From 6aa21f5c1f1525dc84165051d5a29d525e43dc17 Mon Sep 17 00:00:00 2001
From: David Anderson <davea@ssl.berkeley.edu>
Date: Fri, 3 Jun 2005 21:39:38 +0000
Subject: [PATCH] *** empty log message ***

svn path=/trunk/boinc/; revision=6285
---
 checkin_notes          | 15 +++++++++
 doc/index.php          |  2 ++
 doc/links.php          |  4 ++-
 sched/handle_request.C | 73 +++++++++++++++++++++++++++++-------------
 sched/sched_util.C     |  7 ++--
 5 files changed, 74 insertions(+), 27 deletions(-)

diff --git a/checkin_notes b/checkin_notes
index 442cf82b92..7b161ad650 100755
--- a/checkin_notes
+++ b/checkin_notes
@@ -7380,3 +7380,18 @@ David  2 June 2005
         client_types.C
         cs_apps.C
         cs_scheduler.C
+
+David  3 June 2005
+    Stuff related to per-host lockfiles in scheduler:
+
+    - enable locking only if <sched_lockfile_dir> is present in config file
+    - delete lockfiles from sched_lockfile_dir (NOT necessarily CGI dir)
+    - if mylockf(fd) fails (can't lock file) close fd before returning
+        Otherwise would overflow fd limit under FCGI
+    - Fix error message if lock_sched() returns -1
+        (-1 is not a PID)
+    - move lock-release logic into separate function (unlock_sched())
+
+    sched/
+        handle_request.C
+        sched_util.C
diff --git a/doc/index.php b/doc/index.php
index a639c702ff..8eb9c40233 100644
--- a/doc/index.php
+++ b/doc/index.php
@@ -86,6 +86,8 @@ computer resources
   <tr><td bgcolor=$light_blue><h2>Resources</h2></td></tr>
   <tr><td>
     <ul>
+    <li> <a href=http://boinc-doc.net/boinc-wiki/>BOINC documentation Wiki</a>:
+        much of the content of this web site, in Wiki (editable) form.
     <li> <a href=links.php>Web sites</a> for BOINC participants
     <li> <a href=community.php#email_lists>Email lists</a>
     <li> An <a href=intro.php>overview</a> of BOINC, and links to papers
diff --git a/doc/links.php b/doc/links.php
index ab886a2f52..dd5906ce2c 100644
--- a/doc/links.php
+++ b/doc/links.php
@@ -21,7 +21,8 @@ page_head("Web sites for BOINC participants");
 
 echo "
 <h2>Project status</h2>
-<a href=http://boincprojectstatus.ath.cx/boinc/>BOINC project status</a>
+<a href=http://boincprojectstatus.ath.cx/boinc/>BOINC project status</a>:
+shows whether the servers of various projects are up or down.
 <h2>Statistics</h2>
 ";
 stats_sites();
@@ -54,6 +55,7 @@ language("Dutch", array(
     )
 ));
 language("English", array(
+    site("http://www.boinc-doc.net", "boinc-doc.net"),
     site("http://www.boincsynergy.com/", "BOINC Synergy"),
     site("http://www.esea.dk/esea/boinc.asp", "Earth Space Exploration Agency"),
     site("http://boinc.mundayweb.com", "boinc.mundayweb.com - stats counters and more")
diff --git a/sched/handle_request.C b/sched/handle_request.C
index 822da79fc0..4960f5fece 100644
--- a/sched/handle_request.C
+++ b/sched/handle_request.C
@@ -53,8 +53,14 @@ using namespace std;
 #include "fcgi_stdio.h"
 #endif
 
-// use advisory locking to establish a lock to run a scheduler
-// instance for this host. Return values same as mylockf().
+// Try to lock a file with name based on host ID,
+// to prevent 2 schedulers from running at same time for same host.
+// Return:
+// 0 if successful
+//    In this case store file descriptor in reply struct so we can unlock later
+//    In other cases store -1 in reply struct
+// PID (>0) if another process has lock
+// -1 if error (e.g. can't create file)
 //
 int lock_sched(SCHEDULER_REPLY& reply) {
     char filename[256];
@@ -63,30 +69,42 @@ int lock_sched(SCHEDULER_REPLY& reply) {
 
     reply.lockfile_fd=-1;
 
-    if (strlen(config.sched_lockfile_dir)) {
-        sprintf(filename, "%s/CGI_%07d", config.sched_lockfile_dir, reply.host.id);
-    } else {
-        sprintf(filename, "CGI_%07d", reply.host.id);
-    }
+    sprintf(filename, "%s/CGI_%07d", config.sched_lockfile_dir, reply.host.id);
 
     fd = open(filename, O_WRONLY|O_CREAT, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH);
     if (fd < 0) return -1;
 
     // if we can't get an advisory write lock on the file,
     // return the PID of the process that DOES hold the lock.
+    // (or -1 if failure)
     //
-    if ((pid=mylockf(fd))) return pid;
+    pid = mylockf(fd);
+    if (pid) {
+        close(fd);
+        return pid;
+    }
 
     // write PID into the CGI_<HOSTID> file and flush to disk
     //
-    count=sprintf(pid_string, "%d\n", getpid());
+    count = sprintf(pid_string, "%d\n", getpid());
     write(fd, pid_string, count);
     fsync(fd);
 
-    reply.lockfile_fd=fd;
+    reply.lockfile_fd = fd;
     return 0;
 }
 
+// unlock and delete per-host lockfile
+//
+void unlock_sched(SCHEDULER_REPLY& reply) {
+    char filename[256];
+
+    if (reply.lockfile_fd < 0) return;
+    sprintf(filename, "%s/CGI_%07d", config.sched_lockfile_dir, reply.host.id);
+    unlink(filename);
+    close(reply.lockfile_fd);
+}
+
 // If user's email addr is munged (i.e. of the form @X_Y,
 // where X is email and Y is authenticator) then unmunge it.
 // This can fail if there's already an account with same email
@@ -931,7 +949,6 @@ void process_request(
     bool have_no_work;
     char buf[256];
     HOST initial_host;
-    int pid_with_lock;
 
     // if different major version of BOINC, just send a message
     //
@@ -1004,13 +1021,25 @@ void process_request(
     );
     ++log_messages;
 
-    if ((pid_with_lock=lock_sched(reply))) {
-        log_messages.printf(SCHED_MSG_LOG::CRITICAL,
-            "Another scheduler instance [PID=%d] is running for this host\n",
-            pid_with_lock
-        );
-        send_message("Another scheduler instance is running for this host", 60, false);
-        goto leave;
+    if (strlen(config.sched_lockfile_dir)) {
+        int pid_with_lock = lock_sched(reply);
+        if (pid_with_lock > 0) {
+            log_messages.printf(SCHED_MSG_LOG::CRITICAL,
+                "Another scheduler instance [PID=%d] is running for this host\n",
+                pid_with_lock
+            );
+        } else if (pid_with_lock) {
+            log_messages.printf(SCHED_MSG_LOG::CRITICAL,
+                "Error acquiring lock for [HOST#%d]\n", reply.host.id
+            );
+        }
+        if (pid_with_lock) {
+            send_message(
+                "Another scheduler instance is running for this host",
+                60, false
+            );
+            goto leave;
+        }
     }
 
     last_rpc_time = reply.host.rpc_time;
@@ -1239,11 +1268,9 @@ void handle_request(
 #endif
     
     sreply.write(fout);
-    if (sreply.lockfile_fd>=0) {
-        char filename[64];
-        sprintf(filename, "CGI_%07d", sreply.host.id);
-        unlink(filename);
-        close(sreply.lockfile_fd);
+
+    if (strlen(config.sched_lockfile_dir)) {
+        unlock_sched(sreply);
     }
 }
 
diff --git a/sched/sched_util.C b/sched/sched_util.C
index 28d685610b..cc429aca9f 100644
--- a/sched/sched_util.C
+++ b/sched/sched_util.C
@@ -234,9 +234,10 @@ int elapsed_time() {
     return (int)(time(0)-execution_time);
 }
 
-// returns zero if we get lock on file with file descriptor fd.
-// returns < 0 if error
-// returns PID > 0 if another process has lock
+// Request lock on the given file with given fd.  Returns:
+// 0 if we get lock
+// PID (>0) if another process has lock
+// -1 if error
 //
 int mylockf(int fd) {
     struct flock fl;