1) Add a base class for all of the instance objects.
2) Normalize usage of logging.
3) General cleanups of ipa-server-install.
4) Make better use of httpinstance.
5) Add webguiinstance.
6) Improve progress reporting during installation.
Works Here (TM), but it would be nice to get someone else
to test since this moves code around a bit.
The current manpage installation gzips the files in
place and requests confirmation before overwriting
existing files. Add -f to prevent prompting. We
should consider not gzipping the files in place.
With the change to run autogen on make all if there
was no makefile present, it became impossible to
force the running of autogen when that is needed. Fix
that by adding a bootstrap-autogen target that checks
the existing of Makefiles and reverting the autogen
target to always run autogen.
Configure ipa servers as an ntp server and clients
to (by default) us the ipa server as an ntp server.
Also corrected the messages about which ports should
be opened.
This add replication setup through two new commands: ipa-replica-prepare
and ipa-replica-install. The procedure is to run ipa-replica-prepare
on an existing master. This will collect information about the realm
and the current master and create a file storing all of the information.
After copying that file to the new replica, ipa-replica-install is
run (with -r to create a read-only replica).
This version of the patch also includes fixes for the sasl mappings
on the replicas.
Remaining features:
- ssl for replication.
- automatic configuration of mesh topology for
master (or a simpler way to replicate multiple
masters.
- tool for view / configuring current replication.
After looking into setting up ntpd on the IPA servers I decided it
was better just to warn admins. There are just too many valid setups
for time synchronization for us to try to get this right. Additionally,
just installing ntp and accepting the default config will result in
a configuration that is perfectly valid for IPA.
This patch checks if ntpd is running and suggests enabling it if it
is not - for client and server. It also adds some suggested next
steps to the server installation.
Current ipa-python imports and calls code from ipaserver (which is in
the ipa-server package). This makes it impossible to use the admin
tools or the ipa-python package on a system without the server bits
installed. This fixes that in a fairly minimal way.
The use of a uuid for the DS instance name is overkill and it is a real
pain. This patch will use ipa-realm-name instead (resulting in something
like slapd-EXAMPLE-COM). All periods are converted to "-" because the DS
can't handle periods in server ids.
> > This largish patch makes the build and installation work on 64bit
> > machines. The only catch here is that to get a 64bit build you need to
> > set LIBDIR on make:
> >
> > make install LIBDIR=/usr/lib64
> >
> > The spec file does this correctly. I couldn't find any reliable way to
> > guess this that works both on real systems and in the almost entirely
> > empty rpm build root (you can't, for example, check for the existence
> > of /usr/lib64).
Here is another patch for the installer. It does a few things:
* use socket.getfqdn() but fallback to gethostname()
* streamlines the hostname prompting
* fixes a bunch of spelling and grammatical errors
* fixes a bug in the hostname reading/verification logic
* allows "yes" and "no" as answers
* modularizes and reuses code where possible
* changes some of the prompts to be more like
the FDS installer - some text is copied (which is easy to use IMO)
* tries to make the prompts fit on smaller screens (<80 chars)
Hope you agree that it is better. :)
Thanks,
Jon
> William Jon McCann wrote:
> > Hi,
> >
> > After playing with the install (repeatedly) I ended up with a lot of
> > duplicate values in:
> > /etc/sysconfig/dirsrv
> > /etc/sysconfig/ipa-kpasswd
> >
> > Here is a patch that should fix this. It modifies the file "in-place"
> > and removes lines that matching the key (or commented key) and then
> > appends the new key=value.
> >
> > Jon
>
> Cool, I've wanted to fix this for a while (and recently aborted a switch
> from open with "a" to "w").
>
> What happens if the file doesn't exist yet? Do we need to wrap the
> fileinput loop in either a try/except or just look to see if the file
> exists first (my vote)?
>
> Something like:
>
> def update_key_val_in_file(filename, key, val):
> if os.path.exists(filename):
> pattern = "^[\s#]*%s\s*=" % re.escape(key)
> p = re.compile(pattern)
> for line in fileinput.input(filename, inplace=1):
> if not p.search(line):
> sys.stdout.write(line)
> fileinput.close()
> f = open(filename, "a")
> f.write("%s=%s\n" % (key, val))
> f.close()
Good point. In genera,l I prefer doing a try because it is a little
less racy but in this case it doesn't make a difference.
Updated patch attached.
Thanks,
Jon
Install the turbogears web gui including an init script. This
patch includes a few related changes:
* create a production configuration
* rename the web gui startup scrip to ipa-webgui
* add an init script
* chkconfig on the ipa-webgui init script
* make the start script properly daemonize the app when not
in a development directory.
* Install everything to the correct places (/usr/sbin/ipa-webgui
and /usr/share/ipa/ipagui mainly).
There are some things still left to do:
* Sort out the logging - the config needs to be adjusted so
that logging messages end up in /var/log.
* Remove the rpmbuild tree with the dist-clean target.
* Move ipa-server-setupssl from /usr/sbin to /usr/share/ipa
* Check in requirement change for generated freeipa-python.spec
* Fix interactive hostname in ipa-server-install.
The default configuration of the apache selinux policy doesn't allow
apache to connect to the turbogears gui. This sets the correct
boolean to allow that connection.
removes the need for LDIF conversion. It will make TurboGears direct
code faster, but should keep xmlrpc about the same speed.
The patch also swaps out ldap.cidict for the IPA CIDict class. IPA code
should only use the CIDict class now.
