[Freeipa-devel] [PATCH] 262-265 Enable psearch by default
Rob Crittenden
rcritten at redhat.com
Fri Jun 8 15:10:52 UTC 2012
Martin Kosek wrote:
> On Tue, 2012-06-05 at 09:32 +0200, Martin Kosek wrote:
>> On Mon, 2012-06-04 at 23:49 -0400, Rob Crittenden wrote:
>>> Martin Kosek wrote:
>>>> On Fri, 2012-05-25 at 17:14 +0200, Martin Kosek wrote:
>>>>> On Fri, 2012-05-25 at 09:25 -0400, Rob Crittenden wrote:
>>>>>> Martin Kosek wrote:
>>>>>>> This set of patches handles enabling psearch both for new installations
>>>>>>> (patch 263) and upgraded IPA servers.
>>>>>>>
>>>>>>> For upgraded IPA servers I needed to make sure that psearch is not
>>>>>>> enabled for every IPA package update, but at most once, when a user
>>>>>>> updates to IPA with this patch for the first time (patch 264). This is
>>>>>>> enabled by a new State store located in /var/lib/ipa/sysupgrade (patch
>>>>>>> 262).
>>>>>>>
>>>>>>> I also improved the way we handled SELinux sebool updates (patch 265),
>>>>>>> this can make ipa-upgradeconfig to finish in 0.4 seconds and not in 150
>>>>>>> seconds as previously. Details are in the patches.
>>>>>>>
>>>>>>> Martin
>>>>>>
>>>>>> 262:
>>>>>> The sysupgrade directory isn't created by the RPM install:
>>>>>>
>>>>>> mkdir -p %{buildroot}/%{_localstatedir}/cache/ipa/sysupgrade
>>>>>
>>>>> Fixed.
>>>>>
>>>>>>
>>>>>> 263:
>>>>>>
>>>>>> It looks like zone_refresh is simply disabled in bindinstance.py, why
>>>>>> not remove it completely?
>>>>>
>>>>> zone_refresh is used by bindinstance.py. ipa-server-install or
>>>>> ipa-dns-install may be configured to use zone refresh instead of
>>>>> persistent search mechanism to update the zones (e.g. --zone-refresh
>>>>> 30).
>>>>>
>>>>>>
>>>>>> 264:
>>>>>>
>>>>>> Small nit, worth doing case-insensitive compare of psearch enabled status?
>>>>>
>>>>> Petr2 told me that arg value for boolean configuration option is
>>>>> case-insensitive, so we can do that - fixed.
>>>>>
>>>>>>
>>>>>> We're updating named.conf in place so I don't know that we need to reset
>>>>>> permissions. It at least shouldn't get modified by the write.
>>>>>
>>>>> Right, I was being too defensive. I removed the check.
>>>>>
>>>>> I made the upgrade more robust, now it won't crash for example when
>>>>> named.conf does not exist. I also made sure the upgrade script works
>>>>> correctly when the IPA is configured without DNS.
>>>>>
>>>>> Martin
>>>>
>>>> I rebased the patches for current master. I also slightly reworked patch
>>>> 265, the error message printed in case of an unsuccessful setsebool was
>>>> not printed right.
>>>>
>>>> Martin
>>>
>>> Trailing whitespace in 264:
>>>
>>> # git am /tmp/freeipa-mkosek-264-3-enable-psearch-on-upgrades.patch
>>> Applying: Enable psearch on upgrades
>>> /home/rcrit/redhat/freeipa-nossh/.git/rebase-apply/patch:108: trailing
>>> whitespace.
>>> root_logger.error('Cannot update connections in %s:
>>> %s',
>>> warning: 1 line adds whitespace errors.
>>
>> Fixed.
>>
>>>
>>> I don't think the DNS detection is adequate in 264, testing for
>>> named.conf is not enough. What if someone is running a non-IPA DNS
>>> server on the box?
>>
>> I assume you are referring to this line:
>> + if not bindinstance.named_conf_exists():
>>
>> It checks both if the named.conf exists + if it has bind-dyndb-ldap
>> configured for IPA:
>> if line.startswith('dynamic-db "ipa"'):
>>
>>>
>>> I know that I've recently done similar config changes but in 265 is
>>> using line.startswith() going to be fragile?
>>
>> I assume you mean patch 264. This should be OK - user would need to mess
>> with the configuration generated by our install scripts to break it. But
>> in this case, other regex-es would fail too. I did not want to get too
>> wild with regex-es to keep it simple and safe. The worst case scenario
>> should be that named.conf is not updated and psearch is not turned on.
>>
>>>
>>> In 266 I'd merge in the ipa-upgradeconfig change into 265 or some other
>>> patch.
>>
>> I assume you mean patch 265. I had this change moved to 264 right after
>> I sent the patches :-)
>>
>>>
>>> In the 'for setting, state' loop should it be catching a
>>> CalledProcessException rather than raw Exception? I think that is all
>>> that should be raised there.
>>
>> Right, fixed.
>>
>>>
>>> I did an upgrade and it seemed to work ok, ended up with these scary
>>> messages in /var/log/messages:
>>>
>>> Jun 4 23:39:17 localhost named[18753]: LDAP error: Can't contact LDAP
>>> server
>>> Jun 4 23:39:17 localhost named[18753]: connection to the LDAP server
>>> was lost
>>> Jun 4 23:39:17 localhost named[18753]: bind to LDAP server failed:
>>> Can't contact LDAP server
>>> Jun 4 23:39:17 localhost named[18753]: ldap_psearch_watcher failed to
>>> handle LDAP connection error. Reconnection in 60s
>>> Jun 4 23:39:17 localhost named[18753]: LDAP error: Can't contact LDAP
>>> server
>>> Jun 4 23:39:17 localhost named[18753]: connection to the LDAP server
>>> was lost
>>> Jun 4 23:39:17 localhost named[18753]: bind to LDAP server failed:
>>> Can't contact LDAP server
>>> Jun 4 23:39:17 localhost ns-slapd[18798]: [04/Jun/2012:23:39:17 -0400]
>>> - Information: Non-Secure Port Disabled
>>> Jun 4 23:40:17 localhost named[18753]: handle_connection_error failed
>>> to obtain ldap error code
>>> Jun 4 23:40:17 localhost named[18753]: connection to the LDAP server
>>> was lost
>>> Jun 4 23:40:17 localhost named[18753]: bind to LDAP server failed:
>>> Can't contact LDAP server
>>> Jun 4 23:40:17 localhost named[18753]: ldap_psearch_watcher failed to
>>> handle LDAP connection error. Reconnection in 60s
>>> Jun 4 23:41:17 localhost named[18753]: handle_connection_error failed
>>> to obtain ldap error code
>>> Jun 4 23:41:17 localhost named[18753]: connection to the LDAP server
>>> was lost
>>>
>>> DNS does seem to be working fine from the cli.
>>
>> I think this was caused by ipa-ldap-updater which shut down the
>> Directory Server to perform the LDAP upgrade.
>>
>> Btw I asked Petr to file a ticket for bind-dyndb-ldap to report when it
>> report success after when it returns back from an error state:
>> https://fedorahosted.org/bind-dyndb-ldap/ticket/71
>> This way, we cannot know that the LDAP connection has been restored
>> besides doing a test DNS query.
>>
>>>
>>> The tests are another matter. named crashed in 0:1.1.0-0.10.b2.fc17 in
>>> the test cleanup.
>>>
>>> I upgraded to bind-dyndb-ldap-1.1.0-0.11.rc1.fc17 and got this stack trace:
>>>
>>> Program received signal SIGABRT, Aborted.
>>> [Switching to Thread 0x7f68e50db700 (LWP 19367)]
>>> 0x00007f68e6188915 in raise () from /lib64/libc.so.6
>>> (gdb) where
>>> #0 0x00007f68e6188915 in raise () from /lib64/libc.so.6
>>> #1 0x00007f68e618a0c8 in abort () from /lib64/libc.so.6
>>> #2 0x00007f68e91171fb in assertion_failed (file=<optimized out>,
>>> line=<optimized out>, type=<optimized out>, cond=<optimized out>)
>>> at ./main.c:219
>>> #3 0x00007f68e73a6c3a in isc_assertion_failed (
>>> file=file at entry=0x7f68e8a82deb "zone.c", line=<optimized out>,
>>> type=type at entry=isc_assertiontype_require,
>>> cond=cond at entry=0x7f68e8a82fe7 "zone->db != ((void *)0)")
>>> at assertions.c:57
>>> #4 0x00007f68e8a2ba67 in zone_detachdb (zone=<optimized out>) at
>>> zone.c:12944
>>> #5 zone_detachdb (zone=0x7f68dc57fef0) at zone.c:12943
>>> #6 0x00007f68e8a2baa1 in zone_unload (zone=zone at entry=0x7f68dc57fef0)
>>> at zone.c:9092
>>> #7 0x00007f68e8a2fcc4 in dns_zone_unload (zone=0x7f68dc57fef0) at
>>> zone.c:9040
>>> #8 0x00007f68e3584b9e in ldap_delete_zone2
>>> (inst=inst at entry=0x7f68e90b0f10,
>>> name=name at entry=0x7f68e50dad10, lock=lock at entry=isc_boolean_true)
>>> at ldap_helper.c:786
>>> #9 0x00007f68e3586554 in ldap_delete_zone (dn=<optimized out>,
>>> inst=0x7f68e90b0f10, lock=<optimized out>) at ldap_helper.c:811
>>> #10 update_action (task=<optimized out>, event=0x7f68e37de6a0)
>>> at ldap_helper.c:2763
>>> #11 0x00007f68e73c613e in dispatch (manager=0x7f68e908f010) at task.c:1109
>>> #12 run (uap=0x7f68e908f010) at task.c:1279
>>> #13 0x00007f68e6d7bd14 in start_thread () from /lib64/libpthread.so.0
>>> #14 0x00007f68e624494d in clone () from /lib64/libc.so.6
>>>
>>> rob
>>
>> Thanks for digging out the traceback, I already reported this error to
>> bind-dyndb-ldap:
>> https://bugzilla.redhat.com/show_bug.cgi?id=827401
>>
>> Petr, what's the status of this bug? I guess we cannot push this set of
>> patches to enable the psearch by default until this is fixed. Otherwise
>> bind-dyndb-ldap would crash _every_ DNS unit test case.
>>
>> Updated set of patches attached.
>>
>> Martin
>
> Petr^2 fixed the bug in bind-dyndb-ldap causing it to crash during DNS
> unit tests. A re-tested the new version with IPA and it worked fine.
>
> Attached a rebased set of patches with proper bind-dyndb-ldap version
> enforced. I would like this to get acked soon so that psearch is tested
> by a broader audience and we are able to stabilize it faster.
>
> Martin
These work ok so conditional ACK based on the following:
The tests all pass but I saw this in messages:
Jun 8 10:13:37 localhost named[1624]: psearch moddn change is not
implemented
Jun 8 10:13:37 localhost named[1624]: psearch_update failed for
idnsname=testdnsres-renamed,idnsname=dnszone.test,cn=dns,dc=example,dc=com
zone. Zone can be outdated, run `rndc reload`
Other than immediately seeing new zones are there any other consequences
to disabling psearch? Are features eventually going to not be available
if it is not enabled? I assume that if/when that happens the man page
will be updated at that point?
rob
More information about the Freeipa-devel
mailing list