; healthy_at: string | null }[]).map(row => [row.server, row]));
{ id: string; server: string; action: string; state: string }[]).map(row => [row.server, row]));
const restarts = new Map((handle.prepare('SELECT server, requested_at, healthy_at, operation_id FROM restart_rollout').all() as
{ server: string; requested_at: string; healthy_at: string | null; operation_id: string | null }[]).map(row => [row.server, row]));
const submissions = latestGameSubmissions();
const reports = new Map<string, GameTelemetry>();
for (const row of handle.prepare('SELECT server, received_at, payload FROM server_telemetry WHERE payload IS NOT NULL').all() as
@@ -32,7 +32,7 @@ export function observationReader() {
const operation = operations.get(server.name), restart = restarts.get(server.name);
return serverObservation(server, {
provisioning: !!operation && ['create','rebuild','start'].includes(operation.action) && ['queued','running','awaiting-health'].includes(operation.state),
settling: !!(restart && !restart.healthy_at && Date.now() - Date.parse(restart.requested_at) < 15_000),
settling: recoverySettling(restart, operation),
telemetry: reports.get(server.name) ?? null, submission: submissions.get(server.name) ?? null
});
};
+3 -19
View File
@@ -9,6 +9,7 @@ import type { RconFailure } from '../integrations/rcon.js';
import { setServerFields, type ServerRow } from '../fleet/servers.js';
import { inspectForMonitoring, rconForMonitoring, prepareMonitoringRcon, monitoringStatus, monitoringStatusWriter, monitoringIdentity, MonitoringChangedError, MonitoringSkippedError } from '../fleet/health-observations.js';
import { serverObservation } from '../fleet/status.js';
import { successfulStatusFields, activityFields } from '../fleet/status-updates.js';
import { STATUS_REFRESH_MS } from '../../server-freshness.js';
import { recordMetrics } from '../fleet/metrics.js';
import { selectedZombieMap } from '../../map-aliases.js';
@@ -220,13 +221,7 @@ async function refreshServerStatus(server: string, snapshot: MonitoringSnapshot,
updates.last_reply_snippet = result.rawReply ?? null;
if (result.ok) {
capacity = result.status.capacity ?? null;
updates.last_players = result.status.players;
updates.last_map = result.status.map;
updates.last_gametype = result.status.gametype;
updates.last_roster = JSON.stringify(result.status.roster);
updates.last_seen_at = now;
updates.last_error = null;
updates.rcon_failures = 0;
Object.assign(updates, successfulStatusFields(result, now));
} else {
// Only distinct accepted checks in the current lifecycle count.
// Skipped/superseded checks return above without changing the streak.
@@ -247,18 +242,7 @@ async function refreshServerStatus(server: string, snapshot: MonitoringSnapshot,
}
const current = { ...(snapshot.get(srv.name)?.server ?? srv), ...updates } as ServerRow;
const status = snapshot.observe(current);
// Occupancy timers and match clearing follow the same source as the UI.
if (status.health === 'online' && status.last_players !== null) {
if (status.last_players > 0) {
updates.active_since = current.active_since ?? status.last_seen_at;
updates.inactive_since = null;
} else {
updates.active_since = null;
updates.inactive_since = current.inactive_since ?? status.last_seen_at;
updates.game_started_at = null;
updates.game_ended_at = null;
}
}
Object.assign(updates, activityFields(current, status));
flush();
return capacity;
}
@@ -0,0 +1,37 @@
import { ONLINE_MAX_MS } from '../../server-freshness.js';
import { monitoringIdentity } from '../fleet/health-observations.js';
import { getServer, setServerFields, type ServerRow } from '../fleet/servers.js';
import { latestGameSubmission, telemetryReport, type VerifiedGameStatus } from '../fleet/status-evidence.js';
import { serverObservation } from '../fleet/status.js';
import { activityFields, successfulStatusFields } from '../fleet/status-updates.js';
/** Called inside the completion transaction; evidence must still describe this process. */
export function handoffVerifiedHealth(server: string, identity: string,
container: { state: string; observedAt: string }, game: VerifiedGameStatus): boolean {
const current = getServer(server);
if (!current || current.desired_state !== 'running' || monitoringIdentity(server) !== identity) return false;
const fresh = (at: string) => {
const age = Date.now() - Date.parse(at);
return age >= 0 && age <= ONLINE_MAX_MS;
};
const gameAt = game.source === 'telemetry' ? game.receivedAt : game.observedAt;
const submission = latestGameSubmission(server);
if (!fresh(container.observedAt) || !fresh(gameAt) || (submission && gameAt <= submission)) return false;
const updates: Partial<ServerRow> = {};
if (!current.container_checked_at || current.container_checked_at <= container.observedAt) {
updates.container_state = container.state;
updates.container_checked_at = container.observedAt;
}
if (game.source === 'rcon' && (!current.last_checked_at || current.last_checked_at <= game.observedAt)) {
Object.assign(updates, successfulStatusFields(game.rcon, game.observedAt));
}
const candidate = { ...current, ...updates };
// Evaluate the health readers will see after the operation and recovery slot
// complete. Newer contradictory evidence must keep verification pending.
const status = serverObservation(candidate, { provisioning: false, settling: false,
submission, telemetry: telemetryReport(server) });
if (status.health !== 'online') return false;
setServerFields(server, { ...updates, ...activityFields(candidate, status) });
return true;
}
+13 -3
View File
@@ -23,6 +23,8 @@ import { trackActivity } from '../pipeline/activity.js';
import { fleetSettings } from '../settings.js';
import { resolveRecoveredHealthOperation } from '../maintenance/recovery.js';
import { probeServerStatus, probePlayerEvidence } from '../fleet/status-evidence.js';
import { monitoringIdentity } from '../fleet/health-observations.js';
import { handoffVerifiedHealth } from './health-handoff.js';
import { checkRelease, freshRelease, probeVersion } from '../integrations/plutonium.js';
import { recordScript, removeScript } from '../fleet/inventory.js';
import { sha256Buffer } from '../fleet/files.js';
@@ -544,7 +546,9 @@ async function processJob(job: Job, eligibility: (job: Job) => string | null) {
recordProgress(job.id, { checks, ...(value.result === 'pass' ? { checkpoint: { key: phase, label: value.label } } : {}) });
};
deploymentStage(job.id, 'Checking container state with host agent', { phase: job.action === 'stop' ? 'verify-stopped' : 'verify-container', blocker: null, checks });
const verificationIdentity = monitoringIdentity(job.server);
const health = await nativeRequest<{ healthy: boolean; state: string }>(instance, 'instance.inspect');
const containerObservedAt = new Date().toISOString();
if (!stillActive(job)) return;
if (job.action === 'stop') {
const stopped = ['stopped','exited','created','dead'].includes(health.state);
@@ -584,9 +588,15 @@ async function processJob(job: Job, eligibility: (job: Job) => string | null) {
if (revision === null || target === null) { hold(job, 'Waiting to verify the Plutonium runtime revision.'); return; }
if (revision < target) { finishDeployment(job.id, { ok: false, message: `Running Plutonium r${revision} is older than production r${target}.` }); return; }
}
db().prepare('UPDATE servers SET container_state=?,container_checked_at=? WHERE name=?').run(health.state, new Date().toISOString(), job.server);
recordProgress(job.id, { blocker: null, checks });
complete(job); return;
const completed = db().transaction(() => {
if (!stillActive(job) || !handoffVerifiedHealth(job.server, verificationIdentity,
{ state: health.state, observedAt: containerObservedAt }, status)) return false;
recordProgress(job.id, { blocker: null, checks });
complete(job);
return true;
})();
if (!completed) hold(job, 'Waiting for current health evidence before completing verification.');
return;
}
const reason = eligibility(job);
if (reason) { hold(job, reason); return; }
+16 -3
View File
@@ -23,11 +23,24 @@ export function restartRecoveryPending(server: string): boolean {
return slotFor(server)?.healthy_at === null;
}
/** The queued restart may still expose the previous healthy container. */
type SettlingSlot = Pick<RestartSlot, 'requested_at' | 'healthy_at' | 'operation_id'>;
interface SettlingOperation { id: string; action: string; state: string }
/** Submitted lifecycle work stays Starting until verification ends, including slow restarts. */
export function recoverySettling(slot: SettlingSlot | undefined, operation?: SettlingOperation | null): boolean {
if (!slot || slot.healthy_at) return false;
if (slot.operation_id) return operation?.id === slot.operation_id
&& ['start', 'restart', 'rebuild'].includes(operation.action)
&& ['running', 'awaiting-health'].includes(operation.state);
// Legacy recovery slots have no operation whose verification can be followed.
return Date.now() - Date.parse(slot.requested_at) < HEALTH_POLL_MS;
}
export function restartSettling(server: string): boolean {
const slot = slotFor(server);
return !!(slot && !slot.healthy_at
&& Date.now() - Date.parse(slot.requested_at) < HEALTH_POLL_MS);
const operation = slot?.operation_id ? db().prepare(`SELECT id, action, state FROM server_deployments
WHERE server = ? AND server_deleted_at IS NULL ORDER BY rowid DESC LIMIT 1`).get(server) as SettlingOperation | undefined : undefined;
return recoverySettling(slot, operation);
}
export function rolloutWaitReason(server: string, operationId = '', action: RuntimeAction = 'restart'): string | null {