mirror of
https://github.com/kellymichels/zscripts-token-savers
synced 2026-10-07 07:18:18 +00:00
chore(sync): bring zdeploy/ZHelpers up to the private tree, sanitized (#64)
The mirror's deploy pair had drifted ~280 lines behind: it lacked the
transactional .env preserve/restore (an interrupted deploy could destroy
server-side env files), the stderr-flattening step wrapper (a successful
deploy reported failure and skipped its own verification), and the
verification rework (channel re-picked every retry, edge only with a Host
to route by, verify.timeoutSeconds, honest split of "stale build" vs "no
channel answered").
The sync is byte-faithful to the private tree except where the mirror's
own Sanitization suite demands otherwise - and it caught the first copy:
three failures for private project names, a private domain, and a
private-only script name that rode along in comments. Each war story keeps
its lesson and loses its cast, per the convention already in the file
("EvoCivilCode: deploy/" was already published as "(deploy/, infra/,
...)"). That suite going red on an unsanitized copy is exactly what it
exists for.
Also in this change:
- tests/VerifyPlan.Tests.ps1 - the channel-selection rules are pure
functions and Pester pins them (no domain => no edge attempt; the PS 5.1
one-element-unroll trap). First verification tests in the mirror.
- README: the verify block now documents viaProxy/upstream (they shipped
in the docker-network read but were never in the README),
timeoutSeconds, and the channel order with why it re-resolves per retry.
- README.txt: generated plain-text twin, via scripts/readme_txt.py
(vendored from the fleet's reference implementation; the file is
generated, never edited by hand).
- CHANGELOG.md/.txt: entries merged into the existing Unreleased sections.
CHECKSUMS.txt refreshed (42 entries). Pester: 240 passed, 0 failed.
Both synced files parse clean.
Observed, untouched: the Unreleased section carries duplicate "### Fixed"
headings from earlier appends; folding them risks reordering entries whose
prose references their neighbours, so it is left for the next release cut
(zbump #110 rolls Unreleased into the version being cut).
This commit is contained in:
GitHub
parent
471789f530
commit
ac95c47255
35
CHANGELOG.md
35
CHANGELOG.md
@ -11,6 +11,33 @@ Notable changes to the Evomedia.net Token Savers.
|
||||
## Unreleased
|
||||
|
||||
### Changed
|
||||
- **`zdeploy` verification picks its channel on every retry, and never asks
|
||||
the bare IP** — the check used to choose its channel once, before the wait
|
||||
loop, by probing; the probes raced the app restart the check exists to wait
|
||||
through, so the whole window went to the edge fallback. For a project with
|
||||
no `domain` that fallback had no `Host` header, and the proxy can then only
|
||||
answer from its **default vhost — a different product**: that is how one
|
||||
deploy's check compared another app's build number against its own. Now
|
||||
channels are re-resolved each retry in trust order (docker-network
|
||||
`viaProxy` → `localhost:<port>` → edge with the project's `Host`), the edge
|
||||
is skipped entirely when there is no host to route by, and a project with
|
||||
no trustworthy channel is reported as unverifiable instead of guessed at.
|
||||
The final warning also says which failure happened: a version that never
|
||||
matched (stale/failed build) reads differently from channels that never
|
||||
answered (probably still booting). `verify.timeoutSeconds` joins the
|
||||
config so a project that is slow to boot — e.g. one that runs database
|
||||
migrations in its entrypoint — can widen its own window instead of
|
||||
warning on every routine success.
|
||||
- **An interrupted deploy can no longer destroy server-side `.env` files** —
|
||||
the preserve/restore of operator files is transactional: the restore comes
|
||||
from a tarball taken before the tree is replaced, so a deploy that dies
|
||||
mid-flight leaves the previous files in place instead of an empty
|
||||
directory.
|
||||
- **A successful deploy no longer reports failure** — `docker compose
|
||||
restart` writes routine progress to stderr, which PowerShell 5.1 turns
|
||||
into a terminating error under `$ErrorActionPreference = 'Stop'`; four ssh
|
||||
calls bypassed the wrapper that flattens this. All remote steps now run
|
||||
through it and are judged by exit code alone.
|
||||
- **`zversion bump` is once per *release*, not once per PR** — the usage text
|
||||
and the `bump` help line both said "one per PR, one per defect fix". The
|
||||
build number names something that shipped, so a release carrying five PRs
|
||||
@ -21,6 +48,14 @@ Notable changes to the Evomedia.net Token Savers.
|
||||
written.
|
||||
|
||||
### Added
|
||||
- **`tests/VerifyPlan.Tests.ps1`** — the verification channel-selection rules
|
||||
are pure functions in `ZHelpers.ps1` (`Get-VerifyAttempts`,
|
||||
`Get-VerifyTimeout`) and Pester pins them, including "a project with no
|
||||
domain must never produce an edge attempt" and the PowerShell 5.1
|
||||
one-element-unroll trap.
|
||||
- **`scripts/readme_txt.py` and `README.txt`** — a generated plain-text twin
|
||||
of the README for terminals and pagers. `README.txt` is generated, never
|
||||
edited by hand.
|
||||
- **Read a live build from inside the docker network, not through the public
|
||||
proxy** — `zdeploy`, `zec2` and `zec2online` now prefer
|
||||
`docker exec <viaProxy> curl http://<upstream>/api/build-version` when a
|
||||
|
||||
@ -10,6 +10,33 @@ Notable changes to the Evomedia.net Token Savers.
|
||||
Unreleased
|
||||
----------
|
||||
Changed
|
||||
- zdeploy verification picks its channel on every retry, and never asks
|
||||
the bare IP - the check used to choose its channel once, before the wait
|
||||
loop, by probing; the probes raced the app restart the check exists to
|
||||
wait through, so the whole window went to the edge fallback. For a
|
||||
project with no domain that fallback had no Host header, and the proxy
|
||||
can then only answer from its default vhost - a different product: that
|
||||
is how one deploy's check compared another app's build number against
|
||||
its own. Now channels are re-resolved each retry in trust order
|
||||
(docker-network viaProxy -> localhost:<port> -> edge with the project's
|
||||
Host), the edge is skipped entirely when there is no host to route by,
|
||||
and a project with no trustworthy channel is reported as unverifiable
|
||||
instead of guessed at. The final warning also says which failure
|
||||
happened: a version that never matched (stale/failed build) reads
|
||||
differently from channels that never answered (probably still booting).
|
||||
verify.timeoutSeconds joins the config so a project that is slow to
|
||||
boot - e.g. one that runs database migrations in its entrypoint - can
|
||||
widen its own window instead of warning on every routine success.
|
||||
- An interrupted deploy can no longer destroy server-side .env files - the
|
||||
preserve/restore of operator files is transactional: the restore comes
|
||||
from a tarball taken before the tree is replaced, so a deploy that dies
|
||||
mid-flight leaves the previous files in place instead of an empty
|
||||
directory.
|
||||
- A successful deploy no longer reports failure - docker compose restart
|
||||
writes routine progress to stderr, which PowerShell 5.1 turns into a
|
||||
terminating error under ErrorActionPreference = Stop; four ssh calls
|
||||
bypassed the wrapper that flattens this. All remote steps now run
|
||||
through it and are judged by exit code alone.
|
||||
- zversion bump is once per RELEASE, not once per PR - the usage text and
|
||||
the bump help line both said "one per PR, one per defect fix". The build
|
||||
number names something that shipped, so a release carrying five PRs moves
|
||||
@ -20,6 +47,14 @@ Changed
|
||||
written.
|
||||
|
||||
Added
|
||||
- tests/VerifyPlan.Tests.ps1 - the verification channel-selection rules
|
||||
are pure functions in ZHelpers.ps1 (Get-VerifyAttempts,
|
||||
Get-VerifyTimeout) and Pester pins them, including "a project with no
|
||||
domain must never produce an edge attempt" and the PowerShell 5.1
|
||||
one-element-unroll trap.
|
||||
- scripts/readme_txt.py and README.txt - a generated plain-text twin of
|
||||
the README for terminals and pagers. README.txt is generated, never
|
||||
edited by hand.
|
||||
- Read a live build from inside the docker network, not through the public
|
||||
proxy — zdeploy, zec2 and zec2online now prefer
|
||||
docker exec <viaProxy> curl http://<upstream>/api/build-version when a
|
||||
|
||||
@ -8,14 +8,14 @@ e35d91a175c29dcbe285b55397371a584000aab9ae3de6839f8b31def1d9dfd0 token-count.ps
|
||||
49d48d55bab1b9ee5bc66cb96340a20fa50241b6a3558d4c518db3e06d4553e6 zchecksums.cmd
|
||||
e0cbc6f263c129d89e87e8e7356fb256f59f21a6a4b209e191e7c70a3611286e zchecksums.ps1
|
||||
bd5563bf74b83423aca49a56450f3712a9aec3d6a59f8b7862d5c5988c6defd2 zdeploy.cmd
|
||||
4a55ed714358b910fd2f326a6222e2ebaa99a1321df025a19edb6bbe0da13e76 zdeploy.ps1
|
||||
869efc42e86c4dd0c208e88598abcf9e3226afd5a4eb3fc4346ad81e933ef177 zdeploy.ps1
|
||||
1883d7307682c68bf3786203f54e9abfe12fcb09e3856190e3a7af5fffb2a16a zec2.cmd
|
||||
f9d0406f97e19303d8b368df8aaf70e5011a99703f67ca7b2b526e82a3c68789 zec2.ps1
|
||||
e1da88e4af6d5c88cc95231dc544ef29440d306e70739d0ba1f72795d359778d zec2_rotatekeys.cmd
|
||||
ea2fcf081491dbb98794bc2ef6ce41563fef8a7c3553840e75aff757ea59d582 zec2_rotatekeys.ps1
|
||||
b30b10ca6d7930021f565a3d90eda967f1ca18366b30a671eab0668f6d5695eb zec2online.cmd
|
||||
5b8c3bfc707255f5782ea6906075817f5bfe8d77fb1a4299a33581d0c70b9569 zec2online.ps1
|
||||
d550f9866c1f746da065021e5150566b3aebc07a51aa01ba6459f29560765b23 ZHelpers.ps1
|
||||
860e51adf035232b9242cff8866e31bb2dfe5117e393e0c698e958ef2f67d701 ZHelpers.ps1
|
||||
1167972175b9d60613d1b8da55053c13a390037fd075fd8e411374d7a402ff72 zkill.cmd
|
||||
0a4c09c2a85bfdd8577ee43179128fc6b14335e5232a2b172f4558542b8706f6 zkill.ps1
|
||||
b81c06fc85045b69b298450b84757c1997ea14354e9774d579c27e41e2f43515 ZKiller.ps1
|
||||
|
||||
15
README.md
15
README.md
@ -378,6 +378,21 @@ Give the project a `verify` block instead, and `zdeploy` checks the app **from t
|
||||
|
||||
`port` is the host port the app publishes on the server; `path` defaults to `/`; `expect` is an optional substring the response must contain (an app version string makes this equivalent to build-number verification). Python-kind projects use `verify` automatically when there's no `build_version_tool.py` — and projects with *neither* a `domain` nor a `verify` block are now honestly reported as **NOT verified** instead of green-lighting the proxy's default page.
|
||||
|
||||
Two more keys make the check unambiguous and patient:
|
||||
|
||||
```json
|
||||
"verify": {
|
||||
"port": 8005, "path": "/health",
|
||||
"viaProxy": "my_edge_proxy", "upstream": "myapp_container:8000",
|
||||
"timeoutSeconds": 120
|
||||
}
|
||||
```
|
||||
|
||||
- **`viaProxy` + `upstream`** — read the version **over the docker network**, by exec-ing a curl inside the named container (usually the edge proxy, since it is on every stack's network). This is the most trustworthy channel there is: it cannot answer from the wrong product, and it works for apps that publish no host port at all.
|
||||
- **`timeoutSeconds`** — how long verification may wait. An app that runs database migrations in its entrypoint exceeds a 30-second window *on every deploy that ships one*, and a warning that fires on routine success teaches you to ignore the one that matters.
|
||||
|
||||
Verification walks its channels in trust order — docker-network `viaProxy`, then `localhost:<port>`, then the edge with the project's own `Host` header — and **re-picks the channel on every retry**. That last part is the point: the check runs while the app is restarting, which is exactly when the good channels are briefly down, and locking the choice in up front is how a whole verification window gets spent asking the wrong thing. A project with no domain is **never** asked for via the edge: without a `Host` header the proxy can only answer from its default vhost — a different product — and no answer beats somebody else's answer.
|
||||
|
||||
`zec2` and `zec2online` use these same endpoints to show what's live and flag local/server version drift.
|
||||
|
||||
---
|
||||
|
||||
436
README.txt
Normal file
436
README.txt
Normal file
@ -0,0 +1,436 @@
|
||||
<!--
|
||||
Evomedia.net Token Savers — https://github.com/evomedia-net/evo.zscripts
|
||||
Created by Kelly Michels · dev@evomedia.net
|
||||
Licensed under the MIT License. See LICENSE.
|
||||
-->
|
||||
|
||||
zscripts Token Savers
|
||||
=====================
|
||||
|
||||
When an AI coding agent orchestrates your infrastructure — starting dev servers, deploying to EC2, diagnosing 502s — it spends hundreds to thousands of tokens per operation on SSH plumbing, Docker output, and retry logic. Those tokens should go to code.
|
||||
|
||||
Token Savers gives you short, one-word commands to run those parts yourself: zdeploy myapp, zrepair myapp, zstart myapp. You handle the deterministic infrastructure; your agent handles code. Running these scripts manually instead of asking your agent to orchestrate them keeps measured script output out of your agent's context window — ~26,500 tokens per active development day at a typical run cadence. Per-run figures are measured; the daily total applies typical run counts. See TOKEN_SAVINGS.md (TOKEN_SAVINGS.md) for the numbers and method.
|
||||
|
||||
Every command is a tiny PowerShell script driven by a single JSON config file. The project key you define in that config is the command argument — add myapp to the config and zstart myapp, zdeploy myapp, zbackup myapp all just work, no script edits needed.
|
||||
|
||||
Requirements: Windows, PowerShell 5.1+, OpenSSH client (ssh/scp, ships with Windows 10/11), and Docker + docker compose on the remote host for the deploy scripts.
|
||||
|
||||
---
|
||||
|
||||
Install
|
||||
-------
|
||||
|
||||
No installer. Clone the repo and add the folder to your PATH:
|
||||
|
||||
git clone https://github.com/evomedia-net/evo.zscripts.git C:\tools\zscripts
|
||||
|
||||
# Add to your user PATH (new terminals pick it up automatically)
|
||||
[Environment]::SetEnvironmentVariable(
|
||||
"Path",
|
||||
[Environment]::GetEnvironmentVariable("Path", "User") + ";C:\tools\zscripts",
|
||||
"User"
|
||||
)
|
||||
|
||||
Open a new terminal and every command below works from any directory. The .cmd wrappers invoke PowerShell with -ExecutionPolicy Bypass, so no execution-policy changes are needed — zstart myapp just works from cmd, PowerShell, or a VS Code terminal.
|
||||
|
||||
---
|
||||
|
||||
Configure
|
||||
---------
|
||||
|
||||
All machine-specific values (server IP, SSH user and key, folder paths, project definitions) live in one file: zconfig.json. It is gitignored — your secrets never leave your machine.
|
||||
|
||||
cd C:\tools\zscripts
|
||||
copy zconfig.example.json zconfig.json
|
||||
notepad zconfig.json
|
||||
|
||||
The example config ships with sample projects named by their kind — pyapp, viteapp, nextapp, edge, analytics. Rename the keys to your own project names; the key is what you type as the command argument. Add as many projects as you like — no script edits ever needed.
|
||||
|
||||
Set the ZCONFIG environment variable to point at a config somewhere else — handy for a second machine profile, or for running against a scratch config without touching your real one.
|
||||
|
||||
Config reference
|
||||
----------------
|
||||
|
||||
{
|
||||
"ec2": {
|
||||
"ip": "203.0.113.10", // your server's public IP
|
||||
"user": "youruser", // SSH user on the server
|
||||
"pemKey": "C:\\Users\\You\\.ssh\\key.pem", // path to your SSH private key
|
||||
"stackRoot": "/home/youruser/stack" // parent dir for all deployed projects
|
||||
},
|
||||
"paths": {
|
||||
"temp": "C:\\dev\\temp", // deploy zips staged here (auto-deleted)
|
||||
"backupsLocal": "C:\\dev\\backups\\projects", // zbackup output
|
||||
"backupsEc2": "C:\\dev\\backups\\ec2", // zbackup_ec2 output
|
||||
"scriptsRoot": "C:\\tools\\zscripts", // this folder
|
||||
"oneDriveBackups": "" // zsync destination ("" disables)
|
||||
},
|
||||
"projects": {
|
||||
"myapp": {
|
||||
"label": "My App", // display name in output
|
||||
"kind": "python", // python | vite | nextjs | edge | docker
|
||||
"localRoot": "C:\\dev\\myapp", // project folder on this machine
|
||||
"startModule": "myapp.main", // python kind: runs "python -m myapp.main"
|
||||
// "startApp": "app.main:app", // ...or, for ASGI/FastAPI: uvicorn app.main:app --port <dev> --reload
|
||||
"install": "-e .", // optional: pip args `zsetup` uses (auto-detects "-e ." / "-r requirements.txt")
|
||||
"ports": { "dev": 8080, "prod": 3000 }, // local dev port / direct server port
|
||||
"domain": "www.myapp.com", // public domain (health checks + verification)
|
||||
"start": { // optional zstart pre-steps
|
||||
"gitPull": true, // git pull --ff-only before starting
|
||||
"env": { "MYAPP_DEBUG": "1" } // env vars for the dev server process
|
||||
},
|
||||
"db": { "user": "dbuser", "name": "dbname" }, // optional: enables db dump/wait steps
|
||||
"migrations": "prisma", // optional: run prisma migrate on deploy
|
||||
"remote": {
|
||||
"path": "/home/youruser/stack/myapp", // deploy target on the server
|
||||
"composeDir": "/home/youruser/stack/myapp/docker", // optional: if compose isn't at path root
|
||||
"appService": "app", // optional: compose service name override
|
||||
"containerName": "myapp" // optional (vite): container to read build-version from
|
||||
},
|
||||
"deploy": {
|
||||
"zipName": "MyAppDeploy.zip", // optional: defaults to <key>Deploy.zip
|
||||
"gitPull": true, // optional: git pull --ff-only before zipping
|
||||
"exclude": ["docs", "big-data-folder"] // optional: extra top-level dirs/files to skip
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Optional blocks do real work:
|
||||
|
||||
- install — how zsetup installs a python project's dependencies into its .venv: the pip args, e.g. "-e .", "-e backend" (deps in a subfolder), or "-r requirements.txt". Omit it and zsetup auto-detects a root pyproject.toml/setup.py (-e .) or requirements.txt (-r requirements.txt). zstart never installs — run zsetup <key> once, then zstart <key>.
|
||||
- start — pre-start steps for zstart: gitPull: true runs git pull --ff-only in the project root first (never starts a stale checkout), and env sets environment variables for the dev-server process (feature flags, reload switches).
|
||||
- db — deploys wait for pg_isready and zbackup_ec2 pulls a pg_dump, both against the compose service named db. Omit it and those steps are skipped cleanly.
|
||||
- deploy.gitPull — git pull --ff-only in the project root before zipping, so a merged PR actually ships. Since zdeploy zips your working tree, a checkout left behind origin would otherwise deploy stale code and still bump the build number — a silent no-op that looks like success. A failed pull (dirty tree that conflicts, diverged history) aborts the deploy rather than shipping uncertain code.
|
||||
- migrations": "prisma" — runs npx prisma migrate deploy inside the app container after each deploy.
|
||||
- Compose service-name conventions — handlers assume the app service is named app (python) or web (nextjs) and the database service db. Override the app service with remote.appService.
|
||||
- Edge extras — an edge-kind project can set proxyContainer (the nginx container's name, used for reloads and stale-container cleanup) and certsSource (a host path with TLS certs, mounted read-only when validating nginx.conf).
|
||||
|
||||
---
|
||||
|
||||
Commands
|
||||
--------
|
||||
|
||||
The .cmd wrappers are the everyday interface. Every command takes one or more project keys from your config; several also accept all. A leading dash is tolerated (zdeploy -myapp works the same as zdeploy myapp) for anyone with switch-style muscle memory.
|
||||
|
||||
| Command | What it does |
|
||||
|---|---|
|
||||
| zsetup <key> ... | Create the project's Python venv + install deps (or npm install for node) |
|
||||
| zstart <key> ... | Start local dev server(s) |
|
||||
| zstartd <key> ... | Same, detached (new window, returns immediately) |
|
||||
| zkill <key> ... \| all | Kill local dev server(s) by port |
|
||||
| zrestart <key> ... | Kill + start in one step |
|
||||
| zrestartd <key> ... | Kill + start detached |
|
||||
| zdeploy <key> ... \| all | Zip → upload → rebuild → verify a project on the server |
|
||||
| zec2 [<key> ...] | Quick reachability check (TCP + HTTP + live build version) |
|
||||
| zec2online [<key> ...] | Deep health check; auto-starts downed stacks, streams diagnostics |
|
||||
| zrepair <key> ... | Audit + repair compose/proxy state on the server |
|
||||
| zec2_rotatekeys <key> | Rotate/reset secret keys in a project's server-side .env (values generated server-side; never printed) |
|
||||
| zbackup <key> ... \| all | Zip local project sources (+ DB dump) to the backups folder |
|
||||
| zbackup_ec2 [<key> ...] | Pull DB dumps + server-side data files down from the server |
|
||||
| zsync [<key>] | Copy new backups offsite (or build + mirror a vite dist) |
|
||||
| zstart_docker | Run a local docker compose stack from scriptsRoot\docker\ |
|
||||
| zchecksums [-Update] | Verify every script against CHECKSUMS.txt (SHA-256) |
|
||||
| zversion [bump \| bump-stage <s> \| set <v>] | Show or advance the toolkit version (stamps every header) |
|
||||
| zrelease [-Verify] | Package the current version as releases/zscripts-<version>.zip + .sha256 |
|
||||
|
||||
Local development
|
||||
-----------------
|
||||
|
||||
zstart — start dev servers
|
||||
--------------------------
|
||||
|
||||
zstart <project> [<project> ...] [-Port N] [-BindHost <host>] [-Detached]
|
||||
|
||||
Starts each project's dev server using the handler for its kind: python runs python -m <startModule> — or, for an ASGI/FastAPI app, uvicorn <startApp> (e.g. app.main:app) with the dev port and --reload — preferring the project's .venv; vite runs npm run dev -- --host --port, nextjs runs npm run dev with PORT set. Runs npm install automatically if node_modules is missing. A project's optional start config block runs first — gitPull fast-forwards the checkout and env sets process environment variables. Two more opt-in conveniences: if the project has a motd/ folder of .txt files, one is shown (rotating) at startup; if it has scripts/build_version_tool.py, the build number is bumped on each start.
|
||||
|
||||
zstart viteapp # dev server on its configured port
|
||||
zstart pyapp -Port 9000 # override the port
|
||||
zstart viteapp -BindHost 0.0.0.0 # expose on the LAN
|
||||
zstartd nextapp # detached: window opens, prompt returns
|
||||
|
||||
zkill — stop dev servers
|
||||
------------------------
|
||||
|
||||
zkill <project> [<project> ...] | all [-Port N] [-KillAll]
|
||||
|
||||
Finds whatever is LISTENING on each project's dev port and kills it — along with its whole process tree, children first. That matters for auto-reloading servers (uvicorn/watchfiles, nodemon): their worker processes inherit the listening socket and would otherwise survive as orphans, serving stale code. -KillAll also hunts down stray node/python/next-server processes whose command line references the project folder. all targets every project that has a dev port — the one-shot "stop everything I've got running locally".
|
||||
|
||||
zkill viteapp # free the port
|
||||
zkill pyapp viteapp nextapp # nuke everything
|
||||
zkill all # stop every project's dev server
|
||||
zkill nextapp -KillAll # also kill orphaned runtime processes
|
||||
|
||||
zrestart — kill then start
|
||||
--------------------------
|
||||
|
||||
zrestart <project> [<project> ...] [-Port N] [-KillAll] [-NoRestart] [-Detached] [-BindHost <host>]
|
||||
|
||||
The "it's wedged, bounce it" command: kill phase, then start phase with the same flags. -NoRestart makes it kill-only; zrestartd restarts detached.
|
||||
|
||||
Server deployment & operations
|
||||
------------------------------
|
||||
|
||||
zdeploy — deploy to the server
|
||||
------------------------------
|
||||
|
||||
zdeploy <project> [<project> ...] [-Note "message"]
|
||||
zdeploy all [-Note "message"]
|
||||
|
||||
The core workflow, per project kind (projects with deploy.gitPull first git pull --ff-only so a merged PR isn't left behind):
|
||||
|
||||
- python / vite / nextjs — zip the local source (excluding .git, node_modules, envs, archives, junk, plus anything in deploy.exclude), free disk space on the server (docker prune; aborts if under 1.5 GB free), scp the zip up, unzip into remote.path preserving all server-side .env* files plus anything listed in deploy.preserve (staged keys, certs, seed data — files or directories), docker compose build + up -d, then verify the live site reports the new build version (see Enabling deploy verification (#enabling-deploy-verification)). nextjs additionally waits for Postgres (db block) and applies migrations (migrations field). Zips are always deleted locally afterward.
|
||||
- edge — uploads every top-level file in the edge folder (nginx.conf, compose, css, htpasswd, …), validates the new config with nginx -t before switching over, then recreates the proxy.
|
||||
- docker — uploads the compose folder's files, docker compose pull + up -d. For stacks that run stock images (analytics, mail, etc.).
|
||||
|
||||
all deploys every project — edge kinds first, then the rest in config order — and stops at the first failure.
|
||||
|
||||
zdeploy viteapp
|
||||
zdeploy pyapp -Note "fix billing banner"
|
||||
zdeploy all -Note "weekly release"
|
||||
|
||||
zec2 — reachability check
|
||||
-------------------------
|
||||
|
||||
zec2 [<project> ...] # no args = every project with a domain
|
||||
|
||||
For each project: TCP connect, then an HTTP GET with the project's domain as the Host header, then the live build version. Fast "is it up?" answer with firewall hints when it isn't.
|
||||
|
||||
zec2online — health check with auto-recovery
|
||||
--------------------------------------------
|
||||
|
||||
zec2online [<project> ...] # no args = every project with a domain
|
||||
|
||||
The heavier sibling: verifies each app over HTTP, compares the local build version against what the server is actually serving (a mismatch means "redeploy?" — or a stale cache), and if a site is down it SSHes in, runs docker compose up -d for the app and the edge proxy, waits up to 30 s, and streams compose logs and system diagnostics if recovery fails.
|
||||
|
||||
zrepair — fix server routing
|
||||
----------------------------
|
||||
|
||||
zrepair <project> [<project> ...]
|
||||
|
||||
Validates the edge proxy's nginx config (if an edge project is defined), shows each stack's compose status, starts anything that's down, and smoke-tests the live domain. For the "deploy succeeded but the site 502s" class of problem.
|
||||
|
||||
zstop.ps1 — stop server stacks
|
||||
------------------------------
|
||||
|
||||
zstop <project> [<project> ...]
|
||||
|
||||
docker compose down for the selected stacks on the server. Data volumes are preserved; zdeploy <project> brings a stack back. (PowerShell script only, no .cmd wrapper.)
|
||||
|
||||
zec2_rotatekeys — rotate server-side secrets
|
||||
--------------------------------------------
|
||||
|
||||
zec2_rotatekeys <project> [-Rotate KEY,KEY] [-Set KEY,KEY] [-EnvFile rel/path] [-Restart] [-WhatIf]
|
||||
|
||||
For when a secret leaks or a deploy overwrites a production .env with dev values: rotate or reset keys in a project's server-side .env without the values ever passing through this machine's shell history, a command argument, or your screen. -Rotate keys are regenerated on the server with openssl rand -hex 32 — the new value is written straight into the .env there and never leaves the box. -Set keys are typed into a masked prompt and streamed to the server over SSH stdin (never a command argument, never echoed), for operator-known values like DATABASE_URL or ADMIN_EMAIL. The current server .env is copied to a timestamped .bak before any change; the KEY line is updated atomically, matching an existing key or appending it. The env file is auto-detected from the project's deploy.preserve (first *.env) or defaults to .env — override with -EnvFile backend/.env. Nothing touches the running app unless you pass -Restart, which recreates the container (docker compose up -d --force-recreate <svc>) so it actually reloads the new .env — a plain restart would keep the old environment. Being high-impact, it confirms before writing; -WhatIf prints the exact plan and changes nothing.
|
||||
|
||||
# Preview only — see exactly what would change, change nothing:
|
||||
zec2_rotatekeys pyapp -Rotate JWT_SECRET -Set DATABASE_URL,ADMIN_EMAIL -WhatIf
|
||||
|
||||
# Regenerate the JWT secret, restore the operator-known values, then restart:
|
||||
zec2_rotatekeys pyapp -Rotate JWT_SECRET -Set DATABASE_URL,ADMIN_EMAIL,ADMIN_PASSWORD -Restart
|
||||
|
||||
Backups
|
||||
-------
|
||||
|
||||
zbackup — local backups
|
||||
-----------------------
|
||||
|
||||
zbackup <project> [<project> ...] [-Tag "label"]
|
||||
zbackup all # every project + this scripts folder
|
||||
zbackup scripts # just this scripts folder ('scripts' is reserved)
|
||||
|
||||
Zips each project's source into paths.backupsLocal\<key>\<timestamp>_<key>[_tag].zip. If the project's .env declares a DATABASE_URL, a Postgres dump is bundled into the zip automatically — quoted values (Prisma-style), postgres:///postgresql+driver:// schemes, and URLs without an explicit port all parse. backend\.env is checked too, for frontend/backend split projects. -Tag labels the archive — handy before risky changes.
|
||||
|
||||
zbackup all # everything
|
||||
zbackup pyapp -Tag "pre-migration"
|
||||
|
||||
zbackup_ec2 — pull backups from the server
|
||||
------------------------------------------
|
||||
|
||||
zbackup_ec2 [<project> ...] # no args = every project with a remote.path
|
||||
|
||||
For projects with a db block, runs pg_dump inside the server's db container. Also zips server-side data dirs (uploads/, archive/, dist/) when present, then downloads everything to paths.backupsEc2 and cleans up the remote temp files.
|
||||
|
||||
zsync — sync backups offsite
|
||||
----------------------------
|
||||
|
||||
zsync # new backup files -> paths.oneDriveBackups
|
||||
zsync <viteproject> -Destination <path> # npm run build, then mirror dist/ to path
|
||||
|
||||
The no-args mode copies only files that don't already exist at the destination (never overwrites, never deletes). The project mode is for mirroring a static build; it also honors $env:ZSYNC_DEST.
|
||||
|
||||
zbackup_and_sync.ps1 — both in one
|
||||
----------------------------------
|
||||
|
||||
zbackup_and_sync.ps1 <project> [<project> ...] | all
|
||||
|
||||
Runs zbackup, then zsync. This is what the scheduled task calls (with all).
|
||||
|
||||
setup_backup_schedule.ps1 — nightly automation
|
||||
----------------------------------------------
|
||||
|
||||
Run as Administrator once. Creates a Windows Scheduled Task that runs zbackup_and_sync.ps1 all daily at 2:00 AM.
|
||||
|
||||
Utilities
|
||||
---------
|
||||
|
||||
zstart_docker — local compose stack
|
||||
-----------------------------------
|
||||
|
||||
zstart_docker [-Build] [-Attached] [-Solo]
|
||||
|
||||
Brings up a docker compose stack from scriptsRoot\docker\docker-compose.yml (or docker-compose.solo.yml with -Solo). Checks that Docker Desktop is actually running and tells you how to unwedge it if not.
|
||||
|
||||
zsetup_mail.ps1 — provision mail accounts
|
||||
-----------------------------------------
|
||||
|
||||
zsetup_mail.ps1 -domain yourdomain.com [-mailHost mail.yourdomain.com]
|
||||
|
||||
Creates admin@ and noreply@ mailboxes (with generated passwords) in a docker-mailserver container on the server, then prints the exact DNS records (MX, SPF, A) and SMTP/IMAP settings to plug into your app.
|
||||
|
||||
ZHelpers.ps1 — shared library
|
||||
-----------------------------
|
||||
|
||||
Not run directly. Dot-sourced by the other scripts; provides config loading (Get-ZConfig, Get-ZProject), SSH helpers (Invoke-Ec2Step), the deploy/backup archiver (New-ProjectArchive), and process-kill helpers. Extend here if you're adding your own scripts.
|
||||
|
||||
---
|
||||
|
||||
Enabling deploy verification
|
||||
----------------------------
|
||||
|
||||
Why not just check for HTTP 200? Because a 200 proves nothing — a stale cached build serves 200 all day. These scripts verify a deploy by comparing build numbers: your app exposes its build version, the deploy expects to see the new number live, and a mismatch means the upload or Docker build failed (or you're looking at a cached build).
|
||||
|
||||
It's optional — deploys still work without it, ending in a WARNING instead of a PASS — but it's the difference between "the server answered" and "the code I just shipped is actually running."
|
||||
|
||||
1. Add a version file to your project
|
||||
-------------------------------------
|
||||
|
||||
// build-version.json (vite: in public/ · nextjs: in public/ · python: project root)
|
||||
{ "productVersion": "1.0", "buildNumber": 42 }
|
||||
|
||||
2. Bump it during the server-side Docker build
|
||||
----------------------------------------------
|
||||
|
||||
The convention: each deploy's Docker build increments buildNumber by one, so the deploy script expects local buildNumber + 1 to show up live. One line in your Dockerfile does it:
|
||||
|
||||
RUN node -e "const f='public/build-version.json',v=require('./'+f);v.buildNumber++;require('fs').writeFileSync(f,JSON.stringify(v))"
|
||||
|
||||
3. Expose it
|
||||
------------
|
||||
|
||||
Vite / static sites — nothing to do: public/build-version.json is served at /build-version.json, which is where verification looks. (If your edge proxy blocks it from outside, set remote.containerName in config and verification reads it inside the container instead.)
|
||||
|
||||
Next.js — add an API route at /api/build-version:
|
||||
|
||||
// app/api/build-version/route.ts
|
||||
import { NextResponse } from "next/server";
|
||||
import bv from "@/public/build-version.json";
|
||||
|
||||
export async function GET() {
|
||||
return NextResponse.json({ build_version: `v${bv.productVersion}.${bv.buildNumber}` });
|
||||
}
|
||||
|
||||
Python (FastAPI shown; any framework works) — expose /api/build-version:
|
||||
|
||||
import json, pathlib
|
||||
|
||||
@app.get("/api/build-version")
|
||||
def build_version():
|
||||
bv = json.loads(pathlib.Path("build-version.json").read_text())
|
||||
return {"build_version": f"v{bv['productVersion']}.{bv['buildNumber']}"}
|
||||
|
||||
Python projects can go further with a scripts/build_version_tool.py supporting get / set / bump subcommands — if present, zdeploy bumps the version inside the running container, records it, and zstart bumps on every dev start.
|
||||
|
||||
Alternative: server-side health check (verify block)
|
||||
----------------------------------------------------
|
||||
|
||||
Not every stack is published through the edge proxy — internal APIs, apps whose host port the firewall blocks, services waiting on a DNS record. For those, the old fallback (GET http://<server-ip>/) was worse than nothing: the edge proxy's default vhost answers with a 200 and the deploy "passes" even if your app never started.
|
||||
|
||||
Give the project a verify block instead, and zdeploy checks the app from the server itself over SSH:
|
||||
|
||||
"verify": { "port": 8005, "path": "/health", "expect": "\"status\":\"ok\"" }
|
||||
|
||||
port is the host port the app publishes on the server; path defaults to /; expect is an optional substring the response must contain (an app version string makes this equivalent to build-number verification). Python-kind projects use verify automatically when there's no build_version_tool.py — and projects with neither a domain nor a verify block are now honestly reported as NOT verified instead of green-lighting the proxy's default page.
|
||||
|
||||
Two more keys make the check unambiguous and patient:
|
||||
|
||||
"verify": {
|
||||
"port": 8005, "path": "/health",
|
||||
"viaProxy": "my_edge_proxy", "upstream": "myapp_container:8000",
|
||||
"timeoutSeconds": 120
|
||||
}
|
||||
|
||||
- viaProxy + upstream — read the version over the docker network, by exec-ing a curl inside the named container (usually the edge proxy, since it is on every stack's network). This is the most trustworthy channel there is: it cannot answer from the wrong product, and it works for apps that publish no host port at all.
|
||||
- timeoutSeconds — how long verification may wait. An app that runs database migrations in its entrypoint exceeds a 30-second window on every deploy that ships one, and a warning that fires on routine success teaches you to ignore the one that matters.
|
||||
|
||||
Verification walks its channels in trust order — docker-network viaProxy, then localhost:<port>, then the edge with the project's own Host header — and re-picks the channel on every retry. That last part is the point: the check runs while the app is restarting, which is exactly when the good channels are briefly down, and locking the choice in up front is how a whole verification window gets spent asking the wrong thing. A project with no domain is never asked for via the edge: without a Host header the proxy can only answer from its default vhost — a different product — and no answer beats somebody else's answer.
|
||||
|
||||
zec2 and zec2online use these same endpoints to show what's live and flag local/server version drift.
|
||||
|
||||
---
|
||||
|
||||
Adding a new project
|
||||
--------------------
|
||||
|
||||
1. Add a key under projects in zconfig.json — copy the sample of the matching kind and rename it.
|
||||
2. That's it: zstart, zkill, zrestart, zbackup, zdeploy, zec2, zec2online, zrepair, zstop all accept the new key immediately.
|
||||
3. A project whose deploy doesn't fit the python/vite/nextjs/edge/docker patterns needs its own Invoke-<Kind>Deploy function in zdeploy.ps1 — copy an existing handler; they're all variations on zip → upload → compose up → verify.
|
||||
|
||||
Downloading without cloning
|
||||
---------------------------
|
||||
|
||||
Each release is packaged as a zip in releases/ (releases/) — grab the latest zscripts-v*.zip, check it, unzip, done:
|
||||
|
||||
sha256sum -c zscripts-v1.0.0.0.0.zip.sha256 # verify the download
|
||||
unzip zscripts-v1.0.0.0.0.zip -d zscripts # extract
|
||||
cd zscripts && sha256sum -c CHECKSUMS.txt # verify the contents
|
||||
|
||||
The zip contains every command, CHECKSUMS.txt, zconfig.example.json, and the docs. Versions follow v{major}.{rc}.{beta}.{alpha}.{build}; every script header carries the release version it shipped in, so even a single copied file can be traced to its release.
|
||||
|
||||
Verifying what you downloaded
|
||||
-----------------------------
|
||||
|
||||
CHECKSUMS.txt holds a SHA-256 for every .ps1 and .cmd in the repo. Check them before running anything:
|
||||
|
||||
zchecksums
|
||||
|
||||
Or with the standard tool on Linux/macOS/WSL — the manifest is sha256sum format:
|
||||
|
||||
sha256sum -c CHECKSUMS.txt
|
||||
|
||||
The hashes are identical on every platform: .gitattributes pins .ps1/.cmd to CRLF everywhere, so a file is byte-for-byte the same whether you cloned on Windows or Linux.
|
||||
|
||||
zchecksums flags three things — a file whose contents changed, a listed file that's gone, and a script on disk that isn't in the manifest (so something added quietly still gets noticed). It exits non-zero on any of them.
|
||||
|
||||
If you edit a script yourself, regenerate and commit the manifest with it:
|
||||
|
||||
zchecksums -Update
|
||||
|
||||
What this does and doesn't prove. CHECKSUMS.txt lives in the same repo as the scripts, so anyone who could alter a script could alter the manifest too. It's an integrity check, not a signature: it reliably catches a truncated clone, a local edit you forgot about, or a file added outside a commit. It does not prove the code came from this project — for that you'd need a signature or a hash published outside this repo.
|
||||
|
||||
Tests
|
||||
-----
|
||||
|
||||
The toolkit has its own Pester (https://pester.dev) suite covering the pure logic — the exclude lists, config lookups, and version-label formatting that the deploy and backup paths depend on:
|
||||
|
||||
Invoke-Pester .\tests
|
||||
|
||||
Needs Pester 5+ (Install-Module Pester -Scope CurrentUser); Windows ships 3.x, which won't run these. The suite injects a fixture config through ZCONFIG, so it never reads your real zconfig.json and runs fine on a machine that has never been configured.
|
||||
|
||||
The high-value case is the deploy-vs-backup split: deploys must exclude .env files and uploads/, backups must keep them. Get that backwards in either direction and you either ship secrets to production or quietly write backups that can't restore — neither fails loudly at runtime.
|
||||
|
||||
Troubleshooting
|
||||
---------------
|
||||
|
||||
- "zconfig.json not found" — you haven't copied zconfig.example.json yet. Every script tells you this and exits.
|
||||
- "Unknown project key" — the argument doesn't match a key in zconfig.json; the error lists the valid keys.
|
||||
- "PEM key not found" — fix ec2.pemKey in zconfig.json.
|
||||
- Deploy aborts with "less than 1.5 GB free" — the server's disk is full even after auto-pruning. Grow the volume, or SSH in and run sudo docker system prune -af.
|
||||
- Deploy ends with a version WARNING — the new build isn't what's being served: check the Docker build output, and see Enabling deploy verification (#enabling-deploy-verification) if you haven't set it up.
|
||||
- Port already in use when starting — zkill <project> first, or just use zrestart.
|
||||
|
||||
License
|
||||
-------
|
||||
|
||||
MIT (LICENSE)
|
||||
174
ZHelpers.ps1
174
ZHelpers.ps1
@ -15,8 +15,8 @@ $script:ArchiveExtensions = @(
|
||||
# exempt from ArchiveExtensions. A project that vendors a dependency as
|
||||
# vendor/*.tgz (common when a bundler cannot resolve `file:` links outside the
|
||||
# project root) needs that tarball in the deploy zip - dropping it makes a
|
||||
# Dockerfile's `COPY vendor ./vendor` fail at image build, which is a confusing
|
||||
# way to discover the archive filter ate a required build input.
|
||||
# Dockerfile's `COPY vendor ./vendor` fail at image build, which is a
|
||||
# confusing way to discover the archive filter ate a required build input.
|
||||
$script:ArchiveKeepDirNames = @('vendor')
|
||||
$script:ScriptExtensions = @('.ps1', '.cmd', '.bat')
|
||||
$script:JunkExtensions = @(
|
||||
@ -42,8 +42,9 @@ $script:JunkDirNames = @(
|
||||
$script:ZConfigCache = $null
|
||||
|
||||
# Where zconfig.json lives. Defaults to next to the scripts; override with the
|
||||
# ZCONFIG environment variable. Useful for pointing a run at an alternate
|
||||
# config, and it is the seam the test suite uses to inject a fixture.
|
||||
# ZCONFIG environment variable, matching the bash port (zhelpers.sh does the
|
||||
# same). Useful for pointing a run at an alternate config, and it is the seam
|
||||
# the test suite uses to inject a fixture.
|
||||
function Get-ZConfigPath {
|
||||
if ($env:ZCONFIG) { return $env:ZCONFIG }
|
||||
return (Join-Path $PSScriptRoot "zconfig.json")
|
||||
@ -122,18 +123,19 @@ function Get-Ec2Home {
|
||||
return "/home/$((Get-ZConfig).ec2.user)"
|
||||
}
|
||||
|
||||
# Options every deploy-path ssh/scp carries. Splat with @sshOpts.
|
||||
# Options every deploy-path ssh/scp carries. Splat with @sshOpts, matching the
|
||||
# idiom in zsetup_mail.ps1 and zec2_rotatekeys.ps1.
|
||||
#
|
||||
# BatchMode=yes is the one that matters. Without it ssh PROMPTS - for a
|
||||
# passphrase, a password, a sudo password - and waits forever. The deploy pipes
|
||||
# stderr into the pipeline (2>&1 | ForEach-Object) so the prompt is swallowed
|
||||
# on its way to the screen: the run simply stops under whatever step label was
|
||||
# printed last, with nothing to explain it and no obvious reason why that
|
||||
# particular step would be slow. One deploy appeared to hang on "ensure shared
|
||||
# web network", a step whose entire body is `docker network create web
|
||||
# particular step would be slow. `zdeploy umami` appeared to hang on "ensure
|
||||
# shared web network", a step whose entire body is `docker network create web
|
||||
# 2>/dev/null || true` against a network that already existed.
|
||||
#
|
||||
# There is no prompt here you would ever want to answer - a deploy key is
|
||||
# There is no prompt here we would ever want to answer - the deploy key is
|
||||
# unencrypted and sudo on the box is passwordless - so failing immediately is
|
||||
# strictly better than waiting on input that is never coming.
|
||||
#
|
||||
@ -147,16 +149,16 @@ function Get-Ec2Home {
|
||||
# reads its stdin and forwards it to the remote command, and under PowerShell it
|
||||
# inherits the console handle - so it can block forever waiting on input nobody
|
||||
# is going to type. The timeouts above cannot help: they bound a connection that
|
||||
# is dying, and this one was never established. One deploy stopped under
|
||||
# "ensure unzip installed", a step whose body short-circuits when unzip is
|
||||
# already present; the server showed no ssh session at all (`who` empty, no
|
||||
# docker build running), which is what a client-side stdin block looks like
|
||||
# from the other end. The zip had uploaded and prod stayed a release behind.
|
||||
# is dying, and this one was never established. A deploy on 2026-08-19 stopped
|
||||
# under "ensure unzip installed", a step whose body short-circuits on an already
|
||||
# installed unzip; the server showed no ssh session at all (`who` empty, no
|
||||
# docker build running), which is what a client-side stdin block looks like from
|
||||
# the other end. The zip had uploaded and prod stayed a PR behind.
|
||||
#
|
||||
# Safe here because nothing that pipes stdin INTO ssh uses these options:
|
||||
# zdeploy passes only command strings. Any script that DOES pipe into ssh must
|
||||
# build its own option array - adding -n to those would break them, so do not
|
||||
# hoist this beyond the deploy path.
|
||||
# zdeploy passes only command strings, and the scripts that do pipe
|
||||
# (one-off registration and key-rotation scripts) build their own
|
||||
# option arrays. Adding -n to those would break them - do not hoist it.
|
||||
function Get-Ec2SshOpts {
|
||||
return @(
|
||||
'-n',
|
||||
@ -170,9 +172,9 @@ function Get-Ec2SshOpts {
|
||||
|
||||
# The same options for scp, which does NOT accept -n: OpenSSH's scp exits 1 with
|
||||
# "unknown option -- n" and prints its usage block. That failure is easy to
|
||||
# misread, because the caller's own error text is what the operator sees while
|
||||
# the usage text scrolls past above it - one deploy reported "Likely server disk
|
||||
# space" on a box with plenty of room.
|
||||
# misread, because the caller's own error text is what the operator sees and the
|
||||
# usage text scrolls past above it - a deploy on 2026-08-19 reported "Likely
|
||||
# server disk space" while the box sat at 79% with 8.1 GB free.
|
||||
#
|
||||
# Derived from Get-Ec2SshOpts rather than duplicated, so the timeouts can never
|
||||
# drift apart between the two transports.
|
||||
@ -217,11 +219,11 @@ function Invoke-Ec2Step {
|
||||
#
|
||||
# Deploys ship the default branch, so this SWITCHES to it rather than pulling
|
||||
# whatever branch happens to be checked out. The old behaviour pulled the
|
||||
# current branch, which breaks as soon as the remote deletes branches on merge:
|
||||
# a checkout still sitting on its just-merged PR branch pulls a ref the merge
|
||||
# deleted, and the deploy dies on "no such ref was fetched". Worse, when the ref
|
||||
# DID still exist, pulling the feature branch meant a deploy could ship a
|
||||
# branch rather than the default.
|
||||
# current branch, which broke the day delete_branch_on_merge went on
|
||||
# fleet-wide (2026-08-07): a repo still sitting on its just-merged PR branch
|
||||
# pulls a ref the merge deleted, and the deploy dies on "no such ref was
|
||||
# fetched". Worse, when the ref DID still exist, pulling the feature branch
|
||||
# meant a deploy could ship a branch, not main.
|
||||
#
|
||||
# The switch refuses to run over local changes: a dirty tree aborts the deploy
|
||||
# with the file list rather than risk tangling uncommitted work. The stale
|
||||
@ -328,32 +330,36 @@ function Get-ServerSideVersionCommand {
|
||||
$null when the project has not configured one.
|
||||
|
||||
.DESCRIPTION
|
||||
Reading a live build number through the public proxy only works while
|
||||
that endpoint IS public - and a build stamp is something many sites
|
||||
deliberately do not serve to the world. Blocking it at the proxy
|
||||
without moving the readers first leaves every tool quietly reporting
|
||||
"unknown", which looks identical to "could not reach it".
|
||||
Four tools read a live build number, and all four did it by asking
|
||||
the public edge: zdeploy's post-deploy check, zec2, zec2online, and
|
||||
bash/zhelpers.sh. That works only for as long as the endpoint is
|
||||
public, and it should not be: www's /build-version.json has been
|
||||
blocked at the edge since the 2026-05-29 security pass, and evo.ehs
|
||||
answering /api/build-version to anyone is the inconsistency this
|
||||
closes.
|
||||
|
||||
Going through the proxy is also how a check reads the WRONG service:
|
||||
the proxy answers from whichever vhost matches the Host header, so a
|
||||
container with no public route gets another site's version back.
|
||||
Going through the edge is also how a check reads the WRONG product.
|
||||
The proxy answers from whichever vhost matches the Host header, so a
|
||||
service with no public route gets somebody else's version back --
|
||||
evo-ai's deploy check compared evo.ehs's build against its own and
|
||||
reported a failure on a deploy that had worked (evo.scripts#101).
|
||||
|
||||
A service reached only on a shared docker network cannot be curled
|
||||
from the host when it publishes no port. It IS reachable by name from
|
||||
another container on that network, which also exercises the real HTTP
|
||||
path - so this proves the app is serving, not merely that its
|
||||
A project reached only on the shared docker network cannot be curled
|
||||
from the host: evoehs_app publishes no port. It IS reachable by name
|
||||
from another container on that network, which also exercises the real
|
||||
HTTP path -- so this proves the app is serving, not merely that its
|
||||
database knows a version.
|
||||
|
||||
Config, on the project's `verify` block:
|
||||
|
||||
"verify": {
|
||||
"path": "/api/build-version",
|
||||
"viaProxy": "edge_proxy_container",
|
||||
"upstream": "app_container:80"
|
||||
"viaProxy": "evo_edge_proxy",
|
||||
"upstream": "evoehs_app:80"
|
||||
}
|
||||
|
||||
Returns $null when either key is missing, so every project without
|
||||
this config keeps the behaviour it has today.
|
||||
this config keeps exactly the behaviour it has today.
|
||||
#>
|
||||
param($Proj)
|
||||
|
||||
@ -370,9 +376,10 @@ function Get-LabelFromVersionJson {
|
||||
The build label out of a version endpoint's JSON text, or $null.
|
||||
|
||||
.DESCRIPTION
|
||||
Apps disagree about the field name - some answer `build_version`,
|
||||
others `version`. Both mean "the build that is live", so both are
|
||||
accepted rather than making an app rename its own field.
|
||||
Two field names in the fleet: evo.ehs answers `build_version` on
|
||||
/api/build-version, evo-ai answers `version` on /health. Both mean
|
||||
"the build that is live", so both are accepted rather than making an
|
||||
app rename its own field.
|
||||
#>
|
||||
param([string]$Text)
|
||||
|
||||
@ -832,3 +839,86 @@ function Stop-ZTracking {
|
||||
Remove-Item -LiteralPath $tp -Force -ErrorAction SilentlyContinue
|
||||
Write-ZTrailer -FinalNote $FinalNote
|
||||
}
|
||||
|
||||
# ── Deploy-verification planning (pure; unit-tested in tests/) ──────────────
|
||||
|
||||
function Get-VerifyTimeout {
|
||||
<#
|
||||
.SYNOPSIS
|
||||
Seconds the live-build verification may wait, per project.
|
||||
|
||||
.DESCRIPTION
|
||||
verify.timeoutSeconds in zconfig.json lets a project that is slow to
|
||||
BOOT say so, instead of every deploy of it warning on a success. An
|
||||
app that runs database migrations in its entrypoint exceeds a 30s
|
||||
window on every deploy that ships one - and a warning that fires on
|
||||
routine success trains people to ignore the one that matters
|
||||
(evo.scripts#101).
|
||||
#>
|
||||
param($Proj, [int]$DefaultSec)
|
||||
if ($Proj.verify -and $Proj.verify.timeoutSeconds) {
|
||||
return [int]$Proj.verify.timeoutSeconds
|
||||
}
|
||||
return $DefaultSec
|
||||
}
|
||||
|
||||
function Get-VerifyAttempts {
|
||||
<#
|
||||
.SYNOPSIS
|
||||
The ordered ways to read this project's live build, most-trustworthy
|
||||
first. Pure: config in, plan out - so the ordering rules are testable
|
||||
without ssh.
|
||||
|
||||
.DESCRIPTION
|
||||
Three channels exist, and their order is the whole point (#101):
|
||||
|
||||
exec - docker-network read via verify.viaProxy/upstream. Cannot
|
||||
answer from the wrong product, works for apps with no
|
||||
published port.
|
||||
port - localhost:<verify.port> on the server. Same-box, still
|
||||
unambiguous; used only if the body carries a version.
|
||||
edge - http://<ip> with a Host header. The proxy answers from
|
||||
whichever vhost MATCHES that header, so without one this
|
||||
channel can only reach the default vhost - which is a
|
||||
different product (that is how evo-ai's check once read
|
||||
evo.ehs's build number). It is therefore included ONLY
|
||||
when the project has a host to route by, and never
|
||||
otherwise: no answer at all beats somebody else's answer.
|
||||
|
||||
The caller must walk this list EVERY retry, not once up front: the
|
||||
verification runs while the app is restarting, which is exactly when
|
||||
the good channels are briefly down. Deciding the channel before the
|
||||
wait loop is how the whole window got spent on the worst one.
|
||||
#>
|
||||
param($Proj, [string]$ExecCmd)
|
||||
$attempts = @()
|
||||
if ($ExecCmd) {
|
||||
$attempts += [pscustomobject]@{
|
||||
Kind = 'exec'
|
||||
Label = "docker network: $($Proj.verify.upstream) (via $($Proj.verify.viaProxy))"
|
||||
}
|
||||
}
|
||||
if ($Proj.verify -and $Proj.verify.port) {
|
||||
$vPath = "/api/build-version"
|
||||
if ($Proj.verify.path) { $vPath = [string]$Proj.verify.path }
|
||||
$attempts += [pscustomobject]@{
|
||||
Kind = 'port'
|
||||
Port = [int]$Proj.verify.port
|
||||
Path = $vPath
|
||||
Label = "server localhost:$($Proj.verify.port)$vPath"
|
||||
}
|
||||
}
|
||||
$verifyHost = $null
|
||||
if ($Proj.deploy -and $Proj.deploy.verifyHost) { $verifyHost = [string]$Proj.deploy.verifyHost }
|
||||
elseif ($Proj.domain) { $verifyHost = [string]$Proj.domain }
|
||||
if ($verifyHost) {
|
||||
$attempts += [pscustomobject]@{
|
||||
Kind = 'edge'
|
||||
HostHeader = $verifyHost
|
||||
Label = "edge with Host: $verifyHost"
|
||||
}
|
||||
}
|
||||
# The comma stops PS 5.1 unrolling a one-element array into a bare
|
||||
# object - the same pipeline trap that deadlocked ztests day 2.
|
||||
return ,$attempts
|
||||
}
|
||||
|
||||
76
scripts/readme_txt.py
Normal file
76
scripts/readme_txt.py
Normal file
@ -0,0 +1,76 @@
|
||||
# Evomedia.net Token Savers — https://github.com/evomedia-net/evo.zscripts
|
||||
# Created by Kelly Michels · dev@evomedia.net
|
||||
# Licensed under the MIT License. See LICENSE.
|
||||
|
||||
"""Render README.md to README.txt with the markdown markup removed.
|
||||
|
||||
README.txt exists for terminals, pagers and anywhere markdown doesn't
|
||||
render. It is generated - never edit it by hand:
|
||||
|
||||
python scripts/readme_txt.py # rewrite README.txt
|
||||
python scripts/readme_txt.py --check # exit 1 if it is out of sync
|
||||
|
||||
The test suite runs --check, so a README.md edit that forgets to
|
||||
regenerate fails CI rather than shipping a stale mirror.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import re
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
ROOT = Path(__file__).resolve().parent.parent
|
||||
|
||||
|
||||
def _inline(text: str) -> str:
|
||||
text = re.sub(r"!\[([^\]]*)\]\([^)]*\)", r"\1", text) # images -> alt text
|
||||
text = re.sub(r"\[([^\]]+)\]\(([^)]+)\)", r"\1 (\2)", text) # links -> text (url)
|
||||
text = re.sub(r"\*\*([^*]+)\*\*", r"\1", text) # bold
|
||||
text = re.sub(r"(?<!\*)\*([^*\n]+)\*(?!\*)", r"\1", text) # italic
|
||||
text = re.sub(r"`([^`]+)`", r"\1", text) # inline code
|
||||
return text
|
||||
|
||||
|
||||
def render(md: str) -> str:
|
||||
out: list[str] = []
|
||||
in_fence = False
|
||||
for line in md.splitlines():
|
||||
if line.lstrip().startswith("```"):
|
||||
# Drop the fence markers; the code itself stays, indented so it
|
||||
# still reads as a block without the backticks.
|
||||
in_fence = not in_fence
|
||||
continue
|
||||
if in_fence:
|
||||
out.append((" " + line) if line else "")
|
||||
continue
|
||||
heading = re.match(r"^(#{1,6})\s+(.*)$", line)
|
||||
if heading:
|
||||
text = _inline(heading.group(2))
|
||||
out.append(text)
|
||||
out.append(("=" if len(heading.group(1)) == 1 else "-") * len(text))
|
||||
continue
|
||||
out.append(_inline(line))
|
||||
text = "\n".join(out)
|
||||
text = re.sub(r"\n{3,}", "\n\n", text)
|
||||
return text.strip() + "\n"
|
||||
|
||||
|
||||
def main() -> int:
|
||||
source = (ROOT / "README.md").read_text(encoding="utf-8")
|
||||
rendered = render(source)
|
||||
target = ROOT / "README.txt"
|
||||
if "--check" in sys.argv:
|
||||
current = target.read_text(encoding="utf-8") if target.exists() else ""
|
||||
if current != rendered:
|
||||
print("README.txt is out of sync - run: python scripts/readme_txt.py")
|
||||
return 1
|
||||
print("README.txt is in sync")
|
||||
return 0
|
||||
target.write_text(rendered, encoding="utf-8", newline="\n")
|
||||
print(f"Wrote {target} ({len(rendered.splitlines())} lines)")
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
96
tests/VerifyPlan.Tests.ps1
Normal file
96
tests/VerifyPlan.Tests.ps1
Normal file
@ -0,0 +1,96 @@
|
||||
# Deploy-verification planning (#101).
|
||||
#
|
||||
# Invoke-Pester .\tests
|
||||
#
|
||||
# The wrong-vhost failure was never about parsing a response - it was about
|
||||
# WHICH channel got asked. So the channel-selection rules live in a pure
|
||||
# function (Get-VerifyAttempts) and are pinned here, where they can be tested
|
||||
# without an EC2 box: a project with no domain must never produce an edge
|
||||
# attempt, because an edge request with no Host header can only reach the
|
||||
# default vhost - which is a different product. That exact gap read evo.ehs's
|
||||
# build number during evo-ai deploys twice on 2026-08-31 alone.
|
||||
#
|
||||
# ZHelpers.ps1 is dot-sourced rather than zdeploy.ps1: zdeploy executes its
|
||||
# main flow on load, helpers only define functions.
|
||||
|
||||
BeforeAll {
|
||||
. (Join-Path (Split-Path -Parent $PSScriptRoot) "ZHelpers.ps1")
|
||||
|
||||
function New-Proj {
|
||||
param($Verify = $null, $Domain = $null, $Deploy = $null)
|
||||
$p = [pscustomobject]@{}
|
||||
if ($null -ne $Verify) { $p | Add-Member verify ([pscustomobject]$Verify) }
|
||||
if ($null -ne $Domain) { $p | Add-Member domain $Domain }
|
||||
if ($null -ne $Deploy) { $p | Add-Member deploy ([pscustomobject]$Deploy) }
|
||||
return $p
|
||||
}
|
||||
}
|
||||
|
||||
Describe "Get-VerifyAttempts" {
|
||||
|
||||
It "puts the docker-network read first when configured" {
|
||||
$proj = New-Proj -Verify @{ viaProxy = "evo_edge_proxy"; upstream = "app:80"; port = 8005 } -Domain "x.example"
|
||||
$attempts = Get-VerifyAttempts -Proj $proj -ExecCmd "docker exec ..."
|
||||
$attempts[0].Kind | Should -Be 'exec'
|
||||
($attempts | ForEach-Object Kind) | Should -Be @('exec', 'port', 'edge')
|
||||
}
|
||||
|
||||
It "never asks the edge for a project with no domain (the #101 trap)" {
|
||||
# The shape that hit #101: viaProxy + port, no domain. The old code
|
||||
# fell back to the bare IP here and read another product's counter.
|
||||
$proj = New-Proj -Verify @{ viaProxy = "evo_edge_proxy"; upstream = "deploy-app-1:8000"; port = 8005; path = "/health" }
|
||||
$attempts = Get-VerifyAttempts -Proj $proj -ExecCmd "docker exec ..."
|
||||
($attempts | ForEach-Object Kind) | Should -Not -Contain 'edge'
|
||||
}
|
||||
|
||||
It "returns an empty plan when nothing trustworthy exists" {
|
||||
# No verify config, no domain: the caller must SKIP, not guess.
|
||||
$attempts = Get-VerifyAttempts -Proj (New-Proj) -ExecCmd ""
|
||||
$attempts.Count | Should -Be 0
|
||||
}
|
||||
|
||||
It "keeps the edge for a project with a domain, with its Host header" {
|
||||
$proj = New-Proj -Domain "jwks.example"
|
||||
$attempts = Get-VerifyAttempts -Proj $proj -ExecCmd ""
|
||||
$attempts.Count | Should -Be 1
|
||||
$attempts[0].Kind | Should -Be 'edge'
|
||||
$attempts[0].HostHeader | Should -Be "jwks.example"
|
||||
}
|
||||
|
||||
It "prefers deploy.verifyHost over domain for the edge Host header" {
|
||||
$proj = New-Proj -Domain "old.example" -Deploy @{ verifyHost = "new.example" }
|
||||
$attempts = Get-VerifyAttempts -Proj $proj -ExecCmd ""
|
||||
$attempts[0].HostHeader | Should -Be "new.example"
|
||||
}
|
||||
|
||||
It "carries the verify path into the port attempt, defaulting sensibly" {
|
||||
$proj = New-Proj -Verify @{ port = 8005; path = "/health" }
|
||||
(Get-VerifyAttempts -Proj $proj -ExecCmd "")[0].Path | Should -Be "/health"
|
||||
$proj2 = New-Proj -Verify @{ port = 9000 }
|
||||
(Get-VerifyAttempts -Proj $proj2 -ExecCmd "")[0].Path | Should -Be "/api/build-version"
|
||||
}
|
||||
|
||||
It "survives the PS 5.1 one-element unroll" {
|
||||
# A single attempt must still come back as something with .Count and
|
||||
# index access - the pipeline trap that deadlocked ztests day 2.
|
||||
$proj = New-Proj -Verify @{ port = 8005 }
|
||||
$attempts = Get-VerifyAttempts -Proj $proj -ExecCmd ""
|
||||
$attempts.Count | Should -Be 1
|
||||
$attempts[0].Kind | Should -Be 'port'
|
||||
}
|
||||
}
|
||||
|
||||
Describe "Get-VerifyTimeout" {
|
||||
|
||||
It "uses the caller's default when the project says nothing" {
|
||||
Get-VerifyTimeout -Proj (New-Proj) -DefaultSec 30 | Should -Be 30
|
||||
}
|
||||
|
||||
It "lets a slow-booting project widen its own window" {
|
||||
# An app that runs database migrations in its entrypoint exceeds
|
||||
# 30s on every deploy that ships one, and a warning that fires on
|
||||
# routine success trains people to ignore the real one.
|
||||
$proj = New-Proj -Verify @{ viaProxy = "p"; upstream = "u"; timeoutSeconds = 120 }
|
||||
Get-VerifyTimeout -Proj $proj -DefaultSec 30 | Should -Be 120
|
||||
}
|
||||
}
|
||||
259
zdeploy.ps1
259
zdeploy.ps1
@ -18,12 +18,11 @@
|
||||
# zdeploy all -Note "weekly release"
|
||||
# zdeploy ztokens evo # refresh token-stats.json, then ship the site with it
|
||||
#
|
||||
# "ztokens" is an OPTIONAL pseudo-project, not a zconfig entry: it runs
|
||||
# `ztokens -Publish` from a sibling ztokens checkout, if you have one, to
|
||||
# refresh a token-stats.json a site can chart. With no such checkout the step
|
||||
# prints a skip and the rest of the run is unaffected. `all` runs it first
|
||||
# automatically; called standalone, list it before a site project (as above) so
|
||||
# that project's deploy zip picks up the freshly written file.
|
||||
# "ztokens" is a pseudo-project, not a zconfig entry: it runs `ztokens -Publish`
|
||||
# from the sibling ztokens repo, refreshing the token-stats.json the public
|
||||
# zscripts page charts. `all` runs it first automatically; called standalone,
|
||||
# list it before a site project (as above) so that project's deploy zip picks
|
||||
# up the freshly written file.
|
||||
#
|
||||
# Flow (python/vite/nextjs): zip source -> free server disk space -> scp up ->
|
||||
# unzip into remote.path (preserving server-side .env* files and anything in
|
||||
@ -72,7 +71,7 @@ if ($Projects.Count -eq 0) {
|
||||
Write-Host "Usage: zdeploy <project> [<project> ...] | all | ztokens [-Note `"message`"]" -ForegroundColor Yellow
|
||||
Write-Host " Projects in zconfig.json: $keys" -ForegroundColor Gray
|
||||
Write-Host " 'all' deploys everything (edge kinds first) and stops at the first failure." -ForegroundColor Gray
|
||||
Write-Host " 'ztokens' refreshes live-usage stats, if a sibling ztokens checkout exists." -ForegroundColor Gray
|
||||
Write-Host " 'ztokens' refreshes the live-usage stats published to the zscripts page." -ForegroundColor Gray
|
||||
Stop-ZTracking; exit 1
|
||||
}
|
||||
|
||||
@ -132,10 +131,12 @@ function Invoke-Ec2PreflightCleanup {
|
||||
"echo available_mb=`$avail_mb",
|
||||
"if [ `"`$avail_mb`" -lt 1500 ]; then echo 'ERROR: less than 1.5 GB free on /. Grow the root volume or run: sudo docker system prune -af' >&2; exit 11; fi"
|
||||
) -join '; '
|
||||
ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET $preflightCmd
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
throw "Server pre-flight cleanup failed (exit $LASTEXITCODE). Root volume too full (need ~1.5 GB free, ideally 3+)."
|
||||
}
|
||||
# Also through the wrapper (#119): the out-of-space branch above writes its
|
||||
# ERROR to stderr and exits 11, so a bare ssh would surface a
|
||||
# NativeCommandError instead of the actionable message below - exactly when
|
||||
# the operator most needs to be told what to do. -FailHint keeps it.
|
||||
Invoke-Ec2Step "server pre-flight cleanup" $preflightCmd `
|
||||
-FailHint "Root volume too full (need ~1.5 GB free, ideally 3+). Grow it or run: sudo docker system prune -af"
|
||||
}
|
||||
|
||||
# Post-deploy cleanup: prune build cache and dangling images created during this deploy.
|
||||
@ -170,30 +171,96 @@ function Invoke-RemoteUnzip {
|
||||
Invoke-Ec2Step "unzip $ZipName" $bash
|
||||
}
|
||||
|
||||
# ── Operator-file preservation (issue #2) ────────────────────────────────────
|
||||
# ── Operator-file preservation (issue #2, hardened in #107) ──────────────────
|
||||
# Deploys replace the project directory wholesale, which used to destroy every
|
||||
# operator-managed file except ./.env. These helpers preserve all .env* files
|
||||
# at the project root PLUS any paths listed in deploy.preserve (files or
|
||||
# directories), by tarring them to the home dir before the wipe and extracting
|
||||
# them back after the unzip. Server-side copies win over anything shipped in
|
||||
# the zip — the same semantics ./.env always had.
|
||||
#
|
||||
# WHY THE ARCHIVE IS TIMESTAMPED AND NEVER DELETED (#107)
|
||||
# -------------------------------------------------------
|
||||
# This used to write one fixed preserve_<key>.tgz, and preserve's FIRST action
|
||||
# was `rm -f` on it. That is the opposite of safe. The window between
|
||||
# `sudo rm -rf` on the project directory and the restore step is the only time
|
||||
# the tarball is the sole copy of the production secrets - and an interrupted
|
||||
# run (dropped ssh, exit 255) stops exactly there, leaving a perfect backup
|
||||
# behind. The next run then deleted that backup before doing anything else,
|
||||
# tarred a directory that no longer had the files, and reported success.
|
||||
# `2>/dev/null; true` on the tar is what made it silent.
|
||||
#
|
||||
# That destroyed civilcode's production deploy/.env on 2026-08-30. The site
|
||||
# survived only because the running container still held its environment; a
|
||||
# restart would have made the loss permanent.
|
||||
#
|
||||
# So, three independent changes, any one of which would have prevented it:
|
||||
#
|
||||
# 1. Each run writes its own preserve_<key>_<stamp>.tgz and restore no
|
||||
# longer deletes it. Nothing removes an archive that has not been
|
||||
# superseded - retention below prunes old ones instead.
|
||||
# 2. Preserve first extracts any earlier archives with `tar -k`, which fills
|
||||
# in files a previous interrupted run lost WITHOUT overwriting anything
|
||||
# currently on disk. A hand-repaired .env therefore wins over the stale
|
||||
# copy in the archive.
|
||||
# 3. Preserve refuses to continue if it captured nothing while an earlier
|
||||
# archive for the same key did have contents. Capturing zero files is
|
||||
# normal for a project with no operator files (edge, gitea, landing) and
|
||||
# catastrophic for one that has them; the prior archive is what tells the
|
||||
# difference.
|
||||
#
|
||||
# One stamp per zdeploy process, so preserve and restore agree on the filename
|
||||
# without threading it through every call site.
|
||||
$script:PreserveStamp = Get-Date -Format 'yyyyMMdd-HHmmss'
|
||||
$script:PreserveKeep = 5
|
||||
|
||||
function Get-PreserveTarball {
|
||||
param([string]$Key)
|
||||
"$RemoteHome/preserve_${Key}_$($script:PreserveStamp).tgz"
|
||||
}
|
||||
|
||||
function Save-OperatorFiles {
|
||||
param([string]$Key, $Proj, [string]$RemotePath)
|
||||
$paths = @('.env*')
|
||||
if ($Proj.deploy -and $Proj.deploy.preserve) { $paths += @($Proj.deploy.preserve) }
|
||||
$spec = $paths -join ' '
|
||||
$tarball = "$RemoteHome/preserve_${Key}.tgz"
|
||||
# NOTE: no embedded quotes or $( ) here - PowerShell 5.1 strips embedded
|
||||
# double quotes when passing args to ssh.exe, silently corrupting the
|
||||
# remote command. Globs expand remotely; tar archives whatever exists
|
||||
# and its nonzero exit for missing paths is deliberately swallowed.
|
||||
Invoke-Ec2Step "preserve operator files ($spec)" "rm -f $tarball; cd $RemotePath && tar -czf $tarball $spec 2>/dev/null; true"
|
||||
$tarball = Get-PreserveTarball -Key $Key
|
||||
$list = $tarball -replace '\.tgz$', '.list'
|
||||
$glob = "$RemoteHome/preserve_${Key}_*"
|
||||
$drop = $script:PreserveKeep + 1
|
||||
|
||||
# NOTE: no embedded double quotes or $( ) here - PowerShell 5.1 strips
|
||||
# embedded double quotes when passing args to ssh.exe, silently corrupting
|
||||
# the remote command, and $( ) would be evaluated locally. Remote shell
|
||||
# variables are backtick-escaped so PowerShell leaves them alone. Globs
|
||||
# expand remotely; tar's nonzero exit for missing paths is swallowed, but
|
||||
# an empty capture is NOT (see the guard below).
|
||||
$bash =
|
||||
"cd $RemotePath || exit 9; " +
|
||||
"for t in ${glob}.tgz; do [ -e `$t ] && tar -xzkf `$t -C $RemotePath 2>/dev/null; done; true; " +
|
||||
"tar -czf $tarball $spec 2>/dev/null; " +
|
||||
"tar -tzf $tarball > $list 2>/dev/null; " +
|
||||
"if [ ! -s $list ]; then " +
|
||||
"for p in ${glob}.list; do " +
|
||||
"if [ -s `$p ] && [ `$p != $list ]; then " +
|
||||
"echo PRESERVE CAPTURED NOTHING BUT AN EARLIER ARCHIVE HAS FILES; exit 8; " +
|
||||
"fi; " +
|
||||
"done; " +
|
||||
"fi; " +
|
||||
"ls -1t ${glob}.tgz 2>/dev/null | tail -n +$drop | xargs -r rm -f; " +
|
||||
"ls -1t ${glob}.list 2>/dev/null | tail -n +$drop | xargs -r rm -f; " +
|
||||
"exit 0"
|
||||
|
||||
Invoke-Ec2Step "preserve operator files ($spec)" $bash `
|
||||
-FailHint "Refusing to wipe $RemotePath - see $glob.tgz on the server."
|
||||
}
|
||||
|
||||
function Restore-OperatorFiles {
|
||||
param([string]$Key, [string]$RemotePath)
|
||||
$tarball = "$RemoteHome/preserve_${Key}.tgz"
|
||||
Invoke-Ec2Step "restore operator files" "test -f $tarball && tar -xzf $tarball -C $RemotePath; rm -f $tarball; true"
|
||||
$tarball = Get-PreserveTarball -Key $Key
|
||||
# Overwrites, deliberately: server-side operator files beat whatever the
|
||||
# zip shipped. The archive is left in place - see the header.
|
||||
Invoke-Ec2Step "restore operator files" "test -f $tarball && tar -xzf $tarball -C $RemotePath; true"
|
||||
}
|
||||
|
||||
# ── Deploy verification (build-version match, not just HTTP 200 — a 200 can be
|
||||
@ -247,54 +314,87 @@ function Wait-VerifyStaticBuild {
|
||||
function Wait-VerifyApiBuild {
|
||||
param([string]$Key, $Proj, [string]$ExpectedLabel, [int]$TimeoutSec = 60)
|
||||
Write-Host "`n--- [$Key] Live build verification (expect $ExpectedLabel) ---" -ForegroundColor Cyan
|
||||
$headers = @{}
|
||||
# deploy.verifyHost overrides domain for verification only. The Host header
|
||||
# decides which edge vhost answers, and a project's public host can be
|
||||
# deliberately unroutable while the app is perfectly healthy - that is why
|
||||
# the override exists. Reach for it when a domain is being retired ahead of
|
||||
# its replacement: the old host may be returning 410 while the new one has
|
||||
# no DNS yet, so neither answers even though the app is fine.
|
||||
$verifyHost = if ($Proj.deploy -and $Proj.deploy.verifyHost) { $Proj.deploy.verifyHost } else { $Proj.domain }
|
||||
if ($verifyHost) { $headers['Host'] = $verifyHost }
|
||||
# Preferred when the project configures it: read the version from a
|
||||
# container ON the shared docker network rather than through the public
|
||||
# proxy. A service that publishes no port cannot be curled from the host
|
||||
# at all, and the proxy answers from whichever vhost matches the Host
|
||||
# header - so a container with no public route gets another site's
|
||||
# version back. See Get-ServerSideVersionCommand.
|
||||
# deploy.verifyHost (or domain) decides which edge vhost may be asked;
|
||||
# both are consumed inside Get-VerifyAttempts now, where "no host at all"
|
||||
# excludes the edge channel entirely rather than defaulting to whatever
|
||||
# vhost the proxy serves (#101). verifyHost exists for a domain retired
|
||||
# ahead of its replacement - a takedown once had a project's public host
|
||||
# answering 410 while the app was healthy; no project sets it today.
|
||||
# Every retry walks the channels in trust order - docker-network exec,
|
||||
# then localhost port, then (only with a Host to route by) the edge.
|
||||
# The choice used to be made ONCE, before the loop, by probing each
|
||||
# channel - but the probes ran at the exact moment step [5] had
|
||||
# restarted the app, so both good channels were briefly down and the
|
||||
# whole window was spent on the edge. For a project with no domain that
|
||||
# meant the default vhost: one project's check read a DIFFERENT
|
||||
# project's build number, twice in a single day (#101). Re-resolving per
|
||||
# retry means the right channel is used the moment the app is back.
|
||||
#
|
||||
# The port channel still counts only when the body carries a version:
|
||||
# one project's verify path is a JWKS endpoint - real, healthy, and no
|
||||
# version in it - so it falls through to the edge (it has a domain),
|
||||
# same as it always did.
|
||||
$execCmd = Get-ServerSideVersionCommand -Proj $Proj
|
||||
$useExec = $false
|
||||
if ($execCmd) {
|
||||
$probe = (ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET $execCmd | Out-String).Trim()
|
||||
if (Get-LabelFromVersionJson $probe) { $useExec = $true }
|
||||
}
|
||||
if ($useExec) {
|
||||
Write-Host " Asking on server: $($Proj.verify.upstream) (via $($Proj.verify.viaProxy))" -ForegroundColor DarkGray
|
||||
$attempts = Get-VerifyAttempts -Proj $Proj -ExecCmd $execCmd
|
||||
$TimeoutSec = Get-VerifyTimeout -Proj $Proj -DefaultSec $TimeoutSec
|
||||
if ($attempts.Count -eq 0) {
|
||||
# No trustworthy channel exists: no viaProxy, no port, no host to
|
||||
# route an edge request by. Asking the edge anyway can only reach
|
||||
# the DEFAULT vhost - a different product - and a check that can
|
||||
# only ever read someone else's number is worse than no check.
|
||||
Write-Host " SKIPPED: no way to verify this project without reading the wrong vhost - configure verify.viaProxy/port, or a domain (#101)." -ForegroundColor Yellow
|
||||
return $false
|
||||
}
|
||||
Write-Host " Channels, in order: $(($attempts | ForEach-Object { $_.Label }) -join '; ')" -ForegroundColor DarkGray
|
||||
|
||||
$sawVersion = $false
|
||||
$deadline = (Get-Date).AddSeconds($TimeoutSec)
|
||||
while ((Get-Date) -lt $deadline) {
|
||||
foreach ($attempt in $attempts) {
|
||||
$r = $null
|
||||
try {
|
||||
if ($useExec) {
|
||||
switch ($attempt.Kind) {
|
||||
'exec' {
|
||||
$raw = ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET $execCmd
|
||||
$live = Get-LabelFromVersionJson ($raw | Out-String)
|
||||
} else {
|
||||
$r = Invoke-RestMethod -Uri "http://$EC2_IP/api/build-version" -Headers $headers -TimeoutSec 10 -ErrorAction Stop
|
||||
$live = if ($r -and $r.build_version) { [string]$r.build_version } else { $null }
|
||||
$r = ($raw | Out-String).Trim() | ConvertFrom-Json -ErrorAction Stop
|
||||
}
|
||||
if ($live) {
|
||||
if ($live -eq $ExpectedLabel) {
|
||||
Write-Host " PASS - live build $live matches expected." -ForegroundColor Green
|
||||
return $true
|
||||
'port' {
|
||||
$raw = ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET "curl -s -m 8 http://localhost:$($attempt.Port)$($attempt.Path)"
|
||||
$r = ($raw | Out-String).Trim() | ConvertFrom-Json -ErrorAction Stop
|
||||
}
|
||||
'edge' {
|
||||
$edgeHeaders = @{ 'Host' = $attempt.HostHeader }
|
||||
$r = Invoke-RestMethod -Uri "http://$EC2_IP/api/build-version" -Headers $edgeHeaders -TimeoutSec 10 -ErrorAction Stop
|
||||
}
|
||||
Write-Host " Live build is $live, expected $ExpectedLabel - waiting..." -ForegroundColor DarkYellow
|
||||
}
|
||||
} catch {
|
||||
Write-Host " version endpoint not ready yet - waiting..." -ForegroundColor DarkGray
|
||||
continue # channel not ready; the next one gets its turn
|
||||
}
|
||||
if ($null -eq $r) { continue }
|
||||
# Two field names in the fleet: some apps answer build_version
|
||||
# on /api/build-version, others answer version on /health. Both
|
||||
# are "the build that is live", so accept either rather than
|
||||
# making every app rename its own field.
|
||||
$live = if ($r.build_version) { [string]$r.build_version } elseif ($r.version) { [string]$r.version } else { $null }
|
||||
if (-not $live) { continue } # answered, but not about versions (JWKS etc.)
|
||||
$sawVersion = $true
|
||||
if ($live -eq $ExpectedLabel) {
|
||||
Write-Host " PASS - live build $live matches expected ($($attempt.Label))." -ForegroundColor Green
|
||||
return $true
|
||||
}
|
||||
Write-Host " Live build is $live via $($attempt.Label), expected $ExpectedLabel - waiting..." -ForegroundColor DarkYellow
|
||||
break # one wrong-version read this pass is enough; retry after the sleep
|
||||
}
|
||||
Start-Sleep -Seconds 3
|
||||
}
|
||||
Write-Host " WARNING: live build did not match $ExpectedLabel within ${TimeoutSec}s (a stale build may be cached)." -ForegroundColor Yellow
|
||||
# Say which failure this actually was: a version that never matched is a
|
||||
# stale/failed build; channels that never answered is "could not verify",
|
||||
# and pretending otherwise is how a warning gets ignored.
|
||||
if ($sawVersion) {
|
||||
Write-Host " WARNING: live build did not match $ExpectedLabel within ${TimeoutSec}s (upload or Docker build may have failed, or a stale build is cached)." -ForegroundColor Yellow
|
||||
} else {
|
||||
Write-Host " WARNING: could not verify within ${TimeoutSec}s - no channel answered with a version (app may still be starting; raise verify.timeoutSeconds if this project boots slowly)." -ForegroundColor Yellow
|
||||
}
|
||||
return $false
|
||||
}
|
||||
|
||||
@ -435,7 +535,7 @@ function Invoke-PythonDeploy {
|
||||
# in the container, is written to .build_version below, and is
|
||||
# proven by the /api/build-version check — which is the thing that
|
||||
# actually establishes what is deployed.
|
||||
ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET "echo '$BuildVersion' | sudo tee $remotePath/.build_version > /dev/null"
|
||||
Invoke-Ec2Step "record the live build number" "echo '$BuildVersion' | sudo tee $remotePath/.build_version > /dev/null"
|
||||
$changelogTool = Join-Path $root "scripts\build_changelog_tool.py"
|
||||
if (Test-Path -LiteralPath $changelogTool) {
|
||||
if ([string]::IsNullOrWhiteSpace($ChangeNote)) { $ChangeNote = "Build deployed" }
|
||||
@ -443,8 +543,18 @@ function Invoke-PythonDeploy {
|
||||
}
|
||||
|
||||
Write-Host "`n--- [5] Restarting app to pick up new version ---" -ForegroundColor Cyan
|
||||
ssh @SSH_OPTS -i $PEM_KEY $SSH_TARGET "cd $composeDir && sudo COMPOSE_BAKE=false docker compose restart $appSvc"
|
||||
if ($LASTEXITCODE -ne 0) { throw "App restart after build bump failed (exit $LASTEXITCODE)" }
|
||||
# Through Invoke-Ec2Step, not a bare ssh: `docker compose restart`
|
||||
# writes " Container <name> Restarting" to STDERR as ordinary
|
||||
# progress, and under ErrorActionPreference='Stop' PS 5.1 turns any
|
||||
# native stderr line into a terminating NativeCommandError whatever
|
||||
# the exit code. That threw here on a deploy that had fully
|
||||
# succeeded - and it threw BEFORE Wait-VerifyApiBuild, so the step
|
||||
# that proves what is actually deployed never ran (#119). The
|
||||
# wrapper flattens stderr and judges by exit code alone, and throws
|
||||
# on non-zero itself, so the hand-written check is gone with it.
|
||||
Invoke-Ec2Step "restart $appSvc to pick up the new build" `
|
||||
"cd $composeDir && sudo COMPOSE_BAKE=false docker compose restart $appSvc" `
|
||||
-FailHint "App restart after the build bump failed."
|
||||
|
||||
Wait-VerifyApiBuild -Key $Key -Proj $Proj -ExpectedLabel $BuildVersion -TimeoutSec 30 | Out-Null
|
||||
} elseif ($Proj.verify -and $Proj.verify.port) {
|
||||
@ -646,7 +756,7 @@ function Invoke-NextDeploy {
|
||||
# offline for the whole build - minutes for a Next.js app - and left it
|
||||
# offline if the build failed. That is not hypothetical: one deploy
|
||||
# stopped the stack, the build did not finish, and the site served 502
|
||||
# for 19 hours with no container running at all. The
|
||||
# for 19 hours with no container at all. The
|
||||
# old image keeps serving while the new one builds, so a failed build is
|
||||
# now harmless and the outage is the seconds between down and up.
|
||||
#
|
||||
@ -670,7 +780,7 @@ function Invoke-NextDeploy {
|
||||
# here: a compose stack behind the edge proxy usually publishes to
|
||||
# 127.0.0.1 only, so that probe can never answer and the old "is the port
|
||||
# open in the security group?" warning sent you chasing a firewall rule
|
||||
# for an app that was already up.
|
||||
# for an app that was already up. See issue #28.
|
||||
if ($Proj.verify -and $Proj.verify.port) {
|
||||
Test-DeployHealth -Key $Key -Proj $Proj -TimeoutSec 60 | Out-Null
|
||||
} elseif ($Proj.domain) {
|
||||
@ -752,10 +862,11 @@ function Invoke-EdgeDeploy {
|
||||
}
|
||||
|
||||
# Content subdirectories the proxy serves (fonts/, vendor/, ...) ship too —
|
||||
# only server-side state stays put. Skipping them is how self-hosted assets
|
||||
# silently never reach prod: docker creates empty mount-point dirs and nginx
|
||||
# serves 404s from them, so fonts fall back and vendored JS never loads.
|
||||
$skipDirs = @('nginx-logs', '.git')
|
||||
# only server-side state stays put. Skipping them is how the self-hosted
|
||||
# Chart.js and fonts silently never reached prod (charts rendered blank).
|
||||
# .pytest_cache is a local test artifact, already gitignored; it has no
|
||||
# business on the proxy box and only adds noise to the upload log.
|
||||
$skipDirs = @('nginx-logs', '.git', '.pytest_cache')
|
||||
$dirs = @(Get-ChildItem -LiteralPath $root -Directory | Where-Object { $skipDirs -notcontains $_.Name })
|
||||
foreach ($d in $dirs) {
|
||||
Write-Host " >> uploading $($d.Name)/ (recursive)" -ForegroundColor DarkCyan
|
||||
@ -781,9 +892,9 @@ function Invoke-EdgeDeploy {
|
||||
function Invoke-StaticDeploy {
|
||||
param([string]$Key, $Proj)
|
||||
|
||||
# Plain static sites - no build, no container of their own. A shared web
|
||||
# container serves them straight off disk, so shipping the files IS the
|
||||
# deploy: there is nothing to restart afterwards.
|
||||
# Plain static sites - no build, no container of their own. The landing
|
||||
# container serves them straight off disk out of /srv/$host, so shipping
|
||||
# the files IS the deploy: there is nothing to restart afterwards.
|
||||
$root = Join-Path $Proj.localRoot $Proj.siteDir
|
||||
$remotePath = $Proj.remote.path
|
||||
|
||||
@ -801,11 +912,8 @@ function Invoke-StaticDeploy {
|
||||
}
|
||||
|
||||
# A large media file uploaded in place is served half-written to anyone who
|
||||
# requests it mid-copy. Ship each directory to a sibling, then swap it in -
|
||||
# the swap is a rename, so the switch is atomic and visitors never see a
|
||||
# partial file.
|
||||
$skipDirs = @($script:JunkDirNames) + @('.github')
|
||||
if ($Proj.deploy -and $Proj.deploy.skipDirs) { $skipDirs += @($Proj.deploy.skipDirs) }
|
||||
# requests it mid-copy. Ship the directory to a sibling, then swap it in.
|
||||
$skipDirs = @('.git', '.pytest_cache', 'node_modules')
|
||||
$dirs = @(Get-ChildItem -LiteralPath $root -Directory | Where-Object { $skipDirs -notcontains $_.Name })
|
||||
foreach ($d in $dirs) {
|
||||
Write-Host " >> uploading $($d.Name)/ (recursive, staged)" -ForegroundColor DarkCyan
|
||||
@ -870,11 +978,10 @@ function Invoke-DockerDeploy {
|
||||
Write-DeployLocation -Proj $Proj
|
||||
}
|
||||
|
||||
# Pseudo-project "ztokens": not a zconfig entry, no compose stack, and entirely
|
||||
# optional. Runs `ztokens -Publish` from a sibling ztokens checkout so a site
|
||||
# that charts usage data has something current to ship. Missing checkout, or a
|
||||
# failure, warns rather than aborting the rest of the deploy list - it is a
|
||||
# nice-to-have refresh, not a deploy step.
|
||||
# Pseudo-project "ztokens": not a zconfig entry, no compose stack. Runs
|
||||
# `ztokens -Publish` from the sibling ztokens repo so the public zscripts page
|
||||
# has current data. A failure here warns rather than aborting the rest of the
|
||||
# deploy list - it's a nice-to-have refresh, not a deploy step.
|
||||
function Invoke-ZTokensPublish {
|
||||
Write-Host "`n=== ztokens: refreshing live-usage stats ===" -ForegroundColor Cyan
|
||||
$ztokensScript = Join-Path (Split-Path -Parent $PSScriptRoot) "ztokens\ztokens.ps1"
|
||||
|
||||
Loading…
Reference in New Issue
Block a user