Hide Forgot
Description of problem: If a user makes a mistake when manually editting cluster.conf, subsequent use of the 'ccs' command can destroy all the information in the file Version-Release number of selected component (if applicable): RHEL6.1 How reproducible: Easily Steps to Reproduce: during cluster configuration I modified a line in cluster.conf. But the xml syntax was wrong because closing tag was missing for quroumd ... </rm> <quorumd label="qdisk_cbgc_cluster" min_score="0"> <cman> ... Then I want to increase the version by ccs -h <hostname> --incversion and the whole configuration was reseted to <?xml version="1.0"?> <cluster config_version="2" name="cluster"> <fence_daemon/> <clusternodes/> <cman/> <fencedevices/> <rm> <failoverdomains/> <resources/> </rm> </cluster> see case: https://access.redhat.com/support/cases/00515395 - the salesforce version contains a patch :-) If I used the command ccs -f cluster.conf --incversion then a python xml.parsers.expat.ExpatError occurs.
Patch from Shane Bradley: Reviewing the source code for the error that was created in comment #9 when -f is given: $ ccs -h localhost --incversion --file /root/sdfc/00515395/cluster.conf Traceback (most recent call last): File "/usr/sbin/ccs", line 1974, in <module> main(sys.argv[1:]) File "/usr/sbin/ccs", line 240, in main if (incrementversion): increment_version() File "/usr/sbin/ccs", line 699, in increment_version new_version = int(get_version()) + 1 File "/usr/sbin/ccs", line 690, in get_version dom = minidom.parseString(get_cluster_conf_xml()) File "/usr/sbin/ccs", line 715, in get_cluster_conf_xml dom = minidom.parseString(xml) File "/usr/lib64/python2.6/xml/dom/minidom.py", line 1928, in parseString return expatbuilder.parseString(string) File "/usr/lib64/python2.6/xml/dom/expatbuilder.py", line 940, in parseString return builder.parseString(string) File "/usr/lib64/python2.6/xml/dom/expatbuilder.py", line 223, in parseString parser.Parse(string, True) xml.parsers.expat.ExpatError: mismatched tag: line 37, column 2 ------------------ Here is patch for fixing the above error from src.rpm: ~/redhat/TMP/ricci-0.16.2-35.el6.src.rpm $ cat ../SOURCES/ccs-catchparseerror.patch diff -up ricci-0.16.2/ccs/ccs.org ricci-0.16.2/ccs/ccs --- ricci-0.16.2/ccs/ccs.org 2011-08-15 15:13:22.076064502 -0400 +++ ricci-0.16.2/ccs/ccs 2011-08-15 15:14:07.454047969 -0400 @@ -3,6 +3,7 @@ import getopt, sys import socket, ssl from xml.dom import minidom +from xml.parsers import expat import logging import os import os.path @@ -712,7 +713,13 @@ def get_cluster_conf_xml(): else: xml = send_ricci_command("cluster", "get_cluster.conf") - dom = minidom.parseString(xml) + dom = None + try: + dom = minidom.parseString(xml) + except expat.ExpatError: + print "Cluster.conf file specified is not a valid cluster.conf file and contains invalid tags which cannot be parsed." + sys.exit(1) + if dom.getElementsByTagName('cluster').length > 0: xml = dom.getElementsByTagName('cluster')[0].toxml() if verifyconf and verify_cluster_conf(xml) != 0: ----- Still does not fix error when not using -f
There are two separate bugs here: 1. Ricci will respond to get_cluster_conf with a blank cluster.conf file if it doesn't understand the cluster.conf file present. Which is what happened in the bug description. However, this has been how ricci worked in RHEL6 in 6.0 & 6.1 (as well as all of RHEL5), so this isn't a regression. Also note, that this will cause luci to overwrite a bad configuration as well. 2. ccs should not segfault when encountering a bad configuration, it should just display an error message and exit. But it's important to note that current ccs does not create a blank configuration if it doesn't understand the config, it just gives a less than helpful error. I'm duping this bug to split out the two bugs that need to be fixed.
Not this bug is not a regression, should this still get an exception or blocker flag?
Easy reproducer, start ricci & clustermon on node (in this example it's ask-03) [cfeist@gold ~]$ ccs -h ask-03 --createcluster testcluster [cfeist@gold ~]$ ccs -h ask-03 --getconf <cluster config_version="1" name="testcluster"> <fence_daemon/> <clusternodes/> <cman/> <fencedevices/> <rm> <failoverdomains/> <resources/> </rm> </cluster> ... Now on ask-03 remove the very first character of the file '<' ... [cfeist@gold ~]$ ccs -h ask-03 --getconf <cluster config_version="1" name="cluster"> <fence_daemon/> <clusternodes/> <cman/> <fencedevices/> <rm> <failoverdomains/> <resources/> </rm> </cluster> ... Notice how it responds with the cluster name 'cluster' instead of test cluster, because it has auto generated a new config file ...