Found by reading what the macOS CI runner prints, which is the whole reason that step was added an hour ago. First time anyone had looked at what these probes return on a platform that is not this laptop: / 96 GiB free of 320 GiB /System/Volumes/VM 96 GiB free of 320 GiB /System/Volumes/Preboot 96 GiB free of 320 GiB /System/Volumes/Update 96 GiB free of 320 GiB /System/Volumes/Data 96 GiB free of 320 GiB 11 filesystems · 483 GiB free of 1600 GiB total On a machine with 320 GiB. An APFS container presents each volume as its own filesystem with its own `/dev/diskNsM`, so the device-name dedupe -- which correctly collapses btrfs subvolumes -- cannot see that the space is shared. Two changes. `/System/Volumes` and `/private/var/vm` are skipped: they are not user storage, and on a Mac they are most of the rows. And filesystems are deduped by *pool* as well as by device. Two filesystems reporting byte-identical capacity and byte-identical free space are one store, whatever their device names say -- which also covers bind mounts and thin-provisioned LVM, neither of which the device check catches either. Two genuinely separate disks agreeing to the byte on both figures would cost one row; a storage total inflated fivefold is a number a capacity plan gets built on. Simulated against the exact runner output: eight rows and 2560 GiB become one row and 320 GiB. The Windows runner, by contrast, was correct first time -- `C:` and `D:` are genuinely separate and totalled 179 GiB free of 299 GiB. Worth recording that the reason we know is that we looked, rather than that we reasoned about it.
nesdoctor
Is this machine any good — as a Nestri host, or as a client?
# macOS and Linux
curl -fsSL https://doctor.nestri.io/install.sh | sh
# Windows
powershell -c "irm https://doctor.nestri.io/install.ps1 | iex"
Both download one binary, verify its checksum, run it, and delete it. They
install nothing, need no administrator rights, and touch no system directory.
The scripts those URLs serve are the files in install/ — the
worker fetches them from this repository, so you can read exactly what you are
about to run before you run it.
They pin a release tag rather than using releases/latest, because this
repository ships product releases too and latest is whichever went out most
recently. NESDOCTOR_TAG overrides it.
Or build it yourself:
cargo run --release -p nesdoctor
Checks every hard requirement for running a Nestri box, measures what your connection actually does when it is busy, and asks at most five questions.
Nothing is uploaded
There is no server. No telemetry endpoint exists, and there is no build of this
program that reports home — the network test talks to Cloudflare's public
speed-test sink and to 1.1.1.1, neither of which is ours.
The output is a line on your terminal. If you want us to have it, you paste it somewhere. If you do not, we never had it. That is a property of the design rather than a promise about our intentions: there is nothing to switch on later.
The shareable line carries no hostname, no IP, no username, no game titles and no file paths — a size band rather than a size, and an hour histogram rather than timestamps. It is put on your clipboard at the end so pasting it is one keystroke.
The long version lands in nesdoctor.json: every check with its reason, the
full latency series, and your installed titles with sizes and launch times if
you said yes to Steam. That file is considerably more useful to us than the
line — it is what lets us size a game library and see which requirement
actually stops people — so do have a read through it and send it along if
nothing in there bothers you. Plain JSON, entirely optional, and the one-line
version is already plenty.
Reading your Steam library needs an explicit yes, and the question is asked last, after you have seen what this program does.
The number worth running it for
Everybody knows their download speed. Almost nobody has seen how much latency their connection adds when it is busy, and for anything interactive that is the figure that decides it:
upstream 39 Mbps
latency, idle floor 55 ms
latency, idle typ. 180 ms
latency, loaded 95 ms
added under load +39 ms grade C
Two idle figures because they answer different questions. Floor is the best the path can do, and queueing is everything above it — so that is what the bloat number is measured against. Typical is what a connection actually gets, and where the two differ this much, the route itself is the problem.
A 500 Mbps uplink that queues for 300 ms under load cannot carry a game. A
25 Mbps one with fq_codel or CAKE can. If your grade is C or F it is almost
always a router setting rather than a line you need to upgrade.
Options
--no-net |
Skip the network test (it uploads ~100 MB) |
--no-steam |
Never look at Steam, and do not ask |
--yes |
Take the defaults — for a second run, not a first |
--json <PATH> |
Where to write the full report |
--quiet |
Only the summary line, for scripting |
--no-open |
Do not offer to open a browser; just print the link |
--submit-url |
Where the submit link points (default https://doctor.nestri.io) |
Verdicts
HOST-READY |
Passes everything, and close enough to the network to serve others |
HOST-READY-LOCAL |
Passes everything, but far enough out that it can only serve players nearby |
HOST-NET |
Good machine; the connection is in the way |
HOST-FIXABLE |
Nothing is a hardware limit — what is missing can be installed |
CLIENT |
Not a host. A complete answer, and what most machines are |
UNKNOWN |
A blocking check could not be run. An unknown is not a no |
Your display, and why we ask
presentation path x11 · bspwm
eDP-1 1920x1200 @ 60 Hz, 8-bit
Vulkan decode h264, h265
VA-API decode h264, h265, vp9
Read from your monitor's EDID and your session, not guessed. Resolution, refresh, colour depth, which HDR transfer functions the panel accepts, whether it takes BT.2020, whether it takes 4:2:0 chroma — plus what your hardware can decode.
This decides real choices on our side: whether 10-bit is worth sending, whether BT.2020 is worth encoding, which codec to reach for. Every one of those had been decided against the single panel in one room.
Its other use is attribution. Told only that a stream "looks bad", the cheapest explanation is always that we compressed it too hard — so without knowing that a compositor is rescaling the picture, we would turn down our own quality to pay for somebody else's window manager.
Present mode, tearing and fractional scaling are not here. They need a real window and a swapchain, so they belong in the client, and are reported as unknown rather than guessed.
Sending it back
At the end it prints a link, lists in plain English what the link contains, and opens it when you press Enter. That is the whole submission — no account, no form, no email client.
One thing worth knowing if you are reading install/install.sh: it reopens
stdin on /dev/tty before handing over. Piped into a shell the documented way,
the script's stdin is the pipe and the pipe is at end of file, so without
that line nesdoctor correctly sees a non-terminal stdin and skips every
question — a run that completes, looks fine, and answers nothing. Running the
script from a file works perfectly, which is what makes it worth a comment.
The link is built from readable query parameters rather than an encoded blob. A blob would be shorter and would let us send more; it would also mean you cannot read what you are sending, which is the one thing this program has going for it.
If you would rather not click a link we wrote, the short line is printed too and put on your clipboard.
At the very end it offers to take an email address — optional, blank skips it — so we can come to you when there is something to try. That is the only identifying thing this program collects, it is asked last, and it appears in the disclosure list like everything else so you see it before it is sent.
What it deliberately does not tell you
- A pass is not a promise. Every check is a necessary condition. Nothing here runs under load, so a machine that passes can still fail on block I/O.
vulkaninforeporting the encode extension is not proof the path works. We have had a correct extension list over a broken path before.- The host's
virglrendererand Mesa are not checked at all, on purpose. The box carries its own inside the image it runs in, so the host's copies are not on the path — and a row that could only ever say "present, patch state unknown" told prospective hosts their machine was wrong when it was not.
Building
cargo build --release -p nesdoctor
Four dependencies, three of them serde/clap/anyhow. Everything that could
be done with std is: the VDF parser, the platform probes and the text wrapping
are all in-tree, because a binary handed to strangers has a dependency tree that
is part of its interface.
A static build, for a release someone downloads rather than compiles:
cargo build --release -p nesdoctor --target x86_64-unknown-linux-musl