OpenStack Compute Node - state path and lock file directories

I may have posted on this topic before, because I ran into this issue before.

I was setting up OpenStack for a colleague of mine, and had all sorts of issues getting it to work.

A couple of problems I had were related to services that were not enabled, and when you rebooted the unit(s), the services were not starting up. These were easy to fix.

The difficult issue to find and fix - which took me almost a full business day - had to do with how OpenStack Nova configures itself.

In the /etc/nova.conf file, there is a variable called a state_path. This variable is set to /var/lib/nova - a directory nova creates upon installation and sets permissions to the nova user and group.

In this directory, is a subdirectory called "instances", where Nova puts running instances.

The problem, is that Nova on installation does not seem to check or care about partition and file system sizes. It just assumes.

The issue we had, was that on a CentOS7 default installation, the /var directory is part of the root file system, which is very small (15-20 Gb), as it should normally be (you generally separate root file system from apps and data).

When you would start Nova, even in debug mode, you never saw an ERROR about the fact that Nova had issues with any of its filters (disk, ram, compute, et al). They were being written into the log as DEBUG and WARNING. This made finding the problem like finding a needle in a haystack. And you only saw this evidence after enabling debug in the /etc/nova.conf file.

Eventually, after enabling debug and combing through the logs (on both Controller as well as Compute node), we found a message on the Controller node (NOT THE COMPUTE NODE WHERE YOU WOULD EXPECT IT TO BE) about the disk filter returning 0/1 hosts.

So - we moved the /var/lib/nova to the /home/nova directory (which had hundreds of Gb). We also changed the home directory of nova in /etc/passwd to /home/nova (from /var/lib/nova).

We got further...but it was STILL FAILING.

Further debugging indicated that when we moved the directory, we forgot about another variable in /etc/nova.conf called the lock_file_path. Used for RabbitMQ communication, this variable was still pointing to a lock file in /var/lib directory (that had been moved to /home). This caused Compute Filter issues - also showing up as DEBUG and WARNING messages and not errors.

Ubiquiti EdgeRouter X - Power Over Ethernet (POE)

I tested the POE feature of the router this week. I will cover what I did, and the results.

I had a Polycom VOIP phone that was POE-enabled, and I did not have the power supply for it; hence the use case, and excitement about leveraging this feature when I started using this router.

By default, this router uses eth0 as the Management port (192.168.1.0/24), and eth1 as the WAN / Internet port.

This assumes that only a single ISP is being used with the router which is a good assumption considering this is a small consumer appliance and not an Enterprise router.  The router actually supports dual ISPs (link aggregation), which is certainly an Enterprise feature, but to employ this feature you need to change the ports such that you are using eth0 (leftmost port) and eth4 (rightmost port).

The POE feature on this router is POE Passthrough, which means that you need to insert a POE-enabled device into the "input port" - which happens to be eth0 (leftmost port), and POE output will be supplied on the passthrough port eth4.

1. First, I checked the line to make sure it had POE, by connecting it to the phone directly. Voila', it powered up. So, knowing POE was indeed enabled on the input CAT-5 from the patch panel, I set about trying it on the router.
2. Next, I had to reconfigure the router so that eth0 was the WAN port. This meant removing (or moving) the Management port. Since a Management port is handy to have, I made eth1 the Management Port - essentially switching eth0 and eth1. I will skip the details on how to do this, but basically you can do this through the menu system of the device by going to Dashboard, Services and Firewall.
3. I then connected the POE VOIP phone to eth4 - directly. A common mistake is to use a switch, but if the switch is not a POE switch, this won't work (you can also burn up the switch this way). I made sure that eth4 was on the Switch! For some reason, only eth2 and eth3 were on the switch (boxes ticked). I ticked the eth4 box.
4. On the eth4 POE Output interface is where you enable POE. I was surprised you could not enable it on eth0 input link. I enabled this. Keep in mind the router is STILL PLUGGED IN! 

The POE VOIP phone did not power up.  One theory is that the router could not power up the phone on its own, based on just the voltage from the input CAT-5 (keep in mind it also needs to power the router itself). We then powered up the router with the adaptor, thinking this might supply more juice (voltage) to power up the phone. Nope.

I did not have more time to debug this. I disabled POE on eth4, and used a POE Adaptor, which works just fine. Maybe I can attempt this later, but at first test, POE did work. For me. For my phone. This is the first time I have actually used POE.

OpenVSwitch Round III

Turns out that when I rebooted my boxes, nothing worked. I debugged the issue back to OpenVSwitch.

The problem is definitely the provider bridge on the Controller / Network node.

I *think* the problem is that I am using only a single interface on the Virtual Machine, and using that same interface for two purposes:

1. vxlan tunneling to OpenStack virtual machines (via the br-tun OpenVSwitch managed bridge.)
2. internet connectivity (via the br-provider OpenVSwitch managed bridge)

I have to verify this, but that is the hunch.

I have backed out OpenVSwitch once again, reverted back to linuxbridge-agent, and more research to follow.

NOTE: To use two interfaces on the VM means quite a bit of work. The interface ens3 on the VM connects to a bridge that connects to an adaptor on the host. I probably need to use two adaptors on the host, and I probably need additional network(s) created, which has routing implications.

Essentially it's a network redesign - or, we could say, expansion.

What to do if your VM won't boot

After I made a bunch of changes to OpenVSwitch (see last post), I rebooted my VM and it just hung with a "random init crng" message. The last thing it did was bring up a network interface ens3 before this message appeared. I assumed it was trying to start networking, and was waiting for a timeout.

I waited for what seemed like an eternity...5 minutes? 10? Longer? And a prompt came up. I saw little to nothing in the logs, so I became concerned. How would I debug this if I have to wait that long every time I reboot?

I found a few nifty tricks to get you out of a jam if you're virtual machine won't boot.

First, if the VM will come up at all, you can enable "system-debug" with systemctl:
# systemctl enable system-debug

This service allows you to connect on tty9 to the VM as it is booting. On a virtual machine, you typically need to "send" a command to get to this terminal. So on KVM, there is a "Send Command" option and you can send keystroke commands to the VM (i.e. Ctl-Alt-Backspace, Ctl-Alt-F1...F9, et al).  This proved to be quite handy, and a sense of relief that I could actually get inside of the VM - especially if you have not snapshotted it or backed it up.

If you need to do more, or go further, you may need to interrupt the boot process and edit the line in the grub config file.  For Red Hat users, here is a link to a page that gives some direction on how to get into debug mode for a virtual machine.

https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/system_administrators_guide/sec-terminal_menu_editing_during_boot

OpenStack and OpenVSwitch - Round II

Yesterday I installed OpenVSwitch in conjunction with OpenStack and was excited to see that it worked - first time - after I installed and configured it.

To review, there are two compute nodes. These are NOT virtual machines; they are kvm / libvirtd machines (each kvm host is a compute node).  And there is one OpenStack Controller that IS a virtual machine, and this virtual machine also serves as the OpenStack Network node (it is common to see the Controller and Network node separated, but I run them on the same VM).

The two OpenStack Compute Nodes appear to be working perfectly. Each of these has two bridges:
1. br-tun - used for tunneling to the controller
2. br-int - used as an integration hub (virtual patch cords between bridges)

The Controller node, however, is a bit different in that it needs to have an additional bridge called br-provider. This bridge connects out to the internet.

I am taking this architecture STRAIGHT out of the OpenStack documentation (I am running Newton but these diagrams are also identical and present in the Ocata release of OpenStack), which is at this website: https://docs.openstack.org/newton/networking-guide/deploy-ovs-selfservice.html

Below is a diagram - their diagram - shown here for convenience - to illustrate what I am trying to achieve:

Now in the OpenStack Install Guide, for Newton, they don't have you creating this "Network Node". All of the stuff on a Network Node is on the Controller Node. So the Network Node and Controller node are combined. While I am considering breaking the Network Node out into a separate VM or on a separate small box, currently my Controller and Network Node are all on the same CentOS 7 virtual machine running on a libvirtd on one of the Compute Nodes - external to OpenStack though.

Having it this way can be difficult in some ways. For instance, I have to run an OpenVSwitch Agent on this Network Node - and this is not well-documented (documentation leads you to believe you only need the linuxbridge agent or openvswitch agent on the Compute Nodes). Another issue with this is that there are ports (tap) for every dhcp agent, and every instance, which gets confusing when you look at your OpenVSwitch and see all of these, with these cryptic names beginning in qr, qq, etc. You don't always know which is which and where to start debugging.

Anyway...I had a VERY difficult time getting OpenVSwitch to work. The first issue I had was that the bridges in OpenVSwitch did not seem to have the right interfaces. This was a configuration issue. Another issue had to do with the br-provider bridge. I actually thought I should put those on the Compute Nodes even though the diagram did not have those. I didn't understand why the Compute Nodes could not go straight out - and why they needed to be tunneled over to the Controller / Network nodes, which seemed like a huge unnecessary performance hit.

But the MAIN issue I had was that once configured and after I instantiated virtual machine instances, they would come up but without an IP Address. I set the instantiated VM's IP Address statically, and lo and behold, it could communicate just fine. This told me that the issue was DHCP; that the VMs were not getting an IP Address via DHCP. At least I knew, but this concerned me because it did not seem like it would be easy to debug or fix (it wasn't).

2. I checked the dhcp agent services. They were running. I checked the logs. They looked okay. I also did the same for the l3 agent.

3. I then mentioned this to my boss, and he suggested that the OpenVSwitch was not handling broadcast messages. He suggested I trace the broadcast by checking the endpoints and working in from there.

This led to me having to go into the namespaces with "ip netns exec". What I found was that the interfaces inside of the namespaces had "@" characters in the names, making it difficult to run tcpdump.

I finally ran across the ovs-testing package, which has some utilities such as ovs-tcpdump that help debug bridges and interfaces inside the switch.  When I did this, I found the broadcast traffic reaching the OpenStack Controller.

Finally - I realized that I needed to DELETE the dhcp agents and RECREATE them. The original agents were created when the networks were created, using the linuxbridge. Apparently switching to OpenVSwitch rendered these agents inoperable. They could not "adapt" to that change. So - I deleted those dhcp agents, OpenStack recreated them, and now things are working just fine.

Elasticity and Autoscaling - More Testing

Today I went back and did some further testing on AutoScaling.

Here are some new things that I tested:

1. Two Scaling policies in a single descriptor. 

It does no good to have "just" a Scale Out, if you don't have a corresponding "Scale In"!

You cannot have true Elasticity without the expansion and contraction - obviously - right?

So I did this, and this parsed just fine, as it should have.

I also learned you can put these scaling directives in different levels of descriptors - like the NSD. If you do this, I presume that what will happen is that it will factor in all instances across VNFMs. But I did not test this.

2. I tested to make sure that the scaling MAXED OUT where it should. 

If cumulative average CPU across instances was greater then 35% CPU, then the SCALE_OUT 3 would take affect. This seemed to work. I started with 2 instances, and as I added load to CPUs to boost the cum average up, it would scale out 3 - and then scale out 3 more for a total of 8 no matter what load was on the CPUs. So it maxed out at 8 and stayed put. This test passed.

I was curious to see if the engine would instantiate one VM at a time, or would it instantiate in bunches of 3 (per the descriptor), or would it actually just instantiate up to the max (which would be errant behavior).  Nova in OpenStack staggers the instantiations so it APPEARS to look like it is doing one at a time up to three (i.e. 1-1-1), at which point re-processing may kick off another series of 1-1-1.  So this is probably to-be-expected behavior. The devil is in the details when it comes to the Orchestrator, OpenStack, and the OpenStack Nova API in terms of whether it is possible and to what extent you can instantiate VMs simultaneously.

When a new VM comes up, it takes a while for it to participate in measurements. The scaling engine would actually skip the interval due to a "measurements received less than measurements requested" exception and only start evaluating things until and unless it had all of the VMs reporting in measurements that were expected.  I have to think about whether I like this or not.


3. Elasticity Contraction - by using SCALE_IN_TO parameter.

I set things up so that it would scale in to 2 instances - to ensure at least two instances would always be running. I would do this when cumulative average CPU was less than 15% CPU across instances.

This test, actually, failed. I saw the alarm get generated, and I saw the engine attempting to scale in, but some kind of decision-making policy was rejecting the scale in "because conditions are not met".

We will need to go into the code and debug this, and see what is going on.

Ubiquiti Edge Router ER-X - Impressive

I just love this router.

• The icon on the top that shows colorized ethernet plugs (colorization related to status). Cool.
•  It was sooooo easy to configure it.
•  It has a shell that takes you into Ubuntu Linux (uses Ubuntu = impressive). I'm not even sure it is using BusyBox or some slimmed down quasi-linux. It looks like a custom compile of Ubuntu.
•  One cool feature is that it can support link aggregation. I am not using that feature, but it's cool.
• Has excellent support for IPv6.
  

It can also automatically switch the ports on the router so that you don't need a L3 switch to go with it. 

So for example, you can set up:

• eth0 as the management port
• eth1 as the WAN port, and 
• eth2, eth3 and eth4 are switched so that anything plugged into these are on the same network (you defined the network).  

I don't currently have this router configured a router than actually learns routes. In other words, I am not running BGP, RIP or OSPF on it. I don't really need to learn networks, nor advertise networks, dynamically. All I need is to get out to the internet.

I have it set up with a hairpin NAT, and the firewall rules configured on it are rather trivial at the moment but designed to protect ingress through iptables rules.

This is truly a "power user" routing device, and it can fit into the palm of your hand; it is no bigger than a Raspberry Pi device.

This router also comes with some interesting Wizards that allow you to configure the router for certain use cases, like the WAN+LAN wizard.

So I have not done anything in-depth, but I spent an hour messing around with this device and I'm pretty impressed with it.