watchdogs

svn path=/trunk/boinc/; revision=990

client/client_types.C

@@ -54,7 +54,8 @@ PROJECT::PROJECT() {
     master_fetch_failures = 0;
     resource_debt = 0;
     debt_order = 0;
-    master_url_fetch_pending = 0;
+    master_url_fetch_pending = false;
+    sched_rpc_pending = false;
 }
 
 PROJECT::~PROJECT() {

client/cs_scheduler.C

@@ -128,8 +128,7 @@ PROJECT* CLIENT_STATE::next_project(PROJECT* old) {
     pbest = 0;
     for (i=0; i<projects.size(); i++) {
         p = projects[i];
-    if (p->master_url_fetch_pending)
-        continue;
+        if (p->master_url_fetch_pending) continue;
         if (p->min_rpc_time > now ) continue;
         if (old && p->debt_order <= old->debt_order) continue;
         if (p->debt_order < best) {

client/scheduler_op.C

@@ -309,6 +309,7 @@ bool SCHEDULER_OP::poll() {
         if (http_op.http_op_state == HTTP_STATE_DONE) {
             action = true;
             project->master_url_fetch_pending = false;
+            gstate.set_client_state_dirty("master URL fetch done");
             http_ops->remove(&http_op);
             if (http_op.http_op_retval == 0) {
                 if (log_flags.sched_op_debug) {

sched/handle_request.C

@@ -45,7 +45,23 @@ using namespace std;
 // return true if the WU can be executed on the host
 //
 bool wu_is_feasible(WORKUNIT& wu, HOST& host) {
-    return ((wu.rsc_disk <= host.d_free) && (wu.rsc_memory <= host.m_nbytes));
+    char buf[256];
+
+    if(host.d_free && wu.rsc_disk > host.d_free) {
+        sprintf(buf, "WU %d needs %f disk; host %d has %f\n",
+            wu.id, wu.rsc_disk, host.id, host.d_free
+        );
+        write_log(buf);
+        return false;
+    }
+    if (host.m_nbytes && wu.rsc_memory > host.m_nbytes) {
+        sprintf(buf, "WU %d needs %f mem; host %d has %f\n",
+            wu.id, wu.rsc_memory, host.id, host.m_nbytes
+        );
+        write_log(buf);
+        return false;
+    }
+    return true;
 }
 
 // estimate the time that a WU will take on a host
@@ -177,7 +193,8 @@ int authenticate_user(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
             );
             strcpy(reply.message_priority, "low");
             reply.request_delay = 120;
-            write_log("bad authenticator\n");
+            sprintf(buf, "Bad authenticator: %s\n", sreq.authenticator);
+            write_log(buf);
             return -1;
         }
 
@@ -202,7 +219,8 @@ int authenticate_user(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
                 "Visit this project's web site to get an authenticator.");
             strcpy(reply.message_priority, "low");
             reply.request_delay = 120;
-            write_log("bad authenticator\n");
+            sprintf(buf, "Bad authenticator: %s\n", sreq.authenticator);
+            write_log(buf);
             return -1;
         }
 new_host:

sched/wd.php

@@ -0,0 +1,40 @@
+#! /usr/local/bin/php
+<?php
+
+// General-purpose watchdog script.
+// Run this from crontab.
+// We use the mod time of a file "watchdog_exec_time"
+// to keep track of the last time we ran.
+
+// BOINC uses a number of "error log files".
+// If any error log file has been updated since the last time we ran,
+// sound the alarm.
+
+// TODO: to detect file system full errors,
+// have this program attempt to create/read a file.
+
+function sound_alarm($x) {
+    echo "alarm: $x\n";
+    //email("davea@ssl.berkeley.edu", "BOINC problem", $x);
+}
+
+function check_log_file($file, $last_time) {
+    $t = filemtime($file);
+    if ($t == false) {
+        sound_alarm("log file ".$file." missing");
+    } else if ($t > $last_time) {
+        $lines = file($file);
+        $last_line = $lines[count($lines)-1];
+        sound_alarm($last_line);
+    }
+}
+
+    $last_time = filemtime("watchdog_exec_time");
+    if (!$last_time) {
+        alarm("foo");
+    }
+    touch("watchdog_exec_time");
+
+    check_log_file("error_log", $last_time);
+
+?>

sched/wd_nresults_changing.php

@@ -0,0 +1,65 @@
+#! /usr/local/bin/php
+<?php
+
+// watchdog script to ensure that the number of result records
+// increases at least every X seconds (X = crontab period)
+
+include_once("util.inc");
+
+function fail($x) {
+    $f = fopen("error_log", "a");
+    if (!$f) return false;
+    fputs($f, "[".strftime("%T %D")."] ");
+    fputs($f, $x);
+    fclose($f);
+    exit();
+}
+
+function read_count_file() {
+    if (!file_exists("nresults")) {
+        return false;
+    }
+    $lines = file("nresults");
+    if (!$lines) return false;
+    return $lines[0];
+}
+
+function write_count_file($n) {
+    $f = fopen("nresults", "w");
+    if (!$f) return false;
+    $x = sprintf("%d", $n);
+    fwrite($f, $x);
+    fclose($f);
+    return true;
+}
+
+function get_count_from_db() {
+    $result = mysql_query("select count(*) from result");
+    if (!$result) return false;
+    $count = mysql_fetch_array($result);
+    mysql_free_result($result);
+    return $count[0];
+}
+
+    $retval = db_init();
+    if (!$retval) {
+        fail("Can't open database");
+    }
+    $m = get_count_from_db();
+    echo "Count: $m\n";
+    if ($m == false) {
+        fail("Can't get result count from DB");
+    }
+    $n = read_count_file();
+    echo "Count: $n\n";
+    if ($n == false) {
+        write_count_file($m);
+        exit();
+    }
+    if ($n == $m) {
+        echo "fail\n";
+        fail("Result count hasn't changed\n");
+    }
+    write_count_file($m);
+
+?>