Caravan (Part 1) - A New Plan For A New Place

My partner and I are looking for a new place to be.  Being the computing-dependent nerd that I am, I want to be able to have all the services I depend on in our current place available in the new place as soon as there's power and Internet.  To that end, I decided what I ought to do is build a road-ready rack with a firewall and a few computers in it.  Once we close on the new place and get utilities turned on, I can pop the rack in there, and be up and running in an hour or 2.

Some background -  About a year ago, I had switched from all my "vital services" being on VMs on a Proxmox cluster to hosting them on a couple RasPi 5s.  I consolidated a lot.  Where before I had separate VMs for my RMVOD, Icecast2, Desktop, food inventory web app, file services, and so on, I decided I could do a lot more with a lot less CPU and memory overhead if I just had it all running on one 16GB RasPi 5.  I have a second RP5 that is the Jenkins host and hosts backup media.  Actually there's a third RP5 that acts as a bastion server and rack console.  This resulted in using a lot less power and generating a lot less heat.  The file services are facilitated by 4 x 8TB disks configured in a RAID5 arrangement in a USB3 enclosure, viz:

 

Can be got here: https://www.amazon.com/dp/B0DD3GZSGJ 

 That USB3 4-bay HDD enclosure hangs off the one main 16GB RasPi 5, and holds my audio and video stashes, my personal file share, and various sundry other bits and bobs.  Other than that, and USB connections to printers and backup media, basically my whole "infrastructure" is handled by 3 Raspberry Pi 5s and a tiny Netgate box doing the routing. 

  This transition has worked out well.  My plan is to more or less replicate this success but in a portable package.

 Enter Caravan

 The name "caravan" was just meant as a nod to portability and the kind of "home away from home" self-contained nature of a camping trailer or RV.  I settled on a 6-space rack enclosure, such that it could handle a switch, a shelf for a Netgate 1100, 2U for a RasPi enclosure, and a PDU.

I ended up getting this enclosure:

 

Can be got here:  https://www.amazon.com/dp/B0CZ41ZHD6

 It comes with screws for the rails (it takes 10-32 screws, but prefers them short), so that's handy, but there's a few shortcuts taken by the manufacturer that might cause frustration.  The plastic stinks -- it off-gasses... It claims to be HDPE... and maybe it is, but I've never come across HDPE stinking like this.  Screws holding the rack rails to the enclosure are poorly placed.  Things like that.  On the bright side it's cheap, and I'm not going on tour with Aerosmith (or whoever the kids like today), so extreme toughness and/or ATA-certification is not important for me in this application.  Surviving an 8-hour car ride is the thing. 

 The switch I settled on is a simple TP-link 24-port unmanaged gigabit switch (TL-SG1024S). It fits in 1U, comes with the rack ears and is under a hundred bucks. 

 Can be got here: https://www.amazon.com/dp/B0779R9LJ3 

 The PDU is kind of an odd choice.  I went with one from Pyle.  Yeah, Pyle.  Normally, Tripp-Lite has been my go-to for rack PDUs in my personal/home stuff.  They're pretty bulletproof, and I've found them reliable.  In this instance, tho, I was anticipating having a bunch of wall-wart power supplies (because no one has figured out how to do a rack-mount muli-Pi power supply yet, apparently) and 1U PDUs with traditional outlets are just not happy with that.  This Pyle unit has all pigtails in the back.  It really is a thing of beauty.  Viz:

 

Can be got here: https://www.amazon.com/dp/B072RC5NZQ 

 The big deal, tho, is the enclosure/mount for the Raspberry Pis.  I wanted something that would accommodate the PCIe SSD capability offered by the Pi 5.  I also wanted to be able to stuff it with RasPis, because... who knows what kind of goofy shenanigans I might get up to?  I went with a GeekPi unit which stands the RasPis on their side, and has a daughter board for each of the 10 "caddies" which uses pogo pins for power delivery and some GPIO connections, and connects to the PCIe connector on the Pi for M.2 SSD support via flat-flex.  This enclosure holds up to 10 RasPi 5s.  My only big gripe with this thing is that the screws for the caddies aren't captive.  The daughter boards seem to work great, and packing up to 40 cores of Raspberry Pi goodness into 2U is pretty sweet.  The daughter boards move the power connection to the front panel, so easy power on/off is right there.

 

 Can be got here: https://www.amazon.com/dp/B0DQL2N6V9

I have the rack assembled and 7 of the RasPi slots occupied.  Now I'm building out services. I've currently got 1 16GB, 2 8GB and 4 4GB RasPis mounted.  One of the 8GB units has a 1TB SSD setup as a MariaDB server, with the OS on a MicroSD, and the data on the SSD.  The other 8GB unit also has a 1TB SSD, but the OS is installed on it.  This one is serving as a Desktop machine with xrdp installed.  I have one of the 4GB RasPis serving as a bastion server and rack console, and the other 3 4GB units have not been assigned, although I expect one of them will be a file & print server (and maybe identity) and one will likely be media streaming.  I have designs for setting up local email, chat, groupware, an Asterisk PBX, and a few other services.  The 16GB unit will host most of the dynamic web stuff (minus database and file needs).  I haven't worked it all out yet, and I have some room to grown in the GeekPi enclosure.

More to come

There's plenty more to do and a lot more decisions to take.  I'll update as this progresses, but I thought it might be nice to offer some of the stuff I've done so far in case anybody's considering a similar project.

I think that's all for now. 

 

 

Proxmox Migration - Progress Report #1

I have procured parts to build a "modern" Proxmox host, including an ATX MoBo, 12C/24T AMD Ryzen 7900X, 128GB RAM, a 2TB SSD, and 4 x 1GB NIC.  I've pre-built the new host (Joanna, named after Roz mother from Frasier ...and, yes, I've set up that many machines that I'm going that deep into the bench of Frasier characters) on a "test bench" frame I bought.

The parts... minus the NIC and SSD.  Also ended up
using the cooler that came with the CPU.

I have done some storage experiments, and arrived at the conclusion that using the existing 18TB disks in a XFS RAID5 will be the way to go.  That means I will need to wipe the disks and re-format them when I do the OS Reinstall on Jerome.  For continuity of service, that means I need to port those files off to somewhere else.  I have decided to light up Floyd, my old 20-bay storage server, which has been cold iron for a little over a year (that's when I migrated storage to the disks onboard Jerome).  Floyd is up and operating as expected.  I have migrated my video library back over there, and now I have to migrate my other file shares over there.  Then I need to repoint all my share mounts.  That'll be fun.  This is in preparation for the OS install on Jerome.

Of course, I'll also need to migrate all the VMs currently running on Jerome off to Joanna.  In previous migration attempts, I have not forced the MAC address of the copy to match the MAC of the original.  For these remaining VMs, I really want to make sure the copies on Joanna look just like the originals did on Jerome.  Hopefully I can coordinate everything and set the MACs, and they just come up.

I've done some networking experiments with the new NIC, and managed to get 2 of the 4 interfaces set up, which is the minimum I need to set-up my replacement PFsense VM.  I plan on the other two being "room for expansion".  I've decided to go with a replacement rather than a migration, as I want to be able to "flip the switch" on my other machines -- from VirtualBox-based VM to Proxmox-based VM -- without any down time.  Once Jerome is back up running Proxmox, I'll just set up another PFsense VM there.  ...but I'll be able to take my time and make sure all the network interfaces are set up and working as I expect.

I had hoped that I could setup a "storage" VM on Joanna, and mount the shares from Floyd there, and re-share them via NFS, which, I'd hoped, would eliminate the need to re-write all the /etc/fstab files on all the machines on the network that use the shares twice.  It looks like that's not going to work.  However, I did figure out how I'm going to do the big disk array once Jerome is running Proxmox, and I will be using a storage VM to share those filesystems... I may even try out TrueNAS.  Who knows.  At a minimum, I can just use Debian, as my needs are pretty contained.

Ultimately, my plan is for Jerome to be the "prime" VM host, where all my "full time" VMs (storage, firewall, desktop, streaming servers) will run, and Joanna will be my "secondary" VM host, where "lab machines" and other non-full time VMs will live.  Each VM host will have its own local storage, but the "Big Store" will live on Jerome for the foreseeable future.  Floyd (named, incidentally, for Mayberry's favorite barber -- I used to name my machines after Andy Griffith Show characters) will go back to cold iron once the migration on Jerome is complete.

Today, I plan to move the guts of Joanna into the case I'd originally bought for Zora.  Then, I'll migrate some (all?) the non-PFsense VMs to Joanna.  Maybe I can also set up a PFsense VM on Joanna.  We'll see.  It looks like it'll be a full, rich day.

Migrating from VirtualBox to Proxmox - A beginning

A long time ago, now (2019, as it turns out), I set up a super-duper AMD Threadripper machine with 16 cores and 32 threads, and 128GB of RAM to be a Virtual Machine host for my home lab. Since then, it has evolved into the host for my PFsense router, my "desktop" machine, and a bunch of other VMs, as well as my storage server. At the time, I set it up with a regular Ubuntu install with VirtualBox running on it. All these VMs are being run on VirtualBox. 

I know. 

Well, time-and-a-half went by, and I became aware of Proxmox. I never really had spare hardware with enough grunt to really try it out, so I kinda let it go. Then, my employment relationship with that online furniture retailer (as well as those of about 500 other folks) suddenly ended. The machine I'd put together as a "BYO" desktop for work was suddenly out of a job as well. That was a pretty beefy machine... AMD Ryzen 7, 8C16T with 32GB of RAM... but it had a sort of narcolepsy problem. it would just... for no apparent cause... stop. It didn't seem to be thermally-related... not the power supply. I have no explanation... it just farts out... but I can usually get a good 6-8 hours out of it, sometimes a lot more. ...so it finally dawned on me that this old beast might be a good Proxmox test mule.   That was a couple months ago.  I decided to give it a go, and I like it. I like it enough that I want to migrate my "production" VM host to Proxmox. 

Well. 

I'd hoped to migrate the VMs, one by one, to Proxmox on a different host, and then do an OS re-install on Jerome (the Threadripper box - named after the "good family man" to whom, Frasier and Niles owed a "favor") with Proxmox, and migrate some or all of the VMs back over. 

The old BYO machine (named Zora, after Fraiser's "excitable" Greek aunt... I have a lot of machines named after characters from Frasier) is based on a Mini-ITX motherboard, so it has only one PCIe slot, and that's taken up by a video card. I need more Ethernet ports to run my PFsense router... and, of course, there's the "oops I forgot how to computer" problem. So, I decided to build a new machine, with a full ATX motherboard, and slots... and I managed to find an AMD Ryzen 7 CPU with 12 cores and 24 threads AND a built-in GPU... so that's exciting. Not QUITE as heavy as Jerome, but pretty good... and 128GB of RAM. So, I can put a 4 x 1GB Ethernet board in there, and run my PFsense router, and the board has 4 SATA ports, so I can figure out how to do my storage. 

Ah, yes, storage. Jerome has three 18TB drives in a RAID5 array, and those are run by the host OS. I'm thinking that under Proxmox, I will have the raw drives forwarded up to one of the VMs that will serve as the storage server... but I'm not entirely sure how that will work. I haven't really dug into the docs yet. I know it's been done in other KVM environments, so... I'm confident it should be doable. When I get the new machine up and running I can try it out. It would be lovely to just be able to take the disks as they are, and plug them in on a fresh Proxmox install, build a fileshare VM and just... ta-da! We'll see. 

So, the parts for the new Proxmox host are supposed to show up today. Hopefully, everything goes together without a hitch, and I can begin properly experimenting. 

Updates when they're warranted.

RMVOD v0.9.2 is HERE! It's a BIG DEAL!

New Features:

 0) User Sessions.  You have the option to log in at the browser and track your viewing across multiple devices, such that your recommendations will be consistent across devices, and your tv series viewing history will be aggregated.  User sessions also enables you to bring your settings with you, like playback speed, watch next episode, and resume playback.  In addition to your settings, watch data like "current artifact" and "time offset" are also captured in your session, so you can switch devices easily in mid playback.

1) Recent Episodes. Thanks to the existence of User Sessions, it is now possible to keep track of what the latest episode is that you've watched for a TV Series, and either re-watch or watch the next one with just one click.

Bug Fixes and Minor Tweaks:

Posters now fetched from OMDbAPI - The previous source of poster images became unusable.  RMVOD has been shifted to take advantage of the service on OMDbAPI 

Added "stats" - The "Settings" tab has a clickable link which fetches the current "statistics" for the server, including the current count of movies, tv episodes and tv series, as well as viewing statistics for the top 10 tags among tv series and movies, and the top 10 titles for each tag.

OMDbAPI "Series Fetch" bug fix - Fetching series info from OMDbAPI had a bug, which has been fixed.

Added Playback Speed option to "Settings" - Playback Speed can be set as a option in the Settings tab.

Fix title set on play. - There was an issue with setting the page title when playback starts.

Fix Recs Poster Image thing. - There was an issue in the recommendations artifact details wherein the artifact details were not getting the "oops no poster" when there was no poster for the artifact.  

Reworked statistics to have the period delivered by the client, rather than hard-coded in the API.  

Download the latest stable code here:

https://github.com/taobear68/rmvod/releases/tag/v0.9.2

RMVOD v0.9.1 is HERE! It's a DOOZIE!

 Hi folks!

So, I've created a new tag on the RIBBBITmedia VideoOnDemand code repo, and this one undoes the whole "feature complete for release 1.0" thing.  Many new features.  Oops!

In this tag you'll find several new features.  Notably:

0) Config file.  I'd planned for some time to implement this, and this seemed a really good time to do it.

1) Recommendations (Recs).  This is a new tab in the RMVOD screen, which provides a "carousel-like" interface for browsing artifacts based on connections to viewing trends on the server.  The Recs interface shows artifact details in a floating box when the artifact's image is clicked.  For TV Series artifacts, options are presented to start playing the series from the beginning or to select a season, and browse episodes to select one to play.  Recommendations are cached on the server and expire after a period of days set in the API's config file.  There is a "Refresh Recs" link on the Settings tab to force a refresh of the recommendations.  Recommendations functionality requires a new table in the database.  See README.txt in the root of the code tree for details on upgrading an existing system to support this feature.  For systems where the DB is hosted on the same box as the API, and the box is really-low resource... like a RasPi 2 something even less capable... Recommendations should be disabled in the API's config file.

2) Quick Search Links.  This is a new set of links on the "List/Search" tab which present the lists of artifacts from the Recs tab in the familiar list format.  This also depends on Recs being enabled on the server.

3) "Last Search" display.  Now, when you execute a search, the "search factors" will be displayed below the artifact list to remind you of what you searched for, whether the search is single factor or multi-factor.

4) "Play fullscreen" option.  This is a checkbox on the Settings tab which asks the player to initiate playback in full-screen mode.  This currently only works for playback initiated by the user clicking the artifact to initiate playback.  When playing a series, the second and later episodes played will not be in full-screen.  This is a known issue related to a security feature in popular browsers.  I'm looking for a work around, but for now, it eludes me.

5) "Populate Episode Details" link. In the Artifact Details box for a TV Series, at the bottom, to the right of the "Edit" and "Associate Episodes" links, there is now a "Populate Episode Details" link which initiates an operation on the server to query IMDbAPI for the details relating to the "series" artifact and each of the episode artifacts.  This requires that the series artifact already has episodes associated with it, and has its "imdbid" value set appropriately.  The use of this feature requires a "key" for the API which can be obtained by visiting here https://www.omdbapi.com/ and setting the key value in the API's config file.

Download the latest "stable" code here:

https://github.com/taobear68/rmvod/releases/tag/v0.9.1

RIBBBITmedia VOD -- The Beat Goes On

 RIBBBITmedia VOD -- The Beat Goes On

A lot has happened since October.

• There's a GitHub repository
• The "one page" is redesigned to use a tabbed interface
• There are now search links in the artifact details for people involved
• Multi-Factor Search is working now
• "(series)" now in pale yellow to stand out a little more
• Page header has been simplified, and now contains the artifact details when something is playing
• Cookies are now being used to capture the currently playing artifact and play position to permit resuming playback on page reload
• Cookies are capturing the state of "Settings" to allow them to persist across page loads
• Search factors added: year range, major type, SQL WHERE Clause
• Single and Group artifact creation works in the web UI
• Version numbers are gathered from API, CSS, HTML, and DB, passed to the JS UI, and listed in the footer of the page

Fresh load of the "One Page"

The "One Page" is done with a nearly black background and light gray font to aid readability and minimize shock to the eyes when being used in a "home theatre" setting, with low ambient light.

Artifact playback

Artifact details are included in the header during playback.  A clock display is also included in the header.

Add multiple artifacts

"Add Multiple Artifacts" is a new feature (not that you could tell by the lack of visual sense it makes)

Still to do:

• New function: Add Episodes to a Series
• Add some descriptions, directions, and (maybe) tooltips
• Clean up alignment of new forms

RIBBBIT media VOD

 RIBBBIT media Video On Demand!

Well, what is the point of this?!  

As a long-time cord-cutter, I've made some use of streaming services (mostly YouTube, but also Netflix and Amazon Prime), but I've often been frustrated by available solutions for streaming my own media at home.  Part of what frustrates me about the commercial VOD services is that I'm at the mercy of what their library managers decide will be available to me.  Let's say I want to watch "The Towering Inferno".  Well, if none of the services I'm subscribed to currently think that is worth having online, I'm out of luck.  Well, if I own the DVD, why not just watch it on that?  DVDs are great for getting "ownership" of the media into my possession, but once I have it, I'm very restricted in how I can consume it (to say nothing of maintaining and regularly interacting with a library).   Yes, "ripping" DVDs offers some solace, but now there are data storage issues, and how do you catalog and index such a library? For years, I simply hung an NFS share out there, and watched via VLC on a PC or laptop.  That has obvious limitations.  That DID scratch my itch for a time, but I wanted something more library-ish with indexing and searching and so on.  One obvious (but exhausting to maintain) way to achieve this would be a forest of symbolic links on that NFS share... but another way would be to use an RDBMS to maintain all those relationships, and simply query against that to get the filename of interest.  That has the potential to be much easier to maintain (and expand).

What's it made of?

Since HTML5 includes (among other things) handling for video playback, I figured it couldn't be TOO hard to implement a simple video player page.  I was right.

I've been working on re-ripping and/or transcoding my video library from MPEG2 to MPEG4.  As of this writing, I have over 2,500 TV Episodes and 350 movies in MP4 format (all ripped from DVDs I own) and logged in the DB, many with metadata (such as runtime, people and companies involved, synopsis, and so on).

The system consists of:

• a static HTML page, which loads 2 JavaScript files, and sets up the layout of the page, 
• a Python file which serves both as a Flask REST API and a CLI for interacting with the database, and 
• a MariaDB database.  

All told, the HTML, JavaScript, Python, and SQL DDL currently fit in under 6000 lines of code (including comments and whitespace).  Some refactoring will likely bring that down a bit.

As of now, the ability to edit DB info for individual "artifacts" within the web page exists.  The CLI is for doing things like bulk-creation of artifacts from file lists, creating tags, assigning individual "episodes" to a "series", and so on.  On the bright side, the back-end code to achieve these things exists in the Python file, but I haven't gotten around to hanging a web UI on any of it yet.

Also, there is no authentication on the web side... You are J. Random User.  The back end keeps no stats on you, and currently places no restrictions on whether you can update DB information for any particular artifact.  Obviously, this is not intended for public use.  It's intended for a person (or family) to interact with their own video library.

Call to action(?)

Is this something that interests you?  Would you be interested in seeing this go up on a publicly accessible repo?  Comment!

Netgear SSO? That's a hard no. pfSense to the rescue!

 Some time ago... a couple years, now... my trusty old cheap Netgear router breathed its last, and I bought a new one.  The new one had this pain in the ass "Single Sign On" thing that made you set up an account with Netgear in order to log in to the router.  Hard pass on that.  It's possible to log-in locally, but it's extra steps that piss me off.  I've been ruminating on switching to pfSense for my routing/firewall needs, and just getting one or more cheap Wireless Access Points to handle the WiFi.  I have enough hardware just lying around that I could have done pfSense on a dedicated x86 machine, but the power, space, and noise were things I was not crazy about dealing with.  I'd thought about installing it on a VM on my Threadripper VM Host, but the networking would be a problem without additional hardware.  I finally got a 4x1G ethernet card for my VM Host, and stood up and pfSense VM.  

The config process was not bad, per se, but it was a little opaque in spots.  I have a separate Raspberry Pi running dnsmasq, handling local DNS, DHCP, and DNS forwarding, whereas nearly all of these modern router appliances have built-in, and a pretty enthusiastic about acting as your DHCP and DNS server.  Chalk it up to paranoia, if you like, but I prefer the "component stereo" model here, if for nothing else, it allows be to keep at least some of my more complex config stuff up and running if something like a router quits.  Getting DNS forwarding through pfSense was a little bit confusing at first, but I got it worked out.

Port forwarding turned out to be easy once I figured out that refreshing DHCP leases had to be done on target boxes for return traffic to go back through the router.  I'm smart.

OpenVPN server setup on pfSense has not brought me joy, yet, but it could, still.  I remain hopeful.  Part of the problem, I think, is that my network is on another RFC-1918 network -- that is, I need to traverse another "local" network to get to the Internet.

I ran into a couple hiccups when I tried to get one of the WAPs I bought working.  They're TP-Link AC1200 (TL-WA1201) devices, and they were cheap.  I bought 3, because why not.  The "quick setup" didn't seem to stick, entirely... so I had to kind of work through "not quick setup", and it seemed to take a few tries at that, with resets in between for it to really come up and do its job.  So far, an hour into its service life, the first TP-Link TL-WA1201 seems to be doing the right things.

So, now, my private network is fronted by a VM running pfSense through a dedicated 4x1GB ethernet card, and WiFi is running of a TP-Link WAP, and with any luck, I'll never have to deal with Netgear's stupid SSO nonsense again.

The next bit of excitement for the home network is a 48-port switch.  I know, how can I possibly justify a 48-port switch?!  Well... I'm not going to pretend that I have 48 wired Ethernet machines up and running all the time, but I do have 6 machines in my rack, and 20 or so machines strewn about, and I have little "desktop switches" all over the place... I'm hoping to get away from that.  Of course, that will mean doing more home-runs, but I'm hoping it will make for less confusing networking.