devshell: Fix ssh-machine caching in jj workspaces
CI / Check, build and cache nixfiles (push) Successful in 48m33s
Update docs / update (push) Successful in 1m15s

This commit is contained in:
2026-11-18 21:08:24 +00:00
parent fc92ff9ba1
commit 4307939aab
4 changed files with 20 additions and 13 deletions
+2 -2
View File
@@ -58,8 +58,8 @@ Common ones:
- `ssh-machine <name> [cmd]` — SSH to a NixOS system or home-manager config by name. Resolves the - `ssh-machine <name> [cmd]` — SSH to a NixOS system or home-manager config by name. Resolves the
target and ssh options (identity, port) from its deploy-rs node, so it needs `my.deploy.enable` target and ssh options (identity, port) from its deploy-rs node, so it needs `my.deploy.enable`
(same gate as `deploy`). Connection settings for all nodes are cached together and invalidated (same gate as `deploy`). Connection settings for all nodes are cached together and invalidated
by tracked Git changes; use `ssh-machine --refresh` to force a refresh by checkout changes (the jj working copy, else tracked Git changes); use
(see `docs/deployment.md#ssh-machine`). `ssh-machine --refresh` to force a refresh (see `docs/deployment.md#ssh-machine`).
Boxes default to the `fish` login shell, so pipe multi-statement remote Boxes default to the `fish` login shell, so pipe multi-statement remote
scripts through `bash` (e.g. `ssh-machine <name> bash -s < script.sh`) rather than `&&`/`for`. scripts through `bash` (e.g. `ssh-machine <name> bash -s < script.sh`) rather than `&&`/`for`.
If outbound SSH hangs at the publickey step (flaky `ssh-agent`), disable the agent for the call: If outbound SSH hangs at the publickey step (flaky `ssh-agent`), disable the agent for the call:
+1 -1
View File
@@ -4,7 +4,7 @@ let
devKey = ".keys/dev.key"; devKey = ".keys/dev.key";
sshMachine = pkgs.writeShellApplication { sshMachine = pkgs.writeShellApplication {
name = "ssh-machine"; name = "ssh-machine";
runtimeInputs = with pkgs; [ coreutils git jq openssh ]; runtimeInputs = with pkgs; [ coreutils git jujutsu jq openssh ];
text = builtins.readFile ./ssh-machine.sh; text = builtins.readFile ./ssh-machine.sh;
}; };
in in
+11 -5
View File
@@ -20,11 +20,17 @@ cache_dir=.cache/ssh-machine
cache=$cache_dir/connections.json cache=$cache_dir/connections.json
fingerprint() { fingerprint() {
{ # A jj working-copy commit's ID covers its whole snapshotted tree, so it changes with any
git rev-parse HEAD && # working-copy edit. This also works in secondary workspaces, which have no Git checkout.
git diff --no-ext-diff --no-textconv --binary HEAD -- && if jj root >/dev/null 2>&1; then
git diff --no-ext-diff --no-textconv --binary --cached -- jj log --no-pager --color never --no-graph -r @ -T 'commit_id'
} | sha256sum | cut -d ' ' -f 1 else
{
git rev-parse HEAD &&
git diff --no-ext-diff --no-textconv --binary HEAD -- &&
git diff --no-ext-diff --no-textconv --binary --cached --
} | sha256sum | cut -d ' ' -f 1
fi
} }
key=$(fingerprint) key=$(fingerprint)
+6 -5
View File
@@ -60,11 +60,12 @@ Generated profiles deploy in this order:
`ssh-machine <name> [cmd]` resolves `user@host` and the merged global/node `sshOpts` from `ssh-machine <name> [cmd]` resolves `user@host` and the merged global/node `sshOpts` from
`.#deploy`, using the same `my.deploy.enable` gate and `@` → `-at-` mangling as `deploy`, then runs `.#deploy`, using the same `my.deploy.enable` gate and `@` → `-at-` mangling as `deploy`, then runs
`ssh`. Connection settings for all nodes are cached together in `.cache/ssh-machine/`; `ssh`. Connection settings for all nodes are cached together in `.cache/ssh-machine/`;
normal connections read the cache without running Nix. A fingerprint of Git HEAD and normal connections read the cache without running Nix. A fingerprint of the checkout
staged/unstaged tracked changes automatically invalidates it, including newly added files automatically invalidates it. In a Jujutsu workspace (including secondary workspaces without a
after `git add`. Git checkout) it is the working-copy commit ID, so any snapshotted change counts, including new
Untracked and ignored files do not affect the fingerprint. Any tracked edit, even an unrelated files that are not ignored. In a Git-only checkout it is HEAD plus staged/unstaged tracked
one, causes the next connection to refresh all nodes. Use `ssh-machine --refresh` to force a changes; untracked files only count after `git add`. Ignored files never affect the fingerprint.
Any such edit, even an unrelated one, causes the next connection to refresh all nodes. Use `ssh-machine --refresh` to force a
refresh, or `ssh-machine --refresh <name> [cmd]` to refresh and connect. Refresh failures stop refresh, or `ssh-machine --refresh <name> [cmd]` to refresh and connect. Refresh failures stop
the command; the previous cache is never used as a fallback. the command; the previous cache is never used as a fallback.