From 54a516de8a49bdcdd2b95ecbfe2a365a860adaf4 Mon Sep 17 00:00:00 2001
From: Simon Rettberg
Date: Mon, 21 Jan 2019 14:49:54 +0100
Subject: [statistics] Log if client seems to have crashed

Log when we reset a client's state in the cron job, as well as if we
receive a poweron event even though the state in the DB is still IDLE or
OCCUPIED.
---
 modules-available/statistics/api.inc.php        | 17 +++++++++++++++++
 modules-available/statistics/hooks/cron.inc.php | 13 +++++++++++++
 2 files changed, 30 insertions(+)

diff --git a/modules-available/statistics/api.inc.php b/modules-available/statistics/api.inc.php
index 8f5e9fd0..674cc48d 100644
--- a/modules-available/statistics/api.inc.php
+++ b/modules-available/statistics/api.inc.php
@@ -161,6 +161,11 @@ if ($type{0} === '~') {
 		// Check for suspicious hardware changes
 		if ($old !== false) {
 			checkHardwareChange($old, $new);
+
+			// Log potential crash
+			if ($old['state'] === 'IDLE' || $old['state'] === 'OCCUPIED') {
+				writeClientLog('machine-mismatch-poweron', 'Client sent poweron event, but previous known state is ' . $old['state']);
+			}
 		}
 
 		// Write statistics data
@@ -403,6 +408,18 @@ function writeStatisticLog($type, $username, $data)
 	));
 }
 
+function writeClientLog($type, $description)
+{
+	global $ip, $uuid;
+	Database::exec('INSERT INTO clientlog (dateline, logtypeid, clientip, machineuuid, description, extra) VALUES (UNIX_TIMESTAMP(), :type, :client, :uuid, :description, :longdesc)', array(
+		'type'        => $type,
+		'client'      => $ip,
+		'description' => $description,
+		'longdesc'    => '',
+		'uuid'        => $uuid,
+	));
+}
+
 
 // For backwards compat, we require the . prefix
 if ($type{0} === '.') {
diff --git a/modules-available/statistics/hooks/cron.inc.php b/modules-available/statistics/hooks/cron.inc.php
index 4df7b0d4..f05762bc 100644
--- a/modules-available/statistics/hooks/cron.inc.php
+++ b/modules-available/statistics/hooks/cron.inc.php
@@ -23,6 +23,19 @@ function state_cleanup()
 	// Fix online state of machines that crashed
 	$standby = time() - 86400 * 2; // Reset standby machines after two days
 	$on = time() - 610; // Reset others after ~10 minutes
+	// Query for logging
+	$res = Database::simpleQuery("SELECT machineuuid, clientip, state FROM machine WHERE lastseen < If(state = 'STANDBY', $standby, $on) AND state <> 'OFFLINE'");
+	while ($row = $res->fetch(PDO::FETCH_ASSOC)) {
+		Database::exec('INSERT INTO clientlog (dateline, logtypeid, clientip, machineuuid, description, extra)
+					VALUES (UNIX_TIMESTAMP(), :type, :client, :uuid, :description, :longdesc)', array(
+			'type'        => 'machine-mismatch-cron',
+			'client'      => $row['clientip'],
+			'description' => 'Client timed out, last known state is ' . $row['state'],
+			'longdesc'    => '',
+			'uuid'        => $row['machineuuid'],
+		));
+	}
+	// Update -- yes this is not atomic. Should be sufficient for simple warnings though.
 	Database::exec("UPDATE machine SET state = 'OFFLINE' WHERE lastseen < If(state = 'STANDBY', $standby, $on) AND state <> 'OFFLINE'");
 }
 
-- 
cgit v1.2.3-55-g7522