sed in progress / thoughts are provided "as is", without warranty of any kind

Network Storage Considerations (2023) - Musical Chairs Update

2023-04-13

Background

Since my previous storage considerations, I’ve played a game of musical chairs with rearranging the storage workflow. A fuller picture of what I started out with, is:

sequenceDiagram participant B2 participant Clients participant 8-disk participant 6-disk Clients->8-disk: SMB client R/W Note over Clients,8-disk: Ex: Lightroom, DaVinci 8-disk->>B2: Cloud Sync photo dataset 8-disk->>8-disk: Create dataset snapshots 8-disk->>6-disk: Send snapshots

Current State

I’m now mid-transition where I’ve promoted the 6-disk TrueNAS and demoted the 8-disk. The 8-disk demotion is to prepare for its eventual replacement with a Synology. That comes with a significantly lower power utilization, but means I can’t rely entirely on TrueNAS / zfs utilities as I leave the ecosystem.

That transition now looks like:

sequenceDiagram participant GCP Archive participant M1 Air participant 6-disk participant 8-disk M1 Air->6-disk: SMB client R/W (over 10GbE) Note over M1 Air,6-disk: Ex: Lightroom, DaVinci 6-disk->>6-disk: Create dataset snapshots Note over M1 Air: Arq Backups activate M1 Air 6-disk->>M1 Air: Read data M1 Air->>8-disk: Write backup M1 Air->>GCP Archive: Write backup deactivate M1 Air

The additional goals here were to:

  1. Promote the 6-disk to Primary
  2. Start offloading app hosting and backup scheduling, away from any NAS

Enter Arq

For (2), this was a good opportunity to revisit Arq backups. It’s something I’ve had my eye on for several years, but never needed in the past because I was able to rely on the underlying zfs utilities.

Initial impressions were mixed: I liked the UI/UX and built-in explorer, but the proprietary blob file structure that it drops onto the target filesystem wasn’t what I expected. I went into this assuming it’d be similar to rclone which mirrors the file and directory structure from the source. File-level encryption with rclone simply results in hashed names and data, but in a mirrored directory structure.

That wasn’t the case with arq.

However, there is a published data format as well as a (perhaps older) cloud data format, that can be found in their docs. And their application, even unlicensed, can be used to restore files–if need be. While still a far cry from the open nature of rclone, I believe this should still prove acceptable. There are features that Arq comes with which have tipped the scales in its favor:

  1. Point in time backup restoration
  2. Deduplication
  3. Incremental block updates
  4. Using S3 (and S3-compatible) immutable object-lock

What’s Next

As the last remaining major hurdle, my Windows system still needs to point to the 6-disk TrueNAS.

After that, I can make the optimizations that I’m looking for:

  1. Replacing the 8-disk
  2. Continued offloading of apps onto the M1 Air (or my idle Raspberry Pi 4… or an Intel N5095-based PC?)
  3. Transferring the Lightroom Catalog from my Windows system to the M1 Air
  4. Incorporating a DAS into the editing workflow