[ovirt-users] Testday: json rpc

Antoni Segura Puimedon asegurap at redhat.com
Thu Jul 31 08:14:47 EDT 2014



----- Original Message -----
> From: "Antoni Segura Puimedon" <asegurap at redhat.com>
> To: devel at ovirt.org, "users" <users at ovirt.org>
> Sent: Wednesday, 30 July, 2014 2:15:36 PM
> Subject: Testday: json rpc
> 
> Hi fellow oVirters,
> 
> On this test day I picked JSON RPC for testing:
> 
> My initial environment consisted of a hosted engine setup with three hosts.
> 
> Upgrading
> ===========
> The first task, then, was to upgrade everything to 3.5.
> 
> My guide was http://www.ovirt.org/Hosted_Engine_Howto#Upgrade_Hosted_Engine
> 
> One issue that I encountered was that after setting the 3.5 pre-release yum
> repo on the engine VM and doing yum update, when doing
>     engine-setup
> To make it handle the upgrade, that failed complaining about missing a
> package
> called patternfly1. I remembered that on the list Greg Sheremeta mentioned a
> copr repository for it, so I went ahead, installed it and repeated the
>     engine-setup
> 
> This time it succeeded, however I feel like patternfly1 should probably be an
> ovirt-engine-3.5 dependency and it should be in the ovirt 3.5 repository.
> 
> I also noticed that after upgrading the hosts the amount of free system
> memory
> is much lower, while the VMs continue to run fine. I opened:
>     https://bugzilla.redhat.com/1124451
> 
> Another thing that happened was that restarting ovirt-ha-agent and
> ovirt-ha-broker using systemd fails silently and it says that they were
> killed.
> I only got it to work again by doing:
> 
>     /lib/systemd/systemd-ovirt-ha-broker restart
>     /lib/systemd/systemd-ovirt-ha-agent restart
> 
> However, and obviously, doing it like that escapes from the advisable control
> of systemd. Another bad thing is that after doing all this we get that the
> current host (which has the engine running displays):
> 
>     --== Host 2 status ==--
> 
>     Status up-to-date                  : False
>     Hostname                           : 10.34.63.180
>     Host ID                            : 2
>     Engine status                      : unknown stale-data
>     Score                              : 2000
>     Local maintenance                  : False
>     Host timestamp                     : 1406640293
>     Extra metadata (valid at timestamp):
>          metadata_parse_version=1
>          metadata_feature_version=1
>          timestamp=1406640293 (Tue Jul 29 15:24:53 2014)
>          host-id=2
>          score=2000
>          maintenance=False
>          bridge=True
>          cpu-load=0.0856
>          engine-health={"health": "good", "vm": "up", "detail": "up"}
>          gateway=True
>          mem-free=3192
> 
> Why did the engine status go to unknown stale-data? (it also happened for the
> other two hosts in the setup.

After discussing with Jiři: the 'stale data thing' resulted in
    https://bugzilla.redhat.com/1125244
> 
> Changing hosts to use JSON RPC
> ===============================
> 
> After talking with Piotr and trying it with the webadmin UI, I found out that
> there is no direct way to update a host's settings to use json rpc as its
> connectivity mechanism with the engine. This constitutes a usabiltiy bug
> which I filed:
> 
>     https://bugzilla.redhat.com/1124442
> 
> It is presumable that users will want to move to the newer and better RPC
> mechanism and they should be able to do so by merely putting the host in
> maintenance and ticking some checkbox in the 'edit host' dialog.
> 
> I did the workaround of removing the host and adding it again and that
> worked,
> the host went up.
> 
> Network operations via JSON RPC
> ================================
> 
> After the host went up, I decided to send a setupNetworks command. The
> operation worked out fine, but unfortunately we have a very serious gap that
> makes it impossible for me to use jsonrpc for network operations/development.
> 
> **Logging**
> 
> When doing a network operation with xmlrpc we'd get the following in vdsm.log
>     Thread-21::DEBUG::2014-07-30
>     13:38:11,414::BindingXMLRPC::1127::vds::(wrapper) client
>     [10.34.61.242]::call setupNetworks with ({'10': {'nic': 'em2', 'vlan':
>     '10', 'STP': 'no', 'bridged': 'true', 'mtu': '1500'}}, {},
>     {'connectivityCheck': 'true', 'connectivityTimeout': 120}) {} flowID
>     [686033d4]
>     Thread-21::DEBUG::2014-07-30
>     13:38:32,689::BindingXMLRPC::1134::vds::(wrapper) return setupNetworks
>     with {'status': {'message': 'Done', 'code': 0}}
> 
> As you can see, we get the bare minimum logging one could ask for, an entry
> with the command called and the data it received and another entry with the
> return result data.
> 
> Doing the same with jsonrpc (and ignoring the excessive IOProcess) if I
> search
> for "setupNetworks" the only thing I get is:
>     Thread-23057::DEBUG::2014-07-30
>     13:32:44,126::__init__::462::jsonrpc.JsonRpcServer::(_serveRequest)
>     Looking for method 'Host_setupNetworks' in bridge
> 
> And if I search for the data received, like 'STP', there is nothing
> whatsoever.
> As I said, unless this is fixed and we get entries with the same amount of
> data
> as before, it can't be used in production nor in development.
> 
>     https://bugzilla.redhat.com/1124813
> 
> 
> TL;DR:
> - lack of usability upgrading an environment to use jsonrpc
>   https://bugzilla.redhat.com/1124442
> - Failure on step 7 of upgrade steps:
>   https://bugzilla.redhat.com/1124826
> - JSON RPC logging excessive but insuficient for network call debugging
>   https://bugzilla.redhat.com/1124813
> 


More information about the Users mailing list