[Pacemaker] pacemaker-mgmt (pygui) on fedora 20

Thu Jan 8 13:25:50 UTC 2015

Hi Dori,

On 01/08/2015 10:32 AM, Dori Seliskar wrote:
> Hi all,
> 
> In the past I have put to good use the pacemaker-mgmt (crm_gui) , mostly
> when I have to
> 
> delegate management of cluster to third person, so much easier to
> explain ...
> 
> So for latest cluster I managed to get pacemaker-mgmt working on Fedora
> 20 (since it is not so straight forward
> 
> I included rough procedure below, for those interested), 
Nice! Thanks for doing this :)

> not tested to
> the extent yet, but all frequently used
> 
> operations work (creating resource, starting/stopping resource, put
> cluster/resource in unmanaged state ...)
> 
> My question is following:
> 
> Before the mgmtd was started by corosync like this:
> 
> corosync.conf:
> 
> service {
> 
> 	# Load the Pacemaker Cluster Resource Manager
> 
> 	name: pacemaker
> 
> 	ver:  0
> 
> 	use_mgmtd: 1
> 
> }
> 
> that is apparently not the case anymore, 
This only works if running on corosync-1.x, which is able to help spawn
pacemaker daemons and mgmtd.

> so I start mgmtd l by hand like
> this:
> 
> /usr/lib64/heartbeat/mgmtd -v &
> 
> This mostly work. But if corosync/pacemaker is not started or just
> starting it fails miserably with message Aborted.
> 
> If I wait that cluster comes up and online then it works.
> 
> My question is - how to start mgmtd by cluster itself properly as before ?
Actually, for corosync-2, the decent way would be an init script or a
systemd service file for mgmtd to start it after pacemaker. Patches are
welcome ;)

> 
> I use:
> 
> corosync-2.3.4-1.fc20.x86_64
> 
> corosynclib-devel-2.3.4-1.fc20.x86_64
> 
> pacemaker-doc-1.1.12-2.fc20.x86_64
> 
> corosynclib-2.3.4-1.fc20.x86_64
> 
> pacemaker-libs-1.1.12-2.fc20.x86_64
> 
> pacemaker-cts-1.1.12-2.fc20.x86_64
> 
> pacemaker-cli-1.1.12-2.fc20.x86_64
> 
> pacemaker-libs-devel-1.1.12-2.fc20.x86_64
> 
> pacemaker-1.1.12-2.fc20.x86_64
> 
> pcs-0.9.48-1.fc20.x86_64
> 
> pacemaker-cluster-libs-1.1.12-2.fc20.x86_64
> 
> pacemaker-remote-1.1.12-2.fc20.x86_64
> 
> Best regards,
> 
> ds
> 
> # procedure to get pacemaker-mgmt working on Fedora 20 starting from
> minimal install, rather long, safe to skip ;)
> 
> yum install pacemaker pacemaker-cli pacemaker-cluster-libs pacemaker-doc
> pacemaker-cts pacemaker-libs pacemaker-libs-devel pacemaker-remote
> 
> wget
> http://download.opensuse.org/repositories/network:/ha-clustering:/Stable/Fedora_20/x86_64/cluster-glue-1.0.12-1.15.2.x86_64.rpm
> 
> wget
> http://download.opensuse.org/repositories/network:/ha-clustering:/Stable/Fedora_20/x86_64/cluster-glue-libs-1.0.12-1.15.2.x86_64.rpm
> 
> wget
> http://download.opensuse.org/repositories/network:/ha-clustering:/Stable/Fedora_20/x86_64/cluster-glue-libs-devel-1.0.12-1.15.2.x86_64.rpm
> 
> # --nodeps because of sdb missing dependency, if you are not using sdb
> you should be fine, btw where is sbd nowadays ?
Hmm, I guess sbd has failed to be built in that repo. Anyway, here is
the sbd upstream:
https://github.com/l-mb/sbd

> 
> rpm -ivh --nodeps cluster-glue-1.0.12-1.15.2.x86_64.rpm
> cluster-glue-libs-1.0.12-1.15.2.x86_64.rpm
> cluster-glue-libs-devel-1.0.12-1.15.2.x86_64.rpm
> 
> # install some more dependencies
> 
> yum install OpenIPMI-devel asciidoc autoconf automake docbook-dtds
> docbook-style-xsl libaio-devel libtool openssl-devel python-devel
> libcurl-devel net-snmp-devel openhpi-devel
> 
> yum install intltool gettext gettext-devel gettext-common
> gettext-common-devel gettext-libs gnutls-devel gnutls swig pam-devel
> python-devel pygtk2 xauth
> 
> wget https://github.com/ClusterLabs/pacemaker-mgmt/archive/master.zip
> 
> unzip pacemaker-mgmt/archive/master.zip
> 
> cd pacemaker-mgmt-master/
> 
> #save following text into patch.txt
> 
> ===== SNIP =====
> 
> --- pacemaker-mgmt-master/configure.in 2014-10-20 17:17:56.000000000 +0200
> 
> +++ pacemaker-mgmt-master.patched/configure.in 2015-01-07
> 11:43:09.622664993 +0100
> 
> @@ -55,7 +55,7 @@
> 
> AC_SUBST(GETTEXT_PACKAGE)
> 
> AC_DEFINE_UNQUOTED(GETTEXT_PACKAGE, "$GETTEXT_PACKAGE", [Gettext package])
> 
> #IT_PROG_INTLTOOL([0.40.5], [no-xml])
> 
> -AC_PROG_INTLTOOL(0.35.2)
> 
> +AC_PROG_INTLTOOL
> 
> AM_PO_SUBDIRS
> 
> # Set DATADIRNAME correctly if it is not set yet
> 
> --- pacemaker-mgmt-master/mgmt/daemon/mgmt_crm.c 2014-10-20
> 17:17:56.000000000 +0200
> 
> +++ pacemaker-mgmt-master.patched/mgmt/daemon/mgmt_crm.c 2015-01-07
> 11:46:05.960668547 +0100
> 
> @@ -1652,7 +1652,7 @@
> 
> delete_lrm_rsc(crmd_channel, argv[1], argv[2]);
> 
> refresh_lrm(crmd_channel, NULL);
> 
> - rc = query_node_uuid(cib_conn, argv[1], &dest_node);
> 
> + rc = query_node_uuid(cib_conn, argv[1], &dest_node, NULL);
> 
> if (rc != 0) {
> 
> mgmt_log(LOG_WARNING, "Could not map uname=%s to a UUID: %s\n",
> 
> argv[1], pcmk_strerror(rc));
> 
> @@ -1678,7 +1678,7 @@
> 
> XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh",
> now_s, FALSE);
> 
> #else
> 
> update_attr_delegate(cib_conn, cib_sync_call,
> 
> - XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh",
> now_s, FALSE, NULL);
> 
> + XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh",
> now_s, FALSE, NULL, NULL);
> 
> #endif
> 
> free(now_s);
> 
> ===== SNIP =====
Yeah, correct. There have been some changes of APIs in pacemaker. I was
thinking about to figure out some way to conditionally compile the bits
so that it would not break the compatibility with older pacemaker, but
didn't get the chance. But maybe that would not be necessary any more. I
guess most people are with pacemaker-1.1.11 or 1.1.22 already.

Regards,
  Yan

> 
> # patch it
> 
> patch -p1 < patch.txt
> 
> ./ConfigureMe install --disable-fatal-warnings
> 
> # run mgmtd
> 
> /usr/lib64/heartbeat/mgmtd -v &
> 
> # run gui
> 
> crm_gui &
> 
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
> 

-- 
Gao,Yan <ygao at suse.com>
Senior Software Engineer
SUSE LINUX Products GmbH