diskwatch 0.2.0

What it shows

Plus two single-screen views that are not tabs: 2.0 (--v2, six btop-style boxes) and Lite (--lite, 80×24, six keys).

Where lsblk shows you which disks exist, DiskWatch shows you what's happening on them — capacity trending, IO throughput, p99 latency, SMART health, and the files being written right now — and tells you why in plain English when something's anomalous.

2.0

diskwatch --v2

Six boxes tile the terminal with zero chrome rows — no header bar, no menu bar, no status bar. Identity, uptime, sort state, paging and every keybind live inside the box borders, which a box spends anyway.

╭┤1├─┤ io ├─┤ 4 physical · 6 volumes ├────────────┤ diskwatch 0.2.0  nas-01  up 4d 02:18 ├─╮
│ r 128 MB/s  read      peak 412M  avg 96M  iops 5.0k       avg req 36KB · 4 disks summed │
│ 500M┤                        ⣠⣴⣶⣿⣿⣷⣦⣄                                                   │
│    0┤ ⣀⣠⣤⣶⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿ │
│   2m┤ ────────────────────────────┤ 60s ├───────────────────────────────────────┤ now ├ │
│    0┤ ⠛⠛⠿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿ │
│ 160M┤                              ⠉⠛⠿⠟⠋                                                │
│ w 14 MB/s  write      peak 38M   avg 11M  iops 4.3k                                      │
│ vitals  util ■■■■■■■■········  52%  await 0.35ms  iops 9.3k    p99 4.25ms · headroom 48% │
╰─┤ V view  , settings ├────────────────────────────┤ inflight 2 · 52% util ├─────────────╯

The mirror is earned here. In a CPU monitor a mirrored graph is decoration, because compute has no opposing direction. Disk read/write is two directions of one flow, so it takes the full width: a restore is a cliff above the axis, a backup a cliff below it, and a database at work is roughly symmetric about it.

The latency histogram is the disk-specific part. Seven buckets from <0.1ms to >50ms, drawn as horizontal bars. Bar length comes from the count, but colour comes from the bucket's position — so the right-hand buckets are red even when nearly empty, and you learn where the tail lives before it grows. A mean of 5ms can hide a p99 of 60, and the tail is what you actually feel.

Below 104×32 it falls back to a compact screen — the mirror and the percentiles survive, because they are the identity of the tool; devices, volumes and smart collapse to summary lines.

V cycles full → lite → 2.0 → full from any view, the same convention as netwatch, and the View row in the , settings overlay walks the same list for anyone who finds it there first. Unlike Lite, the 2.0 view carries that overlay: it replaces the 8-tab view rather than deliberately reducing it, so the dials stay reachable without leaving. Every key it binds is printed in a box border — and nothing is printed that it doesn't bind.

Graphs are braille: two samples per character column at four vertical levels per row, with the fill coloured by its height in the graph rather than by which series it belongs to. Axis ceilings come off each series' own measured peak, on a ladder whose rungs are at most 25% apart — so a peak always lands in the top fifth of the axis instead of leaving the top braille row unused.

Lite

diskwatch --lite

One screen at 80×24, six keys, no tabs. Read and write throughput, a capacity line that answers how long have I got rather than how full am I, and the busiest files — sized to run in a tmux split or an SSH session to a NAS.

diskwatch  nas-01 · 4 devices                              ● 2.6 / 3.8 TB

r 128 MB/s read                              peak 412 MB/s  avg 96 MB/s
▁▂▃▅█▇▅▃▂▁▁▂▄▆█▇▄▂▁▁▁▂▃▅▇█▆▄▂▁▁▁▂▃▄▆█▇▅▃▂▁▁▂▃▅▇█▆▄▂▁▁▁▂▃▄▅▇█▆▄▂▁▁▂▃▄▆▇█▅▃▂▁▁▂
w 14 MB/s write                                    peak 38 MB/s  avg 11 MB/s
▁▁▂▁▁▂▃▂▁▁▁▂▁▁▂▂▁▁▁▂▃▂▁▁▁▁▂▁▁▂▂▁▁▁▂▁▁▁▂▂▁▁▁▂▃▂▁▁▁▂▁▁▂▂▁▁▁▂▁▁▂▃▂▁▁▁▂▁▁▂▂▁▁▂▁▁▂
 78s ago ─────────────────────────────────────────────────────────── now
/ 68%   /var 94%   health 4/4   growth +8.2 GB/day    /var 94% · 11 days left

It is not the full tool with tabs hidden — it is a different view for one machine and one question. L jumps straight here and back, V cycles through all three views; the full 8-tab TUI stays the default at every terminal size. Below 80×24 Lite shows a notice rather than a clipped grid.

Same grid, keys and palette as netwatch --lite — only the subject changes.

The capacity projection needs about a minute of observation before it will commit to a number, and stays silent when usage is flat or shrinking. The file table reports events, not bytes: FSEvents and inotify say that a path changed, not who wrote it or how much (see What's real, what's deferred).

Install

brew install diskwatch                # macOS / Linux
nix-shell -p diskwatch                # NixOS / Nix
paru -S diskwatch                     # Arch
cargo install diskwatch               # anywhere with Rust

Or grab a pre-built binary from Releases.

The Nix and Arch packages are maintained by community packagers — thank you. File packaging issues with them; file diskwatch bugs here. The Repology page shows which packages are current.

From source:

git clone https://github.com/matthart1983/diskwatch.git && cd diskwatch
cargo build --release
./target/release/diskwatch

Prerequisites: Rust 1.75+ (only if building from source). No system dependencies on Linux. macOS calls the standard ioreg, diskutil, and system_profiler binaries — all preinstalled. Optional: smartmontools (brew install smartmontools / apt install smartmontools) for full SMART attribute tables — without it, the SMART tab falls back to the basic verified/failing flag from diskutil.

Keys

Themes

Seven built-in themes, matching syswatch's set and cycle order:

dark (default) · light · ocean · solarized · dracula · nord · terminal

terminal pins no colors of its own — every slot resolves to an ANSI palette entry, and foreground and background use your terminal's own defaults. If you theme your whole desktop with pywal, matugen, or a terminal profile, this is the one that follows along.

Switch it live from the settings overlay (, → Theme → Space), or start with it:

diskwatch --theme terminal      # also accepts: system, ansi

The choice isn't persisted between runs — like the other settings, it resets to dark on restart. Use the flag to make it stick.

Graph style

Every chart in the app — the Overview aggregate, the IO tab's per-device panels, and both Lite charts — draws through one renderer, so a single setting changes them all.

diskwatch --graph dots                 # btop-style braille (also: braille, btop)
diskwatch --graph dots --graph-fade    # ...plus btop's gradient and dot grid

bars is the stacked eighth-block look diskwatch shipped with — eight levels per row. dots packs four braille pixel rows into each cell, so a one-row sparkline resolves four distinct heights instead of reading as on/off. On the tall charts the difference is subtler; on Lite's 9-column row sparklines it's the difference between a shape and a blob.

--graph-fade is btop's other half: each chart runs bright at now and dims toward the left edge, over a faint dot grid. It's a separate switch because the gradient reads fine under bars too — and because it interpolates in RGB, so it's ignored under --theme terminal, which exists precisely to pin no RGB. The grid only draws on charts at least 16×4, which means the IO and Overview panels get it and Lite's 3- and 2-row charts don't.

Both are live in the settings overlay (, → Graph style / Graph fade → Space). Lite has no settings overlay of its own — press L, change it, press L back.

Tabs in detail

[1] Overview — 5 KPI tiles (capacity, IO, p99 latency, health, insights), per-device summary, aggregate IO sparkline, top insights, segmented capacity bar.

[2] Devices — block-device table with model, firmware, serial, used %, SMART status. Detail panel for the selected device.

[3] Volumes — APFS containers (macOS) with nested volumes, role, mount, FileVault. mdraid arrays (Linux) with members, slot state [UUUU], resync/recovery progress.

[4] FS — mounted filesystems with inline usage bars, threshold colors, system/user/removable classification.

[5] IO — per-device read / write throughput, 48s sparkline, p50 + p99 latency (read and write) over a 60s rolling window.

[6] SMART — full NVMe / ATA attribute tables when smartctl is on PATH; degraded banner with install instructions when not. Always shows the basic verified/failing flag.

[7] Hot Files — paths by event rate via FSEvents (macOS) / inotify (Linux). Honest footer: this tab can't show bytes/sec or process attribution without root (fs_usage) / Endpoint Security entitlement / eBPF biosnoop.

[8] Insights — anomaly cards over the collected state: capacity warnings, SMART failures, NVMe wear, drive temperature, p99 latency outliers, IO-dominant devices, hot-file runaway, removable drives.

What's real, what's deferred

The 2.0 view renders -- for anything the platform can't measure and keeps the column, so the layout never shifts between machines. Utilisation is the one that matters: macOS has no time-with-IO-in-flight counter, only summed service time, which on a deep-queue NVMe exceeds wall clock and would read as a permanent 100%. Rather than print that, the devices table sorts by throughput there — and says sort ↓ throughput in its own border, so the indicator can never advertise a sort that didn't happen.

Its latency histogram is honest about its resolution too: each 200ms sample contributes its whole op count to the bucket containing that sample's mean service time. It will show a sustained slow stretch; it will smear a single 50ms outlier into whatever its tick averaged. That is why the box is subtitled io completion · sampled.

Design

Inherits the *Watch family chrome — #0c1418 background, terminal-green accent, JetBrains Mono, 130×36 character grid with responsive reflow ≥ 110×30. The same character-grid mockups that drive NetWatch and SysWatch drive DiskWatch.

Anti-goals

• Not multi-host. Use NetWatch Cloud if you need a fleet view.
• Not a daemon. No long-running collector, no persisted DB.
• Not a deduper / cleaner. We surface what's eating disk; we don't delete anything. Mutation is a different tool.
• Not a backup product. Snapshots are observed, not authored.
• Not a benchmark. We measure what's happening, not what's possible.

License

MIT.