Author SHA1 Message Date
mhoennigandClaude Opus 5 fb1db30f2a docs(prs): the init-config-resolution PR-doc
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 18:02:58 +02:00
mhoennigandClaude Opus 5 2cddd76c88 Read init's effective configuration from the repository root
`init --systemd` resolved the configuration through `Paths.get(".")` while
every other file it touches — the two generated config layers, an applied
fragment, the units, the .htaccess — goes through the git top level. The two
agree only when the process happens to run in the repository root; from a
subdirectory the command silently read a different repository's configuration
or none at all, so `--apply`'s promise that the fragment's port reaches the
generated unit did not hold, and the host integration was skipped as if no
publicBaseUrl were configured.

The root is passed down instead, matching every other caller of
`ConfigLoader.load`.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 17:50:58 +02:00
mhoennigandClaude Opus 5 7fd473b88b Warn instead of silently defaulting when init cannot load the config
`init --systemd` read the effective configuration through a catch-all that
swallowed every exception and continued with a default `ServerConfig`. A
repository whose `.werkator.yml` failed validation for any reason then looked
exactly like one with no `publicBaseUrl`: the `.htaccess` and the maintenance
page were skipped without a word, during initial deployment setup of all times.

The fallback stays — the units are still generated — but the exception message
is now printed. The configuration is loaded once per run so the three readers
inside `createSystemdFiles` do not repeat the warning.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 17:47:16 +02:00
092183ca30 Clone the watched repository from Gitea by default (#21)
The repository moved to git.javagil.de, but repo-init still defaulted to
the GitHub mirror, which is no longer updated.
A freshly cloned instance therefore watched a stale main.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: mhoennig <michael@hoennig.de>
Reviewed-on: #21
2026-09-04 13:33:08 +02:00
mhoennig 0761a274db Merge main into the deflake-maxconcurrent-test branch 2026-09-04 12:46:25 +02:00
mhoennigandClaude Opus 5 5c73c4cc21 docs(prs): the deflake-maxconcurrent-test PR-doc
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 08:53:00 +02:00
16543f038b the sandbox config section is werkdock, not bwrap (#19)
`bwrap` named the mechanism one layer below the tool that actually runs it: since
v1.0.0 Werkator does not invoke bwrap at all, it shells out to the werkdock CLI —
which made `bwrap.werkdock` a key naming its own executor.

The section is `werkdock` now and that key is `werkdock.binary`; BwrapConfig,
BwrapOverrides and BwrapBuildRunner follow the name. A file still writing `bwrap`
is read as before and warned about once per file, in `renameLegacySandbox` on the
raw map of every layer before merging — so nothing downstream knows two names, and
the old name is not a way around the pinning either. Renaming rather than refusing,
because the section lives in the machine configuration of every webspace instance,
which no repository tracks; the hard refusal belongs to the release that sets
ConfigVersions.FORMAT_BROKE_IN, where a file declaring no version can be caught
by name at all.

WERKATOR_SANDBOX in tools/remote follows, and still accepts `bwrap`.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: mhoennig <michael@hoennig.de>
Reviewed-on: #19
2026-09-03 20:41:18 +02:00
3ccc901d1b tools/remote drives any host layout (#18)
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: mhoennig <michael@hoennig.de>
Reviewed-on: #18
2026-09-03 20:37:58 +02:00
mhoennigandClaude Opus 5 9221550a2a test(build): gate the maxConcurrent-1 test instead of racing a sleep
The build of branch-a slept one second while the test asserted, without
any synchronization, that branch-b was still PENDING.
Under CPU contention on the mih09 host the sleep could elapse first, so
branch-b was already RUNNING or SUCCESS when the assertion ran.
branch-a now blocks until the test creates a gate file, and the test
first waits for branch-a to be RUNNING; the PENDING assertion no longer
depends on timing.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-03 19:32:22 +02:00
7 changed files with 245 additions and 13 deletions
@@ -0,0 +1,42 @@
> **WARNING:** This document describes only the change applied in this PR.
> It may already be outdated once the next PR is merged.
> Historic PR-documentation is not maintained along with new PRs — treat it as a snapshot, not as current documentation.
## The Problem
The build for commit `7028ca8` on `main` failed on the mih09 production instance with
`BuildExecutorTest > with maxConcurrent 1 a second branch stays PENDING until the first finished`,
while a retry of the very same commit passed, and the test passes locally.
The test was timing-dependent.
It started a build for `branch-a` whose build command was `sleep 1`, immediately started a second build for `branch-b`,
and then asserted — without any synchronization at all — that `branch-b` was still `PENDING`.
That assertion only held as long as the test thread reached it within the one second `branch-a` slept.
On a shared host under CPU contention the executor can get through `branch-a` entirely (queued, running, slept, succeeded) first,
and `branch-b` is then already `RUNNING` or `SUCCESS` when the assertion runs.
The failure therefore says nothing about the executor; it is pure scheduling noise that costs a build and a retry every time it hits.
## Non-Goals
- No change to production code — `BuildExecutor` is not touched, its queueing behaviour is unchanged.
- No sweep of the other timing-sensitive tests in the suite; only the one that actually flaked is fixed.
## The Solution
`branch-a` no longer sleeps for a fixed time, it blocks until the test says so:
its build command is `until [ -f gate ]; do sleep 0.05; done`, and the build workspace is the test's working directory.
The test now
1. waits (via `eventually`) until `branch-a` is `RUNNING`, so the single executor slot is provably occupied,
2. asserts that `branch-b` is `PENDING` — which cannot race anything, because `branch-a` cannot finish before the gate file exists,
3. creates the gate file, and only then awaits both builds' `SUCCESS`.
The assertion on the event transitions (`branch-b` goes `RUNNING` only after `branch-a` reached `SUCCESS`) is unchanged.
A blocking gate was chosen over a mocked `BuildRunner` because it keeps the test on the real `ProcessBuildRunner`,
so it still covers the actual process handling rather than only the executor's bookkeeping.
Verified by running `BuildExecutorTest` five times on an idle machine and three more times with twice `nproc` busy-loops saturating the CPU,
which is the condition that produced the original failure.
- [BuildExecutorTest](../../src/test/kotlin/de/hoennig/werkator/build/BuildExecutorTest.kt)
@@ -0,0 +1,30 @@
> **WARNING:** This document describes only the change applied in this PR.
> It may already be outdated once the next PR is merged.
> Historic PR-documentation is not maintained along with new PRs — treat it as a snapshot, not as current documentation.
## The Problem
The repository's home is `https://git.javagil.de/mi/werkator.git` since the move to the own Gitea instance,
but `tools/remote` still defaulted `WERKATOR_REPO_URL` to the GitHub mirror.
The mih09 production instance was cloned from that mirror and consequently watched a `main` that nobody pushes to any more:
it kept reporting the last GitHub state as green while three merged pull requests sat unbuilt on the real `main`.
The failure that started this — a flaky test already fixed on Gitea's `main` — could not be re-verified live,
because the instance had no way to see the fix.
## Non-Goals
- The existing clone on mih09 is not touched by this change; its remote was repointed by hand (`git remote set-url`), and `repo-init` skips an existing clone.
- The generic placeholder URLs in `docs/deployment.md` stay as they are — they describe cloning *any* watched repository, not werkator's own.
- No decision about mirroring to GitHub; the mirror simply stops being the source an instance builds from.
## The Solution
`REPO_URL` in [tools/remote](../../tools/remote) defaults to the Gitea URL, and the usage comment says so.
The value stays overridable via `WERKATOR_REPO_URL` in the instance's env file,
so an installation that deliberately watches a different remote is unaffected.
Anonymous HTTPS works against Gitea exactly as it did against GitHub, so no deploy key or token is involved.
Note that pull requests opened via AGit-Flow create no branch in Gitea,
so an instance watching this remote sees `main` only — branch builds require pushing real branches.
@@ -0,0 +1,90 @@
> **WARNING:** This document describes only the change applied in this PR.
> It may already be outdated once the next PR is merged.
> Historic PR-documentation is not maintained along with new PRs — treat it as a snapshot, not as current documentation.
## The Problem
`init --systemd` generates the host integration — the systemd unit's resource limits, the Apache `.htaccess` and the maintenance page — from the effective configuration.
It read that configuration through a helper with two independent defects, both of which fail silently.
**It swallowed every error.**
The load sat in `try { … } catch (_: Exception) { ServerConfig() }`.
Any configuration error at all — a missing required field, a malformed layer, a version floor violation — produced a default `ServerConfig` with a blank `publicBaseUrl`.
A repository with a broken `.werkator.yml` then looked exactly like one that simply has no public base URL configured:
the `.htaccess` and the maintenance page were skipped without a word.
This surfaced while verifying [PR#17](2026-09-03-PR%2317-maintenance-page.md) on mih09, where the missing files looked like an unconfigured `publicBaseUrl` and were in fact an unrelated validation error.
**It read from the wrong directory.**
The helper called `configLoader.load(Paths.get("."))` — the process's current directory — while everything else in the command works off the git top level resolved by `GitService.getTopLevel`.
`ConfigLoader.loadRaw` resolves the layers directly under the directory it is given and does not walk up to the repository root, so the two agree only when `init` happens to be invoked from the root itself.
From a subdirectory the command read another repository's configuration, or none.
That also broke `--apply`: the fragment is installed into the repository root deliberately before the systemd files are written, so that its port and limits reach the generated unit, and a current-directory read does not see it.
`init --systemd` runs during initial deployment setup, which is exactly when a silent wrong answer is most expensive.
## Non-Goals
- The fallback itself is kept: a configuration that cannot be loaded is not fatal for `init`, the units are still generated with the defaults.
During the very first bootstrap there is legitimately nothing to load yet.
- No change to `ConfigLoader`, to the configuration schema, or to any other command.
- No sweep for catch-all exception handlers elsewhere in the code base;
the two other `catch` blocks in `InitCommand` already print an `Error:` and abort, so they were only checked, not changed.
## The Scenarios
### Feature: init reports what it read and where it read it from
#### Background
- The *repository root* is the git top level as resolved by `GitService.getTopLevel`, the directory holding `.werkator.yml`, `.git/werkator/.werkator.yml` and an applied fragment.
- The *current directory* is the process working directory, which is the repository root only when `init` is invoked there.
#### Scenario#22.01: A broken configuration is named, not defaulted over
So that a validation error during deployment setup is not mistaken for an unconfigured installation.
- **Given** a repository whose effective configuration cannot be loaded
- **When** `init --systemd` runs
- **Then** the exception message is printed as a warning
- **and** the unit files are still generated with the default settings
- **and** the warning appears exactly once, although three settings are read from the configuration
##### Verified by
- [InitCommandTest: `--systemd warns once when the effective configuration cannot be loaded`](../../src/test/kotlin/de/hoennig/werkator/commands/InitCommandTest.kt)
#### Scenario#22.02: The configuration is read from the repository root
So that the generated host integration reflects the repository being initialized, whatever directory `init` was invoked from.
- **Given** a repository whose root configuration sets `server.publicBaseUrl` and `server.port`
- **and** a current directory that is not that repository root
- **When** `init --systemd` runs
- **Then** the `.htaccess` and the maintenance page are generated
- **and** the `.htaccess` proxies to the port from the root configuration
##### Verified by
- [InitCommandTest: `--systemd reads the configuration from the repository root, not the current directory`](../../src/test/kotlin/de/hoennig/werkator/commands/InitCommandTest.kt)
## The Solution
The catch-all now prints the exception message before falling back:
```
Warning: the effective configuration could not be loaded (<message>)
continuing with default server settings — check the generated unit and host files
```
The configuration is read three times while the systemd files are written (`memoryMax`, `tasksMax`, `publicBaseUrl`), which would repeat the warning three times.
It is therefore loaded once per run and cached in the command, and the cache is reset at the top of `run()` so a reused instance — the command is a Spring singleton — re-reads.
The repository root is passed down into the two accessors instead of `Paths.get(".")`.
This matches every other caller of `ConfigLoader.load` in the code base, all of which pass an explicit working directory;
`InitCommand` was the only one relying on the process's current directory.
Both fixes are the same failure in two forms — the command answered from a configuration it never actually read — which is why they are in one PR.
## Additional Changes
- None.
@@ -48,6 +48,7 @@ class InitCommand(
internal var javaExecutableResolver: () -> Path = { Paths.get(System.getProperty("java.home"), "bin", "java") } internal var javaExecutableResolver: () -> Path = { Paths.get(System.getProperty("java.home"), "bin", "java") }
override fun run() { override fun run() {
cachedServerConfig = null
val normalizedWorkingDir = workingDir.toAbsolutePath().normalize() val normalizedWorkingDir = workingDir.toAbsolutePath().normalize()
val root = val root =
try { try {
@@ -296,14 +297,31 @@ class InitCommand(
* already loadable (re-running `init --systemd` on an installed instance); during * already loadable (re-running `init --systemd` on an installed instance); during
* the very first bootstrap they stay unset and the defaults (no directives) apply. * the very first bootstrap they stay unset and the defaults (no directives) apply.
*/ */
private fun loadedSystemdConfig(): de.hoennig.werkator.config.SystemdConfig = loadedServerConfig().systemd private fun loadedSystemdConfig(root: Path): de.hoennig.werkator.config.SystemdConfig = loadedServerConfig(root).systemd
private fun loadedServerConfig(): de.hoennig.werkator.config.ServerConfig = /** Loaded once per run, so a broken configuration is reported once and not per caller. */
try { private var cachedServerConfig: de.hoennig.werkator.config.ServerConfig? = null
configLoader.load(Paths.get(".")).server
} catch (_: Exception) { /**
de.hoennig.werkator.config * Read from the repository root like every other file this command touches — the
.ServerConfig() * layers sit there, not in whatever directory the process happens to run in, and
* an applied fragment must reach the generated unit even when `init` is invoked
* from a subdirectory.
*
* A configuration error here is not fatal — the units are still generated with defaults —
* but it must not pass for "nothing configured": without the warning a broken `.werkator.yml`
* looks exactly like an unset `publicBaseUrl` and the host integration is skipped silently.
*/
private fun loadedServerConfig(root: Path): de.hoennig.werkator.config.ServerConfig =
cachedServerConfig ?: run {
try {
configLoader.load(root).server
} catch (e: Exception) {
println("Warning: the effective configuration could not be loaded (${e.message})")
println(" continuing with default server settings — check the generated unit and host files")
de.hoennig.werkator.config
.ServerConfig()
}.also { cachedServerConfig = it }
} }
private fun createSystemdFiles( private fun createSystemdFiles(
@@ -327,8 +345,8 @@ class InitCommand(
javaExecutable = javaExecutableResolver(), javaExecutable = javaExecutableResolver(),
jarPath = jarPath, jarPath = jarPath,
envFile = envFile, envFile = envFile,
memoryMax = loadedSystemdConfig().memoryMax, memoryMax = loadedSystemdConfig(root).memoryMax,
tasksMax = loadedSystemdConfig().tasksMax, tasksMax = loadedSystemdConfig(root).tasksMax,
), ),
) )
println("created ${unitFile.toFile().relativeTo(normalizedWorkingDir.toFile())}") println("created ${unitFile.toFile().relativeTo(normalizedWorkingDir.toFile())}")
@@ -351,7 +369,7 @@ class InitCommand(
// generated host integration like the units: only meaningful behind a web // generated host integration like the units: only meaningful behind a web
// frontend, so it needs a public base URL; unused elsewhere and harmless // frontend, so it needs a public base URL; unused elsewhere and harmless
val server = loadedServerConfig() val server = loadedServerConfig(root)
if (server.publicBaseUrl.isNotBlank()) { if (server.publicBaseUrl.isNotBlank()) {
val htaccessFile = werkatorDir.resolve(SystemdServiceFiles.HTACCESS_NAME) val htaccessFile = werkatorDir.resolve(SystemdServiceFiles.HTACCESS_NAME)
htaccessFile.toFile().writeText(SystemdServiceFiles.htaccessContent(server.port)) htaccessFile.toFile().writeText(SystemdServiceFiles.htaccessContent(server.port))
@@ -505,6 +505,8 @@ class BuildExecutorTest : FunSpec() {
} }
test("with maxConcurrent 1 a second branch stays PENDING until the first finished") { test("with maxConcurrent 1 a second branch stays PENDING until the first finished") {
// branch-a blocks on a gate file the test creates, so the PENDING assertion
// below cannot race the first build finishing on a loaded machine
val h = val h =
Harness( Harness(
""" """
@@ -512,7 +514,7 @@ class BuildExecutorTest : FunSpec() {
maxConcurrent: 1 maxConcurrent: 1
branches: branches:
branch-a: branch-a:
buildCommand: "sleep 1" buildCommand: "until [ -f gate ]; do sleep 0.05; done"
cleanCommand: "" cleanCommand: ""
branch-b: branch-b:
buildCommand: "echo ok" buildCommand: "echo ok"
@@ -523,8 +525,13 @@ class BuildExecutorTest : FunSpec() {
h.executor.startBuild(h.repo, "branch-a", "sha-a") h.executor.startBuild(h.repo, "branch-a", "sha-a")
h.executor.startBuild(h.repo, "branch-b", "sha-b") h.executor.startBuild(h.repo, "branch-b", "sha-b")
eventually(30.seconds) {
h.repository.latestFor("branch-a")?.status shouldBe BuildStatus.RUNNING
}
h.repository.latestFor("branch-b")?.status shouldBe BuildStatus.PENDING h.repository.latestFor("branch-b")?.status shouldBe BuildStatus.PENDING
Files.createFile(h.workingDir.resolve("gate"))
awaitStatus(h, "branch-b", BuildStatus.SUCCESS) awaitStatus(h, "branch-b", BuildStatus.SUCCESS)
awaitStatus(h, "branch-a", BuildStatus.SUCCESS) awaitStatus(h, "branch-a", BuildStatus.SUCCESS)
val transitions = h.events.map { it.result.branch to it.result.status } val transitions = h.events.map { it.result.branch to it.result.status }
@@ -252,5 +252,50 @@ class InitCommandTest : FunSpec() {
// This should not throw IllegalArgumentException // This should not throw IllegalArgumentException
initCommand.run() initCommand.run()
} }
test("--systemd reads the configuration from the repository root, not the current directory") {
val tempDir = Files.createTempDirectory("werkator-init-test")
// written before the run, so `init` keeps it instead of creating a template
tempDir.resolve(".werkator.yml").toFile().writeText(
"server:\n publicBaseUrl: \"https://werkator.example.org/\"\n port: 18099\n",
)
initCommand.workingDir = tempDir
initCommand.systemd = true
initCommand.jarPathResolver = { Paths.get("/home/ci/bin/werkator.jar") }
initCommand.javaExecutableResolver = { Paths.get("/usr/bin/java") }
every { gitService.getTopLevel(tempDir) } returns tempDir
every { gitService.getOriginUrl(tempDir) } returns "https://git.example.org/my-org/my-repo.git"
initCommand.run()
// the host integration is generated only when the root's config has a public base URL
val htaccess = tempDir.resolve(".git/werkator/${SystemdServiceFiles.HTACCESS_NAME}")
htaccess.toFile().shouldExist()
htaccess.toFile().readText() shouldContain "18099"
tempDir.resolve(".git/werkator/${SystemdServiceFiles.MAINTENANCE_PAGE_NAME}").toFile().shouldExist()
}
test("--systemd warns once when the effective configuration cannot be loaded") {
val tempDir = Files.createTempDirectory("werkator-init-test")
val brokenLoader = mockk<de.hoennig.werkator.config.ConfigLoader>()
every { brokenLoader.load(any()) } throws IllegalStateException("gitea.owner is required")
val command = InitCommand(gitService, brokenLoader)
command.workingDir = tempDir
command.systemd = true
command.jarPathResolver = { Paths.get("/home/ci/bin/werkator.jar") }
command.javaExecutableResolver = { Paths.get("/usr/bin/java") }
every { gitService.getTopLevel(tempDir) } returns tempDir
every { gitService.getOriginUrl(tempDir) } returns "https://git.example.org/my-org/my-repo.git"
val console = captureConsole { command.run() }
console.stdout shouldContain "gitea.owner is required"
// the three readers of the configuration must not repeat the warning
console.stdout.windowed("Warning:".length).count { it == "Warning:" } shouldBe 1
// the units are still written with the defaults
tempDir.resolve(".git/werkator/${SystemdServiceFiles.unitName(tempDir)}").toFile().shouldExist()
}
} }
} }
+2 -2
View File
@@ -44,7 +44,7 @@
# Optional in the env file: # Optional in the env file:
# WERKATOR_INIT_CONFIG the init fragment to apply (repo-init, instance-start) # WERKATOR_INIT_CONFIG the init fragment to apply (repo-init, instance-start)
# WERKATOR_REPO_URL https clone URL of the watched repository # WERKATOR_REPO_URL https clone URL of the watched repository
# (default: https://github.com/mhoennig/werkator.git) # (default: https://git.javagil.de/mi/werkator.git)
# WERKATOR_REPO_DIR directory of the watched repository, absolute or relative to # WERKATOR_REPO_DIR directory of the watched repository, absolute or relative to
# WERKATOR_PATH (default: werkator); it also names the systemd # WERKATOR_PATH (default: werkator); it also names the systemd
# unit, exactly as `init --systemd` derives it # unit, exactly as `init --systemd` derives it
@@ -122,7 +122,7 @@ require_env WERKATOR_REMOTE WERKATOR_PATH
HOST="$WERKATOR_REMOTE" HOST="$WERKATOR_REMOTE"
TARGET_DIR="$WERKATOR_PATH" TARGET_DIR="$WERKATOR_PATH"
ROOTFS="${WERKATOR_ROOTFS:-$REPO_ROOT/build/werkator-buildenv-trixie-java-go-node.tar.zst}" ROOTFS="${WERKATOR_ROOTFS:-$REPO_ROOT/build/werkator-buildenv-trixie-java-go-node.tar.zst}"
REPO_URL="${WERKATOR_REPO_URL:-https://github.com/mhoennig/werkator.git}" REPO_URL="${WERKATOR_REPO_URL:-https://git.javagil.de/mi/werkator.git}"
# The host layout is three values, not one convention: an installation that grew # The host layout is three values, not one convention: an installation that grew
# before this script existed puts them elsewhere, and the defaults are exactly what # before this script existed puts them elsewhere, and the defaults are exactly what