Our cloud went down today and we realized that only very few of us actually know how it works and how to repair it.
To address this, Kevin has agreed to teach us what he knows about it. We're looking to set up a #fedora-classroom meeting, have zodbot log it, and afterwards turn the log into an infra-docs document.
If you're interested in attending and asking questions, please fill out the following survey so we can find the best time:
http://whenisgood.net/cloud-halp
Please note that you need to set your timezone for that questionnaire to work.
-Ralph
Thanks for setting this up, Ralph. I hope to be able to pick up a few tidbits for the work I'll be doing for the Hat this summer.
David Gay http://oddshocks.com/
On 03/18/2014 10:33 AM, Ralph Bean wrote:
Our cloud went down today and we realized that only very few of us actually know how it works and how to repair it.
To address this, Kevin has agreed to teach us what he knows about it. We're looking to set up a #fedora-classroom meeting, have zodbot log it, and afterwards turn the log into an infra-docs document.
If you're interested in attending and asking questions, please fill out the following survey so we can find the best time:
http://whenisgood.net/cloud-halp
Please note that you need to set your timezone for that questionnaire to work.
-Ralph
_______________________________________________ infrastructure mailing list infrastructure@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/infrastructure
On Tue, Mar 18, 2014 at 10:33:34AM -0400, Ralph Bean wrote:
Our cloud went down today and we realized that only very few of us actually know how it works and how to repair it.
To address this, Kevin has agreed to teach us what he knows about it. We're looking to set up a #fedora-classroom meeting, have zodbot log it, and afterwards turn the log into an infra-docs document.
If you're interested in attending and asking questions, please fill out the following survey so we can find the best time:
http://whenisgood.net/cloud-halpPlease note that you need to set your timezone for that questionnaire to work.
-Ralph
Okay, based on the results of the survey: let's do Tuesday March 25th at 18:00UTC in #fedora-classroom. Not everyone can make it then, but that's the time when the most people can.
I've marked it on the infrastructure calendar: https://apps.fedoraproject.org/calendar/list/infrastructure/
================================================================== #fedora-classroom: Infrastructure Private Cloud Class (2014-03-25) ==================================================================
Meeting started by nirik at 18:00:03 UTC. The full logs are available at http://meetbot.fedoraproject.org/fedora-classroom/2014-03-25/infrastructure-...
Meeting summary --------------- * intro (nirik, 18:00:03)
* History/current setup (nirik, 18:02:51) * LINK: https://fed-cloud02.cloud.fedoraproject.org/dashboard/ (nirik, 18:10:48) * LINK: http://infrastructure.fedoraproject.org/cgit/ansible.git/tree/README has a bunch of cloud specific info. (nirik, 18:11:35) * IDEA: add openid support to openstack's horizon dashboard (threebean, 18:13:19) * LINK: https://wiki.openstack.org/wiki/Nova_openid_service (danofsatx-work, 18:17:06)
* Upcoming plans / TODO (nirik, 18:26:33) * LINK: https://en.wikipedia.org/wiki/OpenStack#Components (threebean, 18:27:00)
* Open Questions (nirik, 18:40:57) * LINK: https://fedoraproject.org/wiki/Infrastructure_private_cloud (nirik, 18:41:36)
Meeting ended at 19:02:55 UTC.
People Present (lines said) --------------------------- * nirik (143) * mirek-hm (29) * danofsatx-work (21) * threebean (19) * tflink (16) * smooge (5) * zodbot (3) * abadger1999 (3) * webpigeon (3) * jsmith (3) * jamielinux (1) * blob (1) * relrod (1) * janeznemanic (1)
18:00:03 <nirik> #startmeeting Infrastructure Private Cloud Class (2014-03-25) 18:00:03 <zodbot> Meeting started Tue Mar 25 18:00:03 2014 UTC. The chair is nirik. Information about MeetBot at http://wiki.debian.org/MeetBot. 18:00:03 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic. 18:00:03 <nirik> #meetingname infrastructure-private-cloud-class 18:00:03 <nirik> #topic intro 18:00:03 <zodbot> The meeting name has been set to 'infrastructure-private-cloud-class' 18:00:22 <nirik> hey everyone. who is around for a bit of talking about fedora's infrastructure private cloud? 18:00:26 * threebean is here 18:00:32 <smooge> hello 18:00:38 <danofsatx-work> aqui! aqui! 18:00:44 * mirek-hm is here 18:00:45 * tflink is here 18:00:53 * blob is here 18:00:57 * relrod here 18:01:01 * webpigeon is here 18:01:11 * jamielinux is here out of curiousity 18:01:53 * abadger1999 is here 18:01:57 <janeznemanic> hi 18:02:24 <nirik> cool. ;) ok, I thought I would give a bit of background/history first, then a bit about ansible integration, then talk about plans... 18:02:51 <nirik> #topic History/current setup 18:02:58 <threebean> that sounds good. can we interrupt you with questions? or should be hold them for later? 18:03:10 <nirik> threebean: please do... questions are good. ;) 18:03:16 <threebean> cool, cool. 18:03:20 * jsmith is late 18:03:23 <nirik> so, our current setup is a openstack folsom cloud 18:03:31 <nirik> It was mostly manually installed by me. 18:03:38 <nirik> Ie, I installed rpms, ran setup, etc... 18:03:55 <nirik> there are currently 7 nodes in use 18:04:07 <jsmith> Nodes = servers? 18:04:08 <nirik> 1 (fed-cloud02.cloud) is the 'head node' 18:04:15 <nirik> 6 are compute nodes 18:04:20 <nirik> jsmith: yeah, physical boxes. 18:04:37 <jsmith> Perfect, just wanted to make sure I was getting the nomenclature right 18:04:53 <nirik> The compute notes only run openstack-nova-compute, and one of them also runs cinder (the storage service will get to that in a few) 18:05:04 <nirik> the head node runs all the other stuff. 18:05:18 <nirik> and acts as the gateway to all the other things. 18:05:34 <nirik> it runs network, the db (mysqld), amqp, etc, etc. 18:05:48 <nirik> It also runs cinder for storage. 18:05:50 <mirek-hm> head node is fed-cloud02.cloud.fedoraproject.org BTW 18:06:29 <nirik> when you fire off an instance openstack looks around and schedules it on a compute node. 18:06:34 <tflink> outside of cinder, is there any shared storage? 18:07:07 <nirik> when you ask for a persistent volume, it allocates it from one of the cinder servers... it makes a lv in a pool and shares it via iscsi 18:07:14 <nirik> tflink: nope. Not currently. 18:07:45 <mirek-hm> aha, that are VG cinder-volumes in lvs output 18:07:46 <nirik> all storage is either cinder from 02 or cinder from 08... or each compute node has a small amount of storage it uses to cache images locally. 18:07:56 <nirik> mirek-hm: yeah. 18:08:04 <tflink> is that a requirement of newer openstack? when I set up my dev/standalone system, I got the impression that something like gluster was highly reccomended for more than 1 mode 18:08:05 <nirik> lets see...what elese. 18:08:24 <nirik> tflink: we tried gluster, but it was really really really really slow at the time. 18:08:42 <nirik> oh, networking: 18:08:44 <danofsatx-work> no swift for image service? 18:09:01 <mirek-hm> nirik: you forget on one exception, Copr-be have one storage allocated on one compute node (800 GB) 18:09:02 <nirik> we have a number of tenants (projects). Each one gets it's own vlan. 18:09:18 <nirik> mirek-hm: thats from the 08 compute node via cinder there. ;) 18:09:44 <nirik> we have a /24 network for external ips. Each instance gets one by default and can also allocated a static one. 18:10:08 <nirik> danofsatx-work: yeah, we do have swift. 18:10:17 <nirik> it uses storage on fed-cloud02. 18:10:34 <nirik> There's a dashboard available for web (horizon): 18:10:48 <nirik> https://fed-cloud02.cloud.fedoraproject.org/dashboard/ 18:11:04 <nirik> The main way we interact tho is via ansible on lockbox01. 18:11:35 <nirik> http://infrastructure.fedoraproject.org/cgit/ansible.git/tree/README has a bunch of cloud specific info. 18:11:45 <nirik> basically we use the ansible ec2 module to spin things up, etc. 18:11:48 <webpigeon> does the web interface work for appentices 18:12:04 <danofsatx-work> fi-apprentice group can't login to dashboard :( 18:12:10 <nirik> webpigeon: it doesn't. It also doesn't interact with our other authentication at all. ;( it's all manually needing to add people. 18:12:10 <mirek-hm> webpigeon: no 18:12:25 <nirik> I keep hoping someday they will add openid support. 18:12:32 * webpigeon wonders if he can get access to have a look :) 18:12:47 <danofsatx-work> I thought I saw that mentioned in the Havana release notes..... 18:12:49 <nirik> sure, we could probibly do that. :) 18:12:53 <mirek-hm> you either have to have account there or you can root login to fed-clooud02 and there is admin password stored in keystonerc file 18:12:55 <nirik> danofsatx-work: oh? excellent. 18:13:19 <threebean> #idea add openid support to openstack's horizon dashboard 18:13:20 <nirik> ok, lets see what else on current setup... oh, we have some playbooks that make transient instances... 18:13:39 <nirik> so it can spin up one and then ansible configures it all for testing something. 18:13:52 <nirik> we also have persistent instances where it's for a specific use. 18:14:31 <nirik> we picked the ec2 interface over the nova one because we wanted to be portable to euca or amazon... we could possibly revisit that, but it seems to do most of what we need. 18:15:06 <nirik> oh, we also hacked onto fed-cloud02 a nginx proxy to proxy https for us. By default folsom had it's ec2 interface http 18:15:42 <nirik> also, one compute node has been down with a bad disk, but should be ready to re-add again now. I will likely do that in the next few days. 18:16:09 <nirik> So, any questions on the current setup? or shall I talk plans and then open up to more questions? 18:16:31 <threebean> so, last week the cloud blew up because... one of the compute nodes got the ip of the router? 18:16:35 <threebean> where did you look to figure that out? 18:16:38 <mirek-hm> if you are logged to fed-cloud2 and want to check status of services you may run openstack-status which should give such output http://fpaste.org/88542/57713621/ 18:17:06 <nirik> threebean: yes. I looked on fed-cloud02 with nova-manage at the ips... or perhaps I saw the .254 one in the dashboard first 18:17:06 <danofsatx-work> https://wiki.openstack.org/wiki/Nova_openid_service 18:18:03 <mirek-hm> openstack services are (re)stared by /etc/init.d/openstack-* init.d files 18:18:06 <nirik> nice! 18:18:13 <nirik> danofsatx-work: thanks for the link//info. 18:18:36 <danofsatx-work> np ;) - I'm teaching an Openstack class at uni right now 18:18:45 <threebean> cool. what group(s) of people are allowed to login to fed-cloud02 (and the others?) its all locally managed? 18:19:14 <nirik> threebean: yes, it's all local. sysadmin-main should have their keys there for root (there arent any local users I don't think). Also a few other folks like mirek-hm 18:19:36 <nirik> the cloud network is completely isolated from our other stuff and rh's internal networks. 18:19:50 <nirik> so if you need to talk to it you have to go via the external ips... 18:20:21 <mirek-hm> nirik: btw I will remove seth ssh key from authorizes_keys 18:20:27 <nirik> mirek-hm: ok. ;( 18:20:44 <tflink> does that have anything to do with the network goofiness where instances can't always talk to eachother on non-public IPs? 18:20:44 <nirik> on to a bit of plans? 18:20:50 <threebean> one more Q 18:20:53 <nirik> tflink: nope, thats a openstack issue. ;) 18:20:56 <threebean> how close to full capacity are we on the compute nodes? 18:20:58 <tflink> ok, wasn't sure 18:21:07 <nirik> it's due to the way that they setup firewalls. ;( 18:21:26 <nirik> instances on different compute nodes can talk to each other fine, but if they happen to be on the same one they cannot. 18:21:34 <nirik> I don't know if thats fixed in newer openstack. 18:21:49 <nirik> threebean: hard to say. ;) There's not any easy "show me my capacity" 18:22:03 <threebean> heh, ok. 18:22:11 <nirik> each compute node reports on itself... 18:22:20 <threebean> perhaps we need a scripts/cloud-info script in our ansible repo. 18:22:20 <nirik> so fed-cloud02 (which is also a compute node) says: 18:22:27 <nirik> 2014-03-25 18:21:44 30752 AUDIT nova.compute.resource_tracker [-] Free ram (MB): 10732 18:22:27 <nirik> 2014-03-25 18:21:44 30752 AUDIT nova.compute.resource_tracker [-] Free disk (GB): 951 18:22:28 <nirik> 2014-03-25 18:21:44 30752 AUDIT nova.compute.resource_tracker [-] Free VCPUS: -4 18:22:57 <mirek-hm> also routing in fedora cloud is little bit tricky as external IP does not work inside of cloud. Therefore if you comunicate from cloud machine to another cloud machine you must use internal IP. 18:23:10 * nirik nods. 18:23:26 <tflink> mirek-hm: that's not always 18:23:27 <tflink> true 18:23:32 <nirik> also, you get automatically a external ip with each instance, but it's dynamically assigned. 18:23:49 <tflink> but I've had trouble figuring out a way to always let instances talk with eachother 18:23:49 <nirik> so if you want a static/known external you have to assign one, and then you have 2 of them. ;( 18:24:16 <nirik> and you cannot give up the dynamic one. 18:24:33 <mirek-hm> nirik: what command you used to get that resource information? 18:25:00 <nirik> threebean: there is a newer project called celiometer or something that is supposed to give you usage info, but thats not available in the old version we have. 18:25:10 <tflink> a way to consistently let instanced talk to eachother over the network, rahter. it's always been a one-off "does this work with instances X and Y" 18:25:10 <nirik> mirek-hm: tail /var/log/nova/compute.log 18:25:24 <danofsatx-work> I thought that was heat, not celiometer..... 18:25:28 * danofsatx-work checks real quick 18:25:35 <mirek-hm> it is celiometer 18:25:48 <mirek-hm> heat is for configuration 18:25:48 <tflink> heat is coordination and setup of instances, no? 18:26:24 <danofsatx-work> ok, yeah....It's the Telemetry service. 18:26:33 <nirik> #topic Upcoming plans / TODO 18:26:48 <danofsatx-work> openstack isn't real good about actually giving the name of each service, you have to dig for it. 18:27:00 <threebean> https://en.wikipedia.org/wiki/OpenStack#Components 18:27:16 <nirik> so, we have a few new machines we are getting in soon (3). My plan is to try and set those up as a seperate cloud, ideally via ansible 18:27:32 <nirik> so we can easily rebuild it from the ground unlike the current one. 18:27:42 * danofsatx-work likes that plan 18:27:53 <nirik> also, if we can do openid that would be lovely. 18:27:54 <danofsatx-work> makes for a cleaner, easier transition 18:28:16 <nirik> once its all working, we could then migrate things over until all is moved 18:28:33 <mirek-hm> yes, becouse if somebody or something break current cloud, we have to manualy rebuild it, which would last ages. 18:28:34 <nirik> (and move compute nodes as needed) 18:28:58 <nirik> sadly, I haven't been able to find a good ansible recipe for setting up a cloud, but I can look some more. 18:29:18 <nirik> it would be nice to just have it in our regular ansible repo with everything else. 18:29:18 <mirek-hm> can we use TripleO for installation? 18:29:33 <nirik> mirek-hm: not sure. I have heard of that, but wasn't clear what it did. 18:29:40 <mirek-hm> I will ask collegues if that can be scripted 18:30:11 <mirek-hm> nirik: it is installation of OpenStack using OpenStack which use OpenStack :) 18:30:19 <nirik> cool. That would be nice. The way I read it is that it needs a cloud installed already? 18:30:34 <danofsatx-work> what about RH's own product, packstack? 18:30:38 <mirek-hm> you first start with disk image, which you boot, then it install undercloud and then it install normal cloud 18:30:59 <nirik> triple-o also looks a bit experemental. 18:31:03 <nirik> danofsatx-work: thats another option yeah. 18:31:15 <nirik> it's puppet/chef, but if we have to... 18:31:47 <mirek-hm> TripleO at devconf http://www.youtube.com/watch?v=Qcpe2gjdRz0&list=PLjT7F8YwQhr928YsRxmOs8h... 18:32:51 <nirik> Also on upcoming plans... in Q4 we are supposed to get some dedicated storage for the cloud... 18:33:23 <threebean> for the new 3-node cloud.. do you expect we would move to a more modern openstack release? 18:33:31 <mirek-hm> nirik: how big? 18:33:40 <threebean> looks like icehouse is supposed to be released in ~2 weeks. 18:33:52 <nirik> threebean: definitely. 18:34:11 <nirik> mirek-hm: unknown yet, as much as we can get for the money. ;) Ideally something thats 2 units in a HA... 18:35:15 <nirik> also in Q3 we are getting some more compute nodes. 18:36:00 <abadger1999> nirik: What's our plans for what we're oging to use the cloud for? (like -- are we going to move more and more services over to it?) 18:36:23 <nirik> abadger1999: currently, I think it's a good fit for some things and not for others. 18:36:31 <nirik> devel and test instances -> definitely 18:36:34 <nirik> coprs -> yep 18:37:04 <nirik> our main applications -> not sure... I like the control we have currently with them 18:37:10 <abadger1999> <nod> 18:37:13 <mirek-hm> nirik: extrapolated data say that copr will run out of disk space on christmass, so Q3 is really last date 18:37:32 <nirik> mirek-hm: well, can you split into 2 volumes? 18:37:49 <nirik> all the other compute nodes have similar space... so we can fire up cinder on them and add more 800GB volumes. 18:38:05 <mirek-hm> nirik: probably, if I can concate them using lvm 18:38:32 <danofsatx-work> is the old cloud hardware being rolled into the new instance, or decommissioned? 18:38:33 <nirik> not sure if that will work, but we can see... 18:38:40 <nirik> danofsatx-work: rolled into the new one. 18:38:51 <nirik> it's got a few more years on it's warentee 18:40:11 <nirik> mirek-hm: we could also try and move the storage to Q3 and move the new compute nodes to Q4. that might be best 18:40:13 <nirik> (if we can do it) 18:40:57 <nirik> #topic Open Questions 18:41:22 <nirik> So, we have a wiki page for our cloud, but it's pretty out of date. We should likely revamp it 18:41:36 <nirik> https://fedoraproject.org/wiki/Infrastructure_private_cloud 18:41:47 <nirik> That has some more older long term plans, etc. 18:42:10 <mirek-hm> will we add some arm64 machine to cloud as they arrive? 18:42:51 <nirik> I'd like to add one there yeah... 18:42:52 <tflink> I assume that the plan going forward is to retain network isolation from the rest of infra? 18:42:57 <nirik> tflink: yep. 18:43:21 <nirik> that has some downsides... 18:43:39 <nirik> but overall I think it's very good given the carefree nature of instances. 18:43:52 <tflink> yeah, I think it's a positive thing overall 18:43:53 <nirik> A few other thoughts to toss out for comment: 18:44:07 <tflink> just not for all my possible use cases :) 18:44:22 <nirik> 1. Should we continute to autoassign a external ip on the new cloud? or require you to get one if you need it? 18:44:29 <nirik> it has advantages/disadvantages 18:44:55 <tflink> how many instances are using non-floating IPs right now? 18:45:04 <mirek-hm> can I reach wild internet when I have internal IP only (e.g via NAT)? 18:45:35 <nirik> mirek-hm: not sure actually. 18:45:38 <danofsatx-work> EC2 only assigns (static) public ip's when asked for, and it's the only IP you get 18:45:53 <nirik> danofsatx-work: we have it set to autoassign one. 18:45:59 <nirik> (in the current cloud) 18:46:16 <nirik> auto_assign_floating_ip = True 18:46:25 <nirik> tflink: I'm not sure how to tell. ;( 18:46:26 <mirek-hm> danofsatx-work: EC2 instance have both, internal and externa IP 18:47:04 <danofsatx-work> I understand, I'm trying to figure out how to translate my thoughts into coherent IRC language ;) 18:48:16 <danofsatx-work> ok, scratch my last....it's not making sense anymore in my head :( 18:48:39 <nirik> there's 101 external ips being used right now. 18:49:20 <threebean> heh, how did you figure that out? :P 18:49:41 <nirik> and 34 instances seem to have 2 external ips. 18:50:07 <nirik> nova-manage floating list | grep 209 | awk '{print $3 " " $2 }' | grep -v None | wc -l 18:50:22 <threebean> great, thanks. 18:50:23 <nirik> ugly, but should be right. 18:50:24 <tflink> ah, not as many as I suspected 18:50:47 <nirik> so, I guess we should see if instances with no external can reach out any. I'm not sure they can. 18:50:55 <nirik> but we can test that on the new cloud. 18:51:32 <danofsatx-work> unless there's a router set up to route the internal traffic out, no they can't get out. 18:51:45 <nirik> question 2. We have been very lax about maint (mostly because this current cloud is so fragile) would folks be ok with a more practive one on the new cloud? ie, more frequent reboots/maint windows? 18:52:14 <nirik> danofsatx-work: all the instances would hit the head node, but I don't know if nova-network will nat them or not. 18:52:19 <danofsatx-work> I would be, but I come from a different background than the rest of y'all 18:52:55 <threebean> yeah, that would be fine. I worry that I'm responsible for wasted resources somewhere, some node I forgot about. 18:53:48 <threebean> erm, I misinterpreted the question.. for some reason I read it as being more proactive about cleanup. 18:53:48 <nirik> last time we rebooted stuff we were able to actually suspend instances... and mostly they came back ok 18:53:56 <nirik> that too. :) 18:54:16 <nirik> more reporting would be nice... like a weekly snapshot of 'here's all running instances' 18:54:24 <nirik> sometimes it's hard to tell what an instance was for tho 18:54:52 <mirek-hm> nirik: +1 to keep auto_assign_floating_ip = True; +1 to planned outages on ne Fedora Cloud 18:54:57 <threebean> oo, send some numbers to collectd? 18:55:53 <tflink> re: more proactive maintenance - I'd like to see coordination with standard backup times. ie - run shortly after backup runs so taht any possible data-loss would be minimized 18:55:57 <nirik> yeah, that might be nice. However, no vpn, so not sure it could talk to log02. 18:56:00 <smooge> well getting that data is hard because of the nat 18:56:12 <nirik> or it could run it's own I guess. 18:56:24 <smooge> I would say we might want to look at having a system on that network which could do that for us 18:56:40 <smooge> the various things that we would like but can't because of seperate networks 18:56:52 <nirik> actually, it could be a cloud instance even. ;) 18:57:20 <mirek-hm> March 2014: Active Instances: 66 Active RAM: 352GB This Month's VCPU-Hours: 40433.81 This Month's GB-Hours: 2067464.96 18:57:29 <smooge> well it might be nice if the box was able to run when the cloud wasn't 18:57:42 <nirik> smooge: details, details. 18:57:49 <nirik> mirek-hm: thats only one tennant tho right? 18:58:23 <mirek-hm> nirik: that is from dashboard overview of admin user, so I would say that it count everything 18:58:24 <nirik> tflink: agreed. 18:58:46 <nirik> mirek-hm: I think it's only the project it has active at the time... if you change that the numbers change right? 18:59:12 <mirek-hm> yes 18:59:24 <nirik> so we would need to sum all those. 19:00:58 <nirik> ok, any other questions or comments? 19:01:02 <nirik> or shall we wrap up? 19:01:13 * nirik is happy to answer anything outside meeting, etc. 19:01:31 <danofsatx-work> any room for an apprectice on the cloud team? 19:02:03 <nirik> sure. always room for help... perhaps assistance setting up the new cloud? 19:02:31 <danofsatx-work> yeah, I can do that...I need to learn ansible, and I am building clouds at school and work anyhow ;) 19:02:39 <nirik> excellent. ;) 19:02:50 <nirik> ok, thanks for coming everyone... lets continue over in #fedora-admin... 19:02:55 <nirik> #endmeeting
On 03/25/2014 08:12 PM, Ralph Bean wrote:
Meeting started by nirik at 18:00:03 UTC. The full logs are available at http://meetbot.fedoraproject.org/fedora-classroom/2014-03-25/infrastructure-...
I promised to ask OpenStack guys two guestions. Here are the answers:
1) Can we put Arm64 into OpenStack? No. Not yet supported.
2) Can be TripleO scripted? Yes, there already exist puppet modules - he was aware of: https://github.com/agroup/tripleo-puppet-elements I asked uncle Google and said that ansible exists as well: https://github.com/redhat-openstack/khaleesi but OS guy said this is for devel setups and for production we should rather use: https://github.com/agroup/instack https://github.com/agroup/instack-undercloud which is intended for production - although that is not ansible.
infrastructure@lists.fedoraproject.org
-
David Gay -
Miroslav Suchý -
Ralph Bean