Send Linux-ha-cvs mailing list submissions to
[email protected]
To subscribe or unsubscribe via the World Wide Web, visit
http://lists.community.tummy.com/mailman/listinfo/linux-ha-cvs
or, via email, send a message with subject or body 'help' to
[EMAIL PROTECTED]
You can reach the person managing the list at
[EMAIL PROTECTED]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of Linux-ha-cvs digest..."
Today's Topics:
1. Linux-HA CVS: crm by andrew from
([email protected])
2. Linux-HA CVS: cts by blaschke from
([email protected])
3. Linux-HA CVS: heartbeat by alan from
([email protected])
----------------------------------------------------------------------
Message: 1
Date: Fri, 21 Apr 2006 03:18:27 -0600 (MDT)
From: [email protected]
Subject: [Linux-ha-cvs] Linux-HA CVS: crm by andrew from
To: [EMAIL PROTECTED]
Message-ID: <[EMAIL PROTECTED]>
linux-ha CVS committal
Author : andrew
Host :
Project : linux-ha
Module : crm
Dir : linux-ha/crm/pengine
Modified Files:
stages.c utils.c pe_utils.h
Log Message:
Log a warning if more than one node has the highest score for running a
given resource.
===================================================================
RCS file: /home/cvs/linux-ha/linux-ha/crm/pengine/stages.c,v
retrieving revision 1.90
retrieving revision 1.91
diff -u -3 -r1.90 -r1.91
--- stages.c 3 Apr 2006 09:51:56 -0000 1.90
+++ stages.c 21 Apr 2006 09:18:26 -0000 1.91
@@ -1,4 +1,4 @@
-/* $Id: stages.c,v 1.90 2006/04/03 09:51:56 andrew Exp $ */
+/* $Id: stages.c,v 1.91 2006/04/21 09:18:26 andrew Exp $ */
/*
* Copyright (C) 2004 Andrew Beekhof <[EMAIL PROTECTED]>
*
@@ -511,11 +511,13 @@
*/
GListPtr nodes = color->details->candidate_nodes;
node_t *chosen = NULL;
+ int multiple = 0;
crm_debug_2("Choosing node for color %d", color->id);
color->details->candidate_nodes = g_list_sort(nodes, sort_node_weight);
+ nodes = color->details->candidate_nodes;
- chosen = g_list_nth_data(color->details->candidate_nodes, 0);
+ chosen = g_list_nth_data(nodes, 0);
color->details->chosen_node = NULL;
color->details->pending = FALSE;
@@ -537,11 +539,38 @@
return FALSE;
}
+
+ slist_iter(candidate, node_t, nodes, lpc,
+ if(chosen->weight > 0
+ && candidate->details->unclean == FALSE
+ && candidate->weight == chosen->weight) {
+ multiple++;
+ } else {
+ break;
+ }
+ );
+
+ if(multiple > 1) {
+ int log_level = LOG_WARNING;
+ char *score = score2char(chosen->weight);
+
+ crm_warn("%d nodes with equal score (%s) for running the"
+ " listed resources (chose %s):",
+ multiple, score, chosen->details->uname);
+ slist_iter(rsc, resource_t,
+ color->details->allocated_resources, lpc,
+ rsc->fns->print(
+ rsc, "\t", pe_print_log|pe_print_rsconly,
+ &log_level);
+ );
+ crm_free(score);
+ }
+
/* todo: update the old node for each resource to reflect its
* new resource count
*/
- crm_debug_2("assigned %s to color %d", chosen->details->uname,
color->id);
+ crm_debug_2("assigned %s to color %d",chosen->details->uname,color->id);
chosen->details->num_resources += color->details->num_resources;
color->details->chosen_node = node_copy(chosen);
===================================================================
RCS file: /home/cvs/linux-ha/linux-ha/crm/pengine/utils.c,v
retrieving revision 1.131
retrieving revision 1.132
diff -u -3 -r1.131 -r1.132
--- utils.c 21 Apr 2006 07:08:04 -0000 1.131
+++ utils.c 21 Apr 2006 09:18:26 -0000 1.132
@@ -1,4 +1,4 @@
-/* $Id: utils.c,v 1.131 2006/04/21 07:08:04 andrew Exp $ */
+/* $Id: utils.c,v 1.132 2006/04/21 09:18:26 andrew Exp $ */
/*
* Copyright (C) 2004 Andrew Beekhof <[EMAIL PROTECTED]>
*
@@ -1855,6 +1855,20 @@
return score_f;
}
+
+char *
+score2char(int score)
+{
+
+ if(score >= INFINITY) {
+ return crm_strdup("+"INFINITY_S);
+
+ } else if(score <= -INFINITY) {
+ return crm_strdup("-"INFINITY_S);
+ }
+ return crm_itoa(score);
+}
+
rsc_to_node_t *
rsc2node_new(const char *id, resource_t *rsc,
int node_weight, node_t *foo_node, pe_working_set_t *data_set)
===================================================================
RCS file: /home/cvs/linux-ha/linux-ha/crm/pengine/pe_utils.h,v
retrieving revision 1.41
retrieving revision 1.42
diff -u -3 -r1.41 -r1.42
--- pe_utils.h 31 Mar 2006 12:05:37 -0000 1.41
+++ pe_utils.h 21 Apr 2006 09:18:26 -0000 1.42
@@ -1,4 +1,4 @@
-/* $Id: pe_utils.h,v 1.41 2006/03/31 12:05:37 andrew Exp $ */
+/* $Id: pe_utils.h,v 1.42 2006/04/21 09:18:26 andrew Exp $ */
/*
* Copyright (C) 2004 Andrew Beekhof <[EMAIL PROTECTED]>
*
@@ -173,6 +173,7 @@
extern const char *fail2text(enum action_fail_response fail);
extern int char2score(const char *score);
+extern char *score2char(int score);
/* free the various structures */
------------------------------
Message: 2
Date: Fri, 21 Apr 2006 10:04:35 -0600 (MDT)
From: [email protected]
Subject: [Linux-ha-cvs] Linux-HA CVS: cts by blaschke from
To: [EMAIL PROTECTED]
Message-ID: <[EMAIL PROTECTED]>
linux-ha CVS committal
Author : blaschke
Host :
Project : linux-ha
Module : cts
Dir : linux-ha/cts
Modified Files:
CTSlab.py.in
Log Message:
fix one missed typo
===================================================================
RCS file: /home/cvs/linux-ha/linux-ha/cts/CTSlab.py.in,v
retrieving revision 1.62
retrieving revision 1.63
diff -u -3 -r1.62 -r1.63
--- CTSlab.py.in 19 Apr 2006 14:52:32 -0000 1.62
+++ CTSlab.py.in 21 Apr 2006 16:04:34 -0000 1.63
@@ -784,7 +784,7 @@
tests.summarize()
if tests.Stats["failure"] > 0:
- sys.exit(test.Stats["failure"])
+ sys.exit(tests.Stats["failure"])
elif tests.Stats["success"] != NumIter:
cm.Env.log("No failure count but success != requested iterations")
------------------------------
Message: 3
Date: Fri, 21 Apr 2006 11:47:46 -0600 (MDT)
From: [email protected]
Subject: [Linux-ha-cvs] Linux-HA CVS: heartbeat by alan from
To: [EMAIL PROTECTED]
Message-ID: <[EMAIL PROTECTED]>
linux-ha CVS committal
Author : alan
Host :
Project : linux-ha
Module : heartbeat
Dir : linux-ha/heartbeat
Modified Files:
heartbeat.c
Log Message:
Put in a bug fix which only occurs if we send lots of packets
before initdead expires, and we're the only node up.
It caused problems in BSC with a 10ms heartbeat time.
What triggered the need for this patch was decreasing the window
size of the protocol to "only" 200 packets (instead of 1000).
===================================================================
RCS file: /home/cvs/linux-ha/linux-ha/heartbeat/heartbeat.c,v
retrieving revision 1.503
retrieving revision 1.504
diff -u -3 -r1.503 -r1.504
--- heartbeat.c 20 Apr 2006 17:14:56 -0000 1.503
+++ heartbeat.c 21 Apr 2006 17:47:46 -0000 1.504
@@ -1,8 +1,4 @@
-/*
- * TODO:
- * 1) Man page update
- */
-/* $Id: heartbeat.c,v 1.503 2006/04/20 17:14:56 alan Exp $ */
+/* $Id: heartbeat.c,v 1.504 2006/04/21 17:47:46 alan Exp $ */
/*
* heartbeat: Linux-HA heartbeat code
*
@@ -283,6 +279,8 @@
#define ALWAYSRESTART_ON_SPLITBRAIN 1
+#define FLOWCONTROL_LIMIT ((seqno_t)(MAXMSGHIST/2))
+
static char hbname []= "heartbeat";
const char * cmdname = hbname;
@@ -334,8 +332,9 @@
static int deadtime_tmpadd_count = 0;
gboolean enable_flow_control = TRUE;
static int send_cluster_msg_level = 0;
+static int live_node_count = 1; /* namely us... */
static void print_a_child_client(gpointer childentry, gpointer unused);
-static seqno_t timer_lowseq = 0;
+static seqno_t timer_lowseq = 0;
static gboolean init_deadtime_passed = FALSE;
static int PrintDefaults = FALSE;
static int WikiOutput = FALSE;
@@ -399,6 +398,7 @@
static void init_xmit_hist (struct msg_xmit_hist * hist);
static void process_rexmit(struct msg_xmit_hist * hist
, struct ha_msg* msg);
+static void update_ackseq(seqno_t new_ackseq) ;
static void process_clustermsg(struct ha_msg* msg, struct link* lnk);
extern void process_registerevent(IPC_Channel* chan, gpointer user_data);
static void nak_rexmit(struct msg_xmit_hist * hist,
@@ -2148,21 +2148,20 @@
HBDoMsg_T_ACKMSG(const char * type, struct node_info * fromnode,
TIME_T msgtime, seqno_t seqno, const char * iface, struct ha_msg
* msg)
{
- const char* ackseq_str = ha_msg_value(msg, F_ACKSEQ);
- seqno_t ackseq;
- struct msg_xmit_hist* hist = &msghist;
- const char* to =
- (const char*)ha_msg_value(msg, F_TO);
- struct node_info* tonode;
- seqno_t old_hist_ackseq;
+ const char* ackseq_str = ha_msg_value(msg, F_ACKSEQ);
+ seqno_t ackseq;
+ struct msg_xmit_hist* hist = &msghist;
+ const char* to = (const char*)ha_msg_value(msg, F_TO);
+ struct node_info* tonode;
+ seqno_t new_ackseq = hist->ackseq;
if (!to || (tonode = lookup_tables(to, NULL)) == NULL
- || tonode != curnode){
+ || tonode != curnode){
return;
}
- if (ackseq_str == NULL||
- sscanf(ackseq_str, "%lx", &ackseq) != 1){
+ if (ackseq_str == NULL
+ || sscanf(ackseq_str, "%lx", &ackseq) != 1){
goto out;
}
@@ -2172,11 +2171,10 @@
goto out;
}
- if (ackseq <= hist->ackseq){
+ if (ackseq <= new_ackseq){
/* late or dup ack
* ignored
*/
-
goto out;
}else if (ackseq > hist->hiseq){
cl_log(LOG_ERR, "HBDoMsg_T_ACK"
@@ -2187,7 +2185,7 @@
goto out;
}
- if ( ackseq < fromnode->track.ackseq){
+ if (ackseq < fromnode->track.ackseq) {
/* late or dup ack
* ignored
*/
@@ -2196,20 +2194,18 @@
fromnode->track.ackseq = ackseq;
- if (hist->lowest_acknode != NULL &&
- STRNCMP_CONST(hist->lowest_acknode->status,
- DEADSTATUS) == 0){
+ if (hist->lowest_acknode != NULL
+ && STRNCMP_CONST(hist->lowest_acknode->status,DEADSTATUS)==0){
/* the lowest acked node is dead
- * we cannnot count on that node
+ * we cannot count on that node
* to update our ackseq
*/
hist->lowest_acknode = NULL;
}
- old_hist_ackseq = hist->ackseq;
- if (hist->lowest_acknode == NULL ||
- hist->lowest_acknode == fromnode){
+ if (hist->lowest_acknode == NULL
+ || hist->lowest_acknode == fromnode){
/*find the new lowest and update hist->ackseq*/
seqno_t minseq;
int minidx;
@@ -2221,79 +2217,103 @@
for (i = 0; i < config->nodecount; i++){
struct node_info* hip = &config->nodes[i];
- if (STRNCMP_CONST(hip->status,DEADSTATUS) == 0
- || hip->nodetype == PINGNODE_I){
+ if (hip->nodetype == PINGNODE_I
+ || STRNCMP_CONST(hip->status, DEADSTATUS) == 0) {
continue;
}
- if (minidx == -1 ||
- hip->track.ackseq < minseq){
+ if (minidx == -1
+ || hip->track.ackseq < minseq){
minseq = hip->track.ackseq;
minidx = i;
}
}
- if (minidx == -1){
- /*each node is in either DEASTATUS or INITSTATUS*/
+ if (minidx == -1) {
+ /* Every node is DEADSTATUS */
goto out;
}
- if (minidx == config->nodecount){
+ if (live_node_count < 2) {
+ /*
+ * Update hist->ackseq so we don't hang onto
+ * messages indefinitely and flow control clients
+ */
+ if ((hist->hiseq - new_ackseq) >= FLOWCONTROL_LIMIT) {
+ new_ackseq = hist->hiseq -
(FLOWCONTROL_LIMIT-1);
+ }
+ hist->lowest_acknode = NULL;
+ goto cleanupandout;
+ }
+ if (minidx >= config->nodecount) {
cl_log(LOG_ERR, "minidx out of bound"
"minidx=%d",minidx );
goto out;
}
- if (minseq != 0){
- hist->ackseq = minseq;
+ if (minseq > 0){
+ new_ackseq = minseq;
}
-
hist->lowest_acknode = &config->nodes[minidx];
-
- if (hist->hiseq - hist->ackseq < MAXMSGHIST/2){
- all_clients_resume();
- }
}
+cleanupandout:
+ update_ackseq(new_ackseq);
+out:
+ return;
+}
+
+static void
+update_ackseq(seqno_t new_ackseq)
+{
+ struct msg_xmit_hist* hist = &msghist;
+ long count;
+ seqno_t start;
+ seqno_t old_ackseq = hist->ackseq;
+
#if 0
- cl_log(LOG_INFO, "hist->ackseq =%ld, old_hist_ackseq=%ld",
- hist->ackseq, old_hist_ackseq);
+ cl_log(LOG_INFO, "new_ackseq = %ld, old_ackseq=%ld"
+ , new_ackseq, old_ackseq);
#endif
- if (hist->ackseq > old_hist_ackseq){
- long count;
- seqno_t start;
- count = hist->ackseq - hist->lowseq - send_cluster_msg_level;
- if (old_hist_ackseq == 0){
- start = 0;
- count = count - 1;
- }else{
- start = hist->lowseq;
+ if (new_ackseq <= old_ackseq){
+ return;
+ }
+ hist->ackseq = new_ackseq;
+
+ if ((hist->hiseq - hist->ackseq) < FLOWCONTROL_LIMIT){
+ all_clients_resume();
+ }
+
+ count = hist->ackseq - hist->lowseq - send_cluster_msg_level;
+ if (old_ackseq == 0){
+ start = 0;
+ count = count - 1;
+ }else{
+ start = hist->lowseq;
+ }
+
+ while(count -- > 0){
+ /*
+ * If the seq number is greater than the lowseq number
+ * the timer set, we should not free any more messages
+ */
+ if (start > timer_lowseq){
+ break;
}
-
- while(count -- > 0){
-
- /*if the seq number is greater than the lowseq number
- the timer set, we should not free any more messages*/
- if (start > timer_lowseq){
- break;
- }
-
- free_one_hist_slot(hist, start%MAXMSGHIST);
- start++;
-
- if (hist->lowseq > hist->ackseq){
- cl_log(LOG_ERR, "lowseq cannnot be greater than
ackseq");
- cl_log(LOG_INFO, "hist->ackseq =%ld,
old_hist_ackseq=%ld",
- hist->ackseq, old_hist_ackseq);
- cl_log(LOG_INFO, "hist->lowseq =%ld,
hist->hiseq=%ld,"
- "send_cluster_msg_level=%d",
- hist->lowseq, hist->hiseq,
send_cluster_msg_level);
- abort();
- }
+
+ free_one_hist_slot(hist, start%MAXMSGHIST);
+ start++;
+ if (hist->lowseq > hist->ackseq){
+ cl_log(LOG_ERR, "lowseq cannnot be greater than
ackseq");
+ cl_log(LOG_INFO, "hist->ackseq =%ld, old_ackseq=%ld"
+ , hist->ackseq, old_ackseq);
+ cl_log(LOG_INFO, "hist->lowseq =%ld, hist->hiseq=%ld"
+ ", send_cluster_msg_level=%d"
+ , hist->lowseq, hist->hiseq,
send_cluster_msg_level);
+ abort();
}
}
-
(void)dump_missing_pkts_info;
#ifdef DEBUG_FOR_GSHI
@@ -2312,13 +2332,8 @@
}
#endif
- out:
-
- return;
}
-
-
static int
getnodes(const char* nodelist, char** nodes, int* num){
@@ -2798,9 +2813,30 @@
, "Status seqno: %ld msgtime: %ld"
, seqno, msgtime);
}
+ /*
+ * IF
+ * It's from a normal node
+ * It isn't from us
+ * The node's old status was dead or init
+ * The node's new status is up or active
+ * THEN
+ * increment the count of live nodes.
+ */
+ if (fromnode->nodetype == NORMALNODE_I
+ && fromnode != curnode
+ && ( STRNCMP_CONST(fromnode->status, DEADSTATUS) == 0
+ || STRNCMP_CONST(fromnode->status, INITSTATUS) == 0)
+ && ( STRNCMP_CONST(status, UPSTATUS) == 0
+ || STRNCMP_CONST(status, ACTIVESTATUS) == 0)) {
+ ++live_node_count;
+ if (live_node_count > config->nodecount) {
+ cl_log(LOG_ERR
+ , "live_node_count too big (%d)"
+ , live_node_count);
+ }
+ }
- strncpy(fromnode->status, status
- , sizeof(fromnode->status));
+ strncpy(fromnode->status, status, sizeof(fromnode->status));
if (!fromnode->status_suppressed) {
QueueRemoteRscReq(PerformQueuedNotifyWorld, msg);
heartbeat_monitor(msg, KEEPIT, iface);
@@ -4190,6 +4226,11 @@
return;
}
+ if (hip->nodetype == NORMALNODE_I
+ && STRNCMP_CONST(hip->status, DEADSTATUS) != 0
+ && STRNCMP_CONST(hip->status, INITSTATUS) != 0) {
+ --live_node_count;
+ }
strncpy(hip->status, DEADSTATUS, sizeof(hip->status));
@@ -5687,7 +5728,7 @@
struct msg_xmit_hist* hist = &msghist;
- return hist->hiseq - hist->ackseq > MAXMSGHIST/2;
+ return hist->hiseq - hist->ackseq > FLOWCONTROL_LIMIT;
}
@@ -5729,20 +5770,30 @@
hist->lastrexmit[slot] = 0L;
hist->lastmsg = slot;
- if (enable_flow_control && hist->hiseq - hist->lowseq > MAXMSGHIST*3 /
4){
- cl_log(LOG_ERR, "Message hist queue is filling up (%d messages
in queue)",
- (int)(hist->hiseq - hist->lowseq));
+ if (enable_flow_control
+ && live_node_count > 1
+ && (hist->hiseq - hist->lowseq) > ((MAXMSGHIST*3)/4)) {
+ cl_log(LOG_ERR
+ , "Message hist queue is filling up"
+ " (%d messages in queue)"
+ , (int)(hist->hiseq - hist->lowseq));
+ hist_display(hist);
}
-
AUDITXMITHIST;
if (enable_flow_control
- && hist->hiseq - hist->ackseq > MAXMSGHIST/2){
- all_clients_pause();
- hist_display(hist);
+ && hist->hiseq - hist->ackseq > FLOWCONTROL_LIMIT){
+ if (live_node_count < 2) {
+ update_ackseq(hist->hiseq - (FLOWCONTROL_LIMIT-1));
+ all_clients_resume();
+ }else{
+ cl_log(LOG_ERR, "Flow control engaged with %d live
nodes"
+ , live_node_count);
+ all_clients_pause();
+ hist_display(hist);
+ }
}
-
}
@@ -6132,6 +6183,13 @@
/*
* $Log: heartbeat.c,v $
+ * Revision 1.504 2006/04/21 17:47:46 alan
+ * Put in a bug fix which only occurs if we send lots of packets
+ * before initdead expires, and we're the only node up.
+ * It caused problems in BSC with a 10ms heartbeat time.
+ * What triggered the need for this patch was decreasing the window
+ * size of the protocol to "only" 200 packets (instead of 1000).
+ *
* Revision 1.503 2006/04/20 17:14:56 alan
* Changed some timing code to not be quite a particular as it had been...
*
------------------------------
_______________________________________________
Linux-ha-cvs mailing list
[email protected]
http://lists.community.tummy.com/mailman/listinfo/linux-ha-cvs
End of Linux-ha-cvs Digest, Vol 29, Issue 113
*********************************************
