Clojure: Modifying a List of Vectors based on Previous Iteration Value

(EDIT: Looks like I muffed the title of this one but I'm going to leave it as is. Should read "Vector of Maps".)

Here's an interesting problem I've been pondering this morning. What is a good functional approach to solving a problem in which you have to iterate through a vector of maps and copy a value from a previous item if the value is nil in the current one?

Problem Statement

This is something that came up while trying to parse output from Cisco Nexus show ip bgp regex....

*|e172.16.1.1/32                                    0 65535 i
*>e                                         0 65535 i

And this results in a map that looks like the value of c here:

(def c [{:next-hop "", :prefix "", :route-status "*|e"}
        {:next-hop "", :prefix nil, :route-status "*>e"}])

The goal of this code is to carry-over the value of the last seen prefix value to all subsequent values that have nil prefixes. In the nature of the output, the last-seen prefix value may not be from the iteration immediately before.

The return of our code is expected to look like this:

[{:next-hop "", :prefix "", :route-status "*|e"}
 {:next-hop "", :prefix "", :route-status "*>e"}]


My initial approach was to initiate an atom in a let-binding for the carry-over value and to use map.

; map with atom
(let [prev (atom nil)]
  (map (fn [r]
         (if (and @prev (not (:prefix r)))
           (assoc r :prefix @prev)
           (do (reset! prev (:prefix r))

Carrying over a value in a mutable atom isn't a very functional way of dealing with the problem so here are a couple more approaches I came up with.

The next approach uses reduce in which each iteration returns a compound accumulator as [<previous value> <accumulating vector>]:

; reduce with compound accumulator
    (fn [[prev acc] v]
      (if (and prev (not (:prefix v)))
        [prev (conj acc (assoc v :prefix prev))]
        [(:prefix v) (conj acc v)]))
    [nil []]

What I like about this approach is it's very clear about what's going into the next iteration. There isn't any mutable state. What I don't like is having to deal with the compound accumulator and having to unpack the result of the reduce using second.

My final approach uses loop/recur:

; loop/recur
(loop [prev nil
       acc []
       coll c]
  (if-let [v (first coll)]
    (if (and prev (not (:prefix v)))
      (recur prev (conj acc (assoc v :prefix prev)) (rest coll))
      (recur (:prefix v) (conj acc v) (rest coll)))

When I first looked at this I thought it was a bit awkward but it's growing on me. The loop's base case is when there are no more items in coll - return the accumulated result. And the recursing code will either update prev or append a modified value.

One might think that this approach may suffer from limits on stack depth but as I understand it, loop/recur has some special handling which makes it not a problem.

Github Gist

Password Recovery for Cisco Nexus NXOS 7.0

Looks like getting into the kickstart for password recovery has changed a bit with NXOS 7.x. NXOS 7.x features a consolidated binary image rather than the previous pair of kickstart and system images we are used to seeing on Nexus 3000.

It's not exactly obvious how one boots to the kickstart. So here's the new sequence:

Step 1 - Power Cycle and Interrupt to the Loader with CTRL-L

When you see this:

Press  ctrl L to go to loader prompt in 2 secs

Press CTRL-L

Step 2 - Set recoverymode=1 and boot your NXOS 7.x image to get the Kickstart

First, ensure you can find your NXOS 7.x binary image using dir.

loader> dir

Then force boot to kickstart and boot the image.

loader> cmdline recoverymode=1
loader> boot bootflash:nxos.7.0.3.I2.2a.bin
Booting kickstart image: bootflash:nxos.7.0.3.I2.2a.bin
 Image valid
INIT: version 2.88 booting
Skipping ata_piix for n3k.

This should land you at the switch(boot)# prompt and you can follow usual procedures from there.

Cisco Nexus 3000 Loader and Kickstart - Getting Unstuck

I found myself in a situation at work where after a downgrade, my Cisco Nexus 3000 switches were sitting at a loader> prompt with no readable images on the bootflash. Here are some things I noticed while trying to get unstuck.

Network Connectivity for Loader and Kickstart

If you end up in a jam where your switch cannot boot to a system image, you will need one of the following to get full network access:

  • a bootable kickstart/system image on the boot flash (assuming it is readable... mine was not)
    • a bootable kickstart and system image on a USB drive which has been connected to the USB port of the switch
    • a TFTP server that can be reached via the Management Ethernet port (MGMT) on the switch if you apply an IP address and, if needed, a default gateway (no dynamic routing).

In this last option, I stress that The MGMT port is the only port you can configure with an IP address in loader or kickstart.

We don’t normally have the MGMT port cabled up, so I needed to reach out to the Data Center engineers to get them move a cable from a normal switch port to the MGMT port. Once I did that, I was able to from the loader to kickstart.

Examples of interaction for loader and kickstart

TBD... I will update this when I get my log files off my work computer for these interactions.

The Case For An Out of Band Management Ethernet

All of this jockeying to try to get network access when the switch is down hard... This makes it very clear that the management interface is special.

There is generally value to being able to reach the switch independent of the routing protocols running on the switch. And if hands-off recovery of a switch that is not able to boot is a requirement, having a management ethernet network built out is a prerequisite.

Alternatively, If you have a solid way of ensuring that a spare switch can be swapped in for a failed one, and that switch will either have or will be configured with a valid configuration, then the case for building out a management ethernet network is less.

Either of these will address the problem for this scenario. You probably don't need both. It's important to remember that having an out of band management ethernet network is a solution to one or more problems. And if you have alternate solutions, it may make sense not to have one.

That being said, a management ethernet network is also useful for stats, logging, and controlling the switch in a manner which is independent of the switch's routing state and access-control-lists applied to the normal switch ports. So there are other scenarios for which being able to swap in doesn't get you the same level of functionality.