Update resource failcount in all required cases, fixed bug #147435.

(Portage version: 2.1-r1)

5 files changed, 354 insertions, 8 deletions

diff --git a/sys-cluster/heartbeat/ChangeLog b/sys-cluster/heartbeat/ChangeLog
index 5d8e569a3641..e40b59a7853e 100644
--- a/sys-cluster/heartbeat/ChangeLog
+++ b/sys-cluster/heartbeat/ChangeLog
@@ -1,6 +1,13 @@
 # ChangeLog for sys-cluster/heartbeat
 # Copyright 2000-2006 Gentoo Foundation; Distributed under the GPL v2
-# $Header: /var/cvsroot/gentoo-x86/sys-cluster/heartbeat/ChangeLog,v 1.56 2006/09/17 22:38:42 xmerlin Exp $
+# $Header: /var/cvsroot/gentoo-x86/sys-cluster/heartbeat/ChangeLog,v 1.57 2006/09/17 22:48:05 xmerlin Exp $
+
+*heartbeat-2.0.7-r2 (17 Sep 2006)
+
+  17 Sep 2006; Christian Zoffoli <xmerlin@gentoo.org>
+  +files/heartbeat-2.0.7-update-resources-failcount.patch,
+  heartbeat-2.0.7-r1.ebuild, +heartbeat-2.0.7-r2.ebuild:
+  Update resource failcount in all required cases, fixed bug #147435.
 
   17 Sep 2006; Christian Zoffoli <xmerlin@gentoo.org>
   -files/heartbeat-1.2.3-lvm.patch,
diff --git a/sys-cluster/heartbeat/files/digest-heartbeat-2.0.7-r2 b/sys-cluster/heartbeat/files/digest-heartbeat-2.0.7-r2
new file mode 100644
index 000000000000..8ff7ae56cec2
--- /dev/null
+++ b/sys-cluster/heartbeat/files/digest-heartbeat-2.0.7-r2
@@ -0,0 +1,3 @@
+MD5 41233e5097c42341bf2162806d4cb99c heartbeat-2.0.7.tar.gz 3327126
+RMD160 8fe70e64fc6714fef4e501ae0b240d870a51cf75 heartbeat-2.0.7.tar.gz 3327126
+SHA256 008214f51fd0c49e168d1bb262cb5c967bc166f99d0c8ff3a2623c8ba51a0ab3 heartbeat-2.0.7.tar.gz 3327126
diff --git a/sys-cluster/heartbeat/files/heartbeat-2.0.7-update-resources-failcount.patch b/sys-cluster/heartbeat/files/heartbeat-2.0.7-update-resources-failcount.patch
new file mode 100644
index 000000000000..cd1e563bac53
--- /dev/null
+++ b/sys-cluster/heartbeat/files/heartbeat-2.0.7-update-resources-failcount.patch
@@ -0,0 +1,223 @@
+--- a/crm/tengine/events.c	Wed Sep 06 11:40:40 2006 +0200
++++ b/crm/tengine/events.c	Tue Sep 12 18:17:10 2006 +0200
+@@ -198,51 +198,81 @@ extract_event(crm_data_t *msg)
+ }
+ 
+ static void
+-update_failcount(crm_action_t *action, int rc) 
+-{
+-	crm_data_t *rsc = NULL;
++update_failcount(crm_data_t *event, const char *event_node, int rc) 
++{
+ 	char *attr_name = NULL;
+ 	
+-	const char *task     = NULL;
+-	const char *rsc_id   = NULL;
+-	const char *on_node  = NULL;
+-	const char *on_uuid  = NULL;
+-	const char *interval = NULL;
++	char *task     = NULL;
++	char *rsc_id   = NULL;
++	const char *on_node  = event_node;
++	const char *on_uuid  = event_node;
++	int interval = 0;
+ 
+ 	if(rc == 99) {
+ 		/* this is an internal code for "we're busy, try again" */
+ 		return;
+ 	}
+ 
+-	interval = g_hash_table_lookup(
+-		action->params, crm_meta_name("interval"));
+-	if(interval == NULL) {
+-		return;
+-	}
+-
+-	CRM_CHECK(action->xml != NULL, return);
+-
+-	rsc = find_xml_node(action->xml, XML_CIB_TAG_RESOURCE, TRUE);
+-	CRM_CHECK(rsc != NULL, return);
+-	rsc_id = ID(rsc);
+-	CRM_CHECK(rsc_id != NULL, return);
+-	
+-	task   = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
+-	on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
+-	on_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
+-
+-	CRM_CHECK(task != NULL, return);
+ 	CRM_CHECK(on_uuid != NULL, return);
+-	CRM_CHECK(on_node != NULL, return);
+-	
+-	attr_name = crm_concat("fail-count", rsc_id, '-');
+-	crm_warn("Updating failcount for %s on %s after failed %s: rc=%d",
+-		 rsc_id, on_node, task, rc);
+-	
+-	update_attr(te_cib_conn, cib_none, XML_CIB_TAG_STATUS,
+-		    on_uuid, NULL,NULL, attr_name, XML_NVPAIR_ATTR_VALUE"++");
+-	
+-	crm_free(attr_name);	
++
++	CRM_CHECK(parse_op_key(ID(event), &rsc_id, &task, &interval),return);
++	CRM_CHECK(task != NULL, crm_free(rsc_id); return);
++	CRM_CHECK(rsc_id != NULL, crm_free(task); return);
++	/* CRM_CHECK(on_node != NULL, return); */
++	
++	if(interval > 0) {
++		attr_name = crm_concat("fail-count", rsc_id, '-');
++		crm_warn("Updating failcount for %s on %s after failed %s: rc=%d",
++			 rsc_id, on_node, task, rc);
++	
++		update_attr(te_cib_conn, cib_none, XML_CIB_TAG_STATUS,
++			    on_uuid, NULL,NULL, attr_name,
++			    XML_NVPAIR_ATTR_VALUE"++");
++		crm_free(attr_name);	
++	}
++
++	crm_free(rsc_id);
++	crm_free(task);
++}
++
++static int
++status_from_rc(crm_action_t *action, int orig_status, int rc)
++{
++	int status = orig_status;
++	const char *target_rc_s = g_hash_table_lookup(
++		action->params, crm_meta_name(XML_ATTR_TE_TARGET_RC));
++
++	if(target_rc_s != NULL) {
++		int target_rc = 0;
++		crm_debug_2("Target rc: %s vs. %d", target_rc_s, rc);
++		target_rc = crm_parse_int(target_rc_s, NULL);
++		if(target_rc == rc) {
++			crm_debug_2("Target rc: == %d", rc);
++			if(status != LRM_OP_DONE) {
++				crm_debug_2("Re-mapping op status to"
++					    " LRM_OP_DONE for rc=%d", rc);
++				status = LRM_OP_DONE;
++			}
++		} else {
++			crm_debug_2("Target rc: != %d", rc);
++			if(status != LRM_OP_ERROR) {
++				crm_info("Re-mapping op status to"
++					 " LRM_OP_ERROR for rc=%d", rc);
++				status = LRM_OP_ERROR;
++			}
++		}
++	}
++	
++	/* 99 is the code we use for direct nack's */
++	if(rc != 99 && status != LRM_OP_DONE) {
++		const char *task, *uname;
++		task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
++		uname  = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
++		crm_warn("Action %s on %s failed (target: %s vs. rc: %d): %s",
++			 task, uname, target_rc_s, rc, op_status2text(status));
++	}
++
++	return status;
+ }
+ 
+ /*
+@@ -255,9 +285,6 @@ match_graph_event(
+ match_graph_event(
+ 	crm_action_t *action, crm_data_t *event, const char *event_node)
+ {
+-	int log_level_fail = LOG_ERR;
+-	int target_rc = 0;
+-	const char *target_rc_s = NULL;
+ 	const char *allow_fail  = NULL;
+ 	const char *this_action = NULL;
+ 	const char *this_node   = NULL;
+@@ -296,15 +323,19 @@ match_graph_event(
+ 		crm_debug_2("Action %d : Node mismatch %s (%s) vs. %s",
+ 			    action->id, this_node, this_uname, event_node);
+ 		return -1;
+-
+-	}
+-
++	}
++	
+ 	crm_debug_2("Matched action (%d) %s", action->id, this_event);
+ 
+ 	CRM_CHECK(decode_transition_magic(
+ 			       magic, &update_te_uuid,
+ 			       &transition_i, &op_status_i, &op_rc_i), return -2);
+ 
++	op_status_i = status_from_rc(action, op_status_i, op_rc_i);
++	if(op_status_i != LRM_OP_DONE) {
++		update_failcount(event, event_node, op_rc_i);
++	}
++	
+ 	if(transition_i == -1) {
+ 		/* we never expect these - recompute */
+ 		crm_err("Detected action %s initiated outside of a transition",
+@@ -334,28 +365,6 @@ match_graph_event(
+ 	/* stop this event's timer if it had one */
+ 	stop_te_timer(action->timer);
+ 	action->confirmed = TRUE;
+-
+-	target_rc_s = g_hash_table_lookup(
+-		action->params,crm_meta_name(XML_ATTR_TE_TARGET_RC));
+-	if(target_rc_s != NULL) {
+-		crm_debug_2("Target rc: %s vs. %d", target_rc_s, op_rc_i);
+-		target_rc = crm_parse_int(target_rc_s, NULL);
+-		if(target_rc == op_rc_i) {
+-			crm_debug_2("Target rc: == %d", op_rc_i);
+-			if(op_status_i != LRM_OP_DONE) {
+-				crm_debug_2("Re-mapping op status to"
+-					    " LRM_OP_DONE for %s",update_event);
+-				op_status_i = LRM_OP_DONE;
+-			}
+-		} else {
+-			crm_debug_2("Target rc: != %d", op_rc_i);
+-			if(op_status_i != LRM_OP_ERROR) {
+-				crm_info("Re-mapping op status to"
+-					 " LRM_OP_ERROR for %s", update_event);
+-				op_status_i = LRM_OP_ERROR;
+-			}
+-		}
+-	}
+ 	
+ 	/* Process OP status */
+ 	switch(op_status_i) {
+@@ -371,18 +380,9 @@ match_graph_event(
+ 		case LRM_OP_DONE:
+ 			break;
+ 		case LRM_OP_ERROR:
+-			/* This is the code we use for direct nack's */
+-			if(op_rc_i == 99) {
+-				log_level_fail = LOG_WARNING;
+-			}
+-			/* fall through */
+ 		case LRM_OP_TIMEOUT:
+ 		case LRM_OP_NOTSUPPORTED:
+ 			action->failed = TRUE;
+-			crm_log_maybe(log_level_fail,
+-				"Action %s on %s failed (target: %d vs. rc: %d): %s",
+-				update_event, this_uname, target_rc,
+-				op_rc_i, op_status2text(op_status_i));
+ 			break;
+ 		case LRM_OP_CANCELLED:
+ 			/* do nothing?? */
+@@ -405,10 +405,6 @@ match_graph_event(
+ 	}
+ 
+ 	if(action->failed) {
+-		/* ignore probes */
+-		if(target_rc != EXECRA_NOT_RUNNING) {
+-			update_failcount(action, op_rc_i);
+-		}
+ 		abort_transition(action->synapse->priority+1,
+ 				 tg_restart, "Event failed", event);
+ 
+@@ -544,6 +540,9 @@ process_graph_event(crm_data_t *event, c
+ 	/* unexpected event, trigger a pe-recompute */
+ 	/* possibly do this only for certain types of actions */
+ 	crm_warn("Event not found.");
++	if(rc != EXECRA_OK) {
++		update_failcount(event, event_node, rc);
++	}
+ 	crm_log_xml_info(event, "match:not-found");
+ 	abort_transition(INFINITY, tg_restart, "Unexpected event", event);
+ 	return;
+
diff --git a/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild b/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild
index d5c371523e74..8290397ae031 100644
--- a/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild
+++ b/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild
@@ -1,6 +1,6 @@
 # Copyright 1999-2006 Gentoo Foundation
 # Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild,v 1.1 2006/08/24 16:24:28 xmerlin Exp $
+# $Header: /var/cvsroot/gentoo-x86/sys-cluster/heartbeat/heartbeat-2.0.7-r1.ebuild,v 1.2 2006/09/17 22:48:05 xmerlin Exp $
 
 inherit flag-o-matic eutils
 
@@ -43,12 +43,6 @@ DEPEND="
 	)
 	"
 
-src_unpack() {
-	unpack ${A}
-	cd ${S}
-	#epatch ${FILESDIR}/${PN}-2.0.4-fix-local-DOS-attack.patch || die
-}
-
 src_compile() {
 	append-ldflags $(bindnow-flags)
 
diff --git a/sys-cluster/heartbeat/heartbeat-2.0.7-r2.ebuild b/sys-cluster/heartbeat/heartbeat-2.0.7-r2.ebuild
new file mode 100644
index 000000000000..480c3b1f11e5
--- /dev/null
+++ b/sys-cluster/heartbeat/heartbeat-2.0.7-r2.ebuild
@@ -0,0 +1,119 @@
+# Copyright 1999-2006 Gentoo Foundation
+# Distributed under the terms of the GNU General Public License v2
+# $Header: /var/cvsroot/gentoo-x86/sys-cluster/heartbeat/heartbeat-2.0.7-r2.ebuild,v 1.1 2006/09/17 22:48:05 xmerlin Exp $
+
+inherit flag-o-matic eutils
+
+DESCRIPTION="Heartbeat high availability cluster manager"
+HOMEPAGE="http://www.linux-ha.org"
+SRC_URI="http://www.linux-ha.org/download/${P}.tar.gz"
+
+LICENSE="GPL-2"
+SLOT="0"
+KEYWORDS="~amd64 -mips ~ppc -s390 ~x86"
+IUSE="ldirectord doc snmp management"
+
+DEPEND="
+	=dev-libs/glib-2*
+	net-libs/libnet
+	dev-util/pkgconfig
+	>=dev-lang/perl-5.8.8
+	net-misc/iputils
+	virtual/ssh
+	net-libs/gnutls
+	ldirectord? (	sys-cluster/ipvsadm
+			dev-perl/Net-DNS
+			dev-perl/libwww-perl
+			dev-perl/perl-ldap
+			perl-core/libnet
+			dev-perl/Crypt-SSLeay
+			dev-perl/HTML-Parser
+			dev-perl/perl-ldap
+			dev-perl/Mail-IMAPClient
+			dev-perl/Mail-POP3Client
+			dev-perl/MailTools
+	)
+	snmp? ( net-analyzer/net-snmp )
+	net-misc/telnet-bsd
+	dev-lang/swig
+	management? (
+		>=dev-lang/python-2.4
+		>=dev-python/pygtk-2.4
+		>=sys-libs/pam-0.78-r3
+	)
+	"
+
+src_unpack() {
+	unpack ${A}
+	cd ${S}
+	epatch ${FILESDIR}/${P}-update-resources-failcount.patch || die
+}
+
+src_compile() {
+	append-ldflags $(bindnow-flags)
+
+	./configure --prefix=/usr \
+		--mandir=/usr/share/man \
+		--sysconfdir=/etc \
+		--localstatedir=/var \
+		--with-group-name=cluster \
+		--with-group-id=65 \
+		--with-ccmuser-name=cluster \
+		--with-ccmuser-id=65 \
+		--enable-checkpointd \
+		--enable-crm \
+		--enable-lrm \
+		$(use_enable management mgmt) \
+		|| die
+	emake -j 1 || die "compile problem"
+}
+
+pkg_preinst() {
+	# check for cluster group, if it doesn't exist make it
+	if ! grep -q cluster.*65 /etc/group ; then
+		groupadd -g 65 cluster
+	fi
+	# check for cluster user, if it doesn't exist make it
+	if ! grep -q cluster.*65 /etc/passwd ; then
+		useradd -u 65 -g cluster -s /dev/null -d /var/lib/heartbeat cluster
+	fi
+}
+
+src_install() {
+	make DESTDIR=${D} install || die
+
+	# heartbeat modules need these dirs
+	#keepdir /var/lib/heartbeat/ckpt /var/lib/heartbeat/ccm /var/lib/heartbeat
+
+	keepdir /var/lib/heartbeat/crm /var/lib/heartbeat/lrm /var/lib/heartbeat/fencing
+	keepdir /var/lib/heartbeat/cores/cluster /var/lib/heartbeat/cores/root /var/lib/heartbeat/cores/nobody
+
+	keepdir /var/run/heartbeat/ccm /var/run/heartbeat/crm
+
+	keepdir /etc/ha.d/conf
+
+	dosym /usr/sbin/ldirectord /etc/ha.d/resource.d/ldirectord || die
+
+	if use management ; then
+		keepdir /var/lib/heartbeat/mgmt
+	fi
+
+	# if ! USE="ldirectord" then don't install it
+	if ! use ldirectord ; then
+		rm ${D}/etc/init.d/ldirectord
+		rm ${D}/etc/logrotate.d/ldirectord
+		rm ${D}/etc/ha.d/resource.d/ldirectord
+		rm ${D}/usr/share/man/man8/supervise-ldirectord-config.8
+		rm ${D}/usr/share/man/man8/ldirectord.8
+		rm ${D}/usr/sbin/ldirectord
+		rm ${D}/usr/sbin/supervise-ldirectord-config
+	fi
+
+	exeinto /etc/init.d
+	newexe ${FILESDIR}/heartbeat-init heartbeat
+
+	dodoc ldirectord/ldirectord.cf doc/*.cf doc/haresources doc/authkeys || die
+	if use doc ; then
+		dodoc README doc/*.txt doc/AUTHORS doc/COPYING  || die
+	fi
+}