Merge branch 'claude/amazing-khayyam-38cad4': interrupt builds on server shutdown
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
@@ -56,6 +56,8 @@ Three places must stay in sync when config keys change: the `GitTallyConfig` dat
|
||||
|
||||
`BuildExecutor` runs builds asynchronously: up to `builds.maxConcurrent` branches concurrently (default 1), but never more than one build per branch at a time. Each branch builds in its own reusable git worktree at `.git/gittally/worktrees/<branchKey>` (`BranchWorkspaces`), checked out detached at the requested commit — the primary checkout is never used for builds. Status transitions are persisted via `BuildResultRepository` (JSON file under `.git/gittally/`), published to Gitea non-fatally, and emitted as `BuildStatusChangedEvent`s. Cancellation addresses a build by artifact key and terminates the whole process tree. Future code (watcher, server, UI) must not assume a single running build.
|
||||
|
||||
On context close (e.g. systemd SIGTERM), a `ContextClosedEvent` listener in `BuildExecutor` terminates the process trees of all executing builds and waits (bounded) until their results are persisted as INTERRUPTED — a shutdown is never recorded as FAILED. Builds still queued stay PENDING and start no process. Both are re-enqueued by the watcher's startup recovery; INTERRUPTED therefore publishes as Gitea state `pending`, not `failure` (`GiteaStateMapping`).
|
||||
|
||||
The runtime is selected per branch behind the `BuildRunner` interface: `DispatchingBuildRunner` (`@Primary`) routes to native `ProcessBuildRunner` (the default) or to `DockerBuildRunner` when `branches.<name>.docker.enabled`. The Docker runner shells out to the `docker` CLI (no SDK): it (re)builds the configured image when the Dockerfile inputs changed (tracked via the `org.gittally.build-inputs-sha256` image label), maintains a per-repo Gradle cache volume, mounts the worktree and the Docker socket into a labelled (`org.hoennig.gittally`) `--rm --init` container, and repairs workspace ownership in-container after each command (under a rootless daemon the container runs as root, which is the host user, and the repair degenerates to `0:0`). Git works inside the container: the primary `.git` is mounted read-only with `.git/gittally/` masked by an empty tmpfs (credential isolation) and the worktree's admin dir mounted read-write (`gitMetadataMounts`). The returned `Process` is the attached `docker run` client, so log streaming and termination work exactly like native builds.
|
||||
|
||||
## Watcher
|
||||
|
||||
@@ -0,0 +1,87 @@
|
||||
> **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
|
||||
|
||||
When the server is stopped (systemd SIGTERM) while a build is executing, the build process dies with the service.
|
||||
`BuildExecutor.execute` then classified the shutdown-induced exit as FAILED, because only the `cancelled` flag was checked.
|
||||
FAILED is terminal and not restartable, so the watcher's startup recovery did not re-enqueue the build.
|
||||
Additionally, a red `failure` status was posted to Gitea for a commit that was never actually built to failure.
|
||||
Observed on vm4006: a running Docker build died during a service restart and the commit silently stayed red.
|
||||
|
||||
## Non-Goals
|
||||
|
||||
- No change to explicit cancellation: a user-cancelled build stays CANCELLED and red.
|
||||
- No handling of `kill -9`: an unclean kill still leaves a stale RUNNING result, which the existing `markStaleRunningAsInterrupted` startup recovery already covers.
|
||||
- No draining of the Spring web layer or watcher — those already have their own shutdown hooks.
|
||||
|
||||
## The Scenarios
|
||||
|
||||
### Feature: builds interrupted by a server shutdown are recovered, not failed
|
||||
|
||||
#### Background
|
||||
|
||||
- INTERRUPTED and PENDING are restartable statuses; `Watcher.recoverOnStartup` re-enqueues the latest build of a branch in either status.
|
||||
- FAILED is terminal and never re-enqueued.
|
||||
|
||||
#### Scenario#000.01: An executing build is recorded as INTERRUPTED on shutdown
|
||||
|
||||
So that a service restart never loses a build or marks its commit as failed.
|
||||
|
||||
- **Given** a build is executing
|
||||
- **When** the application context closes (e.g. systemd SIGTERM)
|
||||
- **Then** the build's process tree is terminated
|
||||
- **and** the result is persisted as INTERRUPTED, not FAILED, before the context finishes closing
|
||||
- **and** the startup recovery re-enqueues the branch on the next start
|
||||
|
||||
##### Verified by
|
||||
|
||||
- [BuildExecutorTest — "shutdown kills an executing build and records INTERRUPTED, not FAILED"](../../src/test/kotlin/de/hoennig/gittally/build/BuildExecutorTest.kt)
|
||||
|
||||
#### Scenario#000.02: A queued build stays PENDING over a shutdown
|
||||
|
||||
So that queued builds survive a restart the same way executing builds do.
|
||||
|
||||
- **Given** a build is queued behind an executing build
|
||||
- **When** the application context closes
|
||||
- **Then** the queued build starts no process and gets no status transition
|
||||
- **and** it stays PENDING for the startup recovery, which re-enqueues it after the restart
|
||||
|
||||
##### Verified by
|
||||
|
||||
- [BuildExecutorTest — "a build still queued at shutdown stays PENDING for the startup recovery"](../../src/test/kotlin/de/hoennig/gittally/build/BuildExecutorTest.kt)
|
||||
|
||||
#### Scenario#000.03: No failure status is posted to Gitea for an interrupted build
|
||||
|
||||
So that a commit does not turn red because of a service restart.
|
||||
|
||||
- **Given** a build transitions to INTERRUPTED
|
||||
- **When** the status is published to Gitea
|
||||
- **Then** the commit-status state is `pending` (description "build interrupted"), not `failure`
|
||||
- **and** the re-enqueued build posts `pending` again after the restart
|
||||
|
||||
##### Verified by
|
||||
|
||||
- [GiteaStateMappingTest](../../src/test/kotlin/de/hoennig/gittally/gitea/GiteaStateMappingTest.kt)
|
||||
- [GiteaClientTest — "maps every build status to the documented Gitea state"](../../src/test/kotlin/de/hoennig/gittally/gitea/GiteaClientTest.kt)
|
||||
|
||||
## The Solution
|
||||
|
||||
`BuildExecutor` gets a `shuttingDown` flag set by a `ContextClosedEvent` listener (`shutdown()`).
|
||||
The listener terminates the process trees of all executing builds and waits (bounded, 20s) until their workers have persisted the INTERRUPTED results — the event fires before bean destruction, so the repository and the Gitea client are still usable.
|
||||
The final-status classification checks the flag: a non-zero exit or an exception during shutdown becomes INTERRUPTED instead of FAILED; explicit cancellation and a clean SUCCESS still win.
|
||||
Workers that pick up a queued build during shutdown return without any transition, leaving it PENDING.
|
||||
The command-start and between-commands gates also check the flag, so no new process is spawned (and orphaned) once shutdown began.
|
||||
`GiteaStateMapping` now publishes INTERRUPTED as `pending`: an interrupted build is re-enqueued by the startup recovery, so red would be wrong.
|
||||
|
||||
The hard invariants hold: nothing is scheduled (the listener is a lifecycle callback), and the listener is a no-op in CLI runs because `build`/`retry` block until completion before the context closes.
|
||||
On a Ctrl-C during a CLI build the same logic applies and correctly records INTERRUPTED.
|
||||
|
||||
## Additional Changes
|
||||
|
||||
- Documented the shutdown behavior in the [architecture skill](../../.claude/skills/architecture/SKILL.md).
|
||||
|
||||
## Follow-up PRs
|
||||
|
||||
- None planned.
|
||||
@@ -5,6 +5,8 @@ import de.hoennig.gittally.config.ConfigLoader
|
||||
import de.hoennig.gittally.gitea.GiteaClient
|
||||
import org.slf4j.LoggerFactory
|
||||
import org.springframework.context.ApplicationEventPublisher
|
||||
import org.springframework.context.event.ContextClosedEvent
|
||||
import org.springframework.context.event.EventListener
|
||||
import org.springframework.stereotype.Service
|
||||
import java.io.IOException
|
||||
import java.io.InputStream
|
||||
@@ -51,6 +53,9 @@ class BuildExecutor(
|
||||
@Volatile
|
||||
private var slots: Semaphore? = null
|
||||
|
||||
/** Set on context close; in-flight builds then finish as INTERRUPTED instead of FAILED. */
|
||||
private val shuttingDown = AtomicBoolean(false)
|
||||
|
||||
/** The builds currently executing, newest last (queued builds are PENDING in the repository). */
|
||||
fun currentBuilds(): List<RunningBuild> = builds.values.filter { it.running }.map { it.runningBuild }
|
||||
|
||||
@@ -108,9 +113,37 @@ class BuildExecutor(
|
||||
return true
|
||||
}
|
||||
|
||||
/**
|
||||
* Runs when the application context closes (e.g. systemd SIGTERM): terminates the
|
||||
* process trees of all executing builds and waits (bounded) until their INTERRUPTED
|
||||
* results are persisted, so a shutdown is never recorded as a build failure.
|
||||
* Builds still queued stay PENDING; the watcher's startup recovery re-enqueues
|
||||
* both PENDING and INTERRUPTED builds after the restart.
|
||||
*/
|
||||
@EventListener(ContextClosedEvent::class)
|
||||
fun shutdown() {
|
||||
if (!shuttingDown.compareAndSet(false, true)) {
|
||||
return
|
||||
}
|
||||
val executing = builds.values.filter { it.running }
|
||||
if (executing.isEmpty()) {
|
||||
return
|
||||
}
|
||||
log.info("shutdown requested; interrupting {} executing build(s)", executing.size)
|
||||
executing.forEach { build -> build.process?.let { destroyProcessTree(it) } }
|
||||
val deadline = System.nanoTime() + Duration.ofMillis(SHUTDOWN_DRAIN_TIMEOUT_MILLIS).toNanos()
|
||||
while (executing.any { builds.containsKey(it.runningBuild.artifactKey) } && System.nanoTime() < deadline) {
|
||||
Thread.sleep(50)
|
||||
}
|
||||
executing
|
||||
.filter { builds.containsKey(it.runningBuild.artifactKey) }
|
||||
.forEach { log.warn("build of branch {} was not recorded as interrupted in time", it.runningBuild.branch) }
|
||||
}
|
||||
|
||||
private fun execute(build: ActiveBuild) {
|
||||
var slot: Semaphore? = null
|
||||
var finalStatus = BuildStatus.FAILED
|
||||
// null keeps the persisted status untouched (a queued build stays PENDING over a shutdown)
|
||||
var finalStatus: BuildStatus? = BuildStatus.FAILED
|
||||
var workspace: Path? = null
|
||||
try {
|
||||
slot = slotsFor(build.workingDir)
|
||||
@@ -119,6 +152,10 @@ class BuildExecutor(
|
||||
finalStatus = BuildStatus.CANCELLED
|
||||
return
|
||||
}
|
||||
if (shuttingDown.get()) {
|
||||
finalStatus = null
|
||||
return
|
||||
}
|
||||
build.running = true
|
||||
build.runningBuild.runningSince = Instant.now()
|
||||
transition(build, BuildStatus.RUNNING, duration = null)
|
||||
@@ -134,20 +171,34 @@ class BuildExecutor(
|
||||
when {
|
||||
build.cancelled.get() -> BuildStatus.CANCELLED
|
||||
exitCode == 0 -> BuildStatus.SUCCESS
|
||||
shuttingDown.get() -> BuildStatus.INTERRUPTED
|
||||
else -> BuildStatus.FAILED
|
||||
}
|
||||
} catch (e: Exception) {
|
||||
finalStatus = if (build.cancelled.get()) BuildStatus.CANCELLED else BuildStatus.FAILED
|
||||
log.error("build of branch {} crashed", build.runningBuild.branch, e)
|
||||
appendToLiveLog(build, "\nbuild crashed: ${e.message}\n")
|
||||
finalStatus =
|
||||
when {
|
||||
build.cancelled.get() -> BuildStatus.CANCELLED
|
||||
shuttingDown.get() -> BuildStatus.INTERRUPTED
|
||||
else -> BuildStatus.FAILED
|
||||
}
|
||||
if (finalStatus == BuildStatus.FAILED) {
|
||||
log.error("build of branch {} crashed", build.runningBuild.branch, e)
|
||||
appendToLiveLog(build, "\nbuild crashed: ${e.message}\n")
|
||||
}
|
||||
} finally {
|
||||
// pure build time, without the queue wait; null when the build never started executing
|
||||
val duration = build.runningBuild.runningSince?.let { Duration.between(it, Instant.now()) }
|
||||
val result = transition(build, finalStatus, duration)
|
||||
try {
|
||||
artifactStore.persist(result, build.runningBuild.stagingDir, workspace)
|
||||
} catch (e: Exception) {
|
||||
log.warn("could not persist artifacts of {}: {}", result.artifactKey, e.message)
|
||||
if (finalStatus == BuildStatus.INTERRUPTED) {
|
||||
log.info("build of branch {} interrupted by shutdown", build.runningBuild.branch)
|
||||
appendToLiveLog(build, "\nbuild interrupted by shutdown\n")
|
||||
}
|
||||
if (finalStatus != null) {
|
||||
// pure build time, without the queue wait; null when the build never started executing
|
||||
val duration = build.runningBuild.runningSince?.let { Duration.between(it, Instant.now()) }
|
||||
val result = transition(build, finalStatus, duration)
|
||||
try {
|
||||
artifactStore.persist(result, build.runningBuild.stagingDir, workspace)
|
||||
} catch (e: Exception) {
|
||||
log.warn("could not persist artifacts of {}: {}", result.artifactKey, e.message)
|
||||
}
|
||||
}
|
||||
builds.remove(build.runningBuild.artifactKey)
|
||||
slot?.release()
|
||||
@@ -193,7 +244,7 @@ class BuildExecutor(
|
||||
return cleanExitCode
|
||||
}
|
||||
}
|
||||
if (build.cancelled.get()) {
|
||||
if (build.cancelled.get() || shuttingDown.get()) {
|
||||
return CANCELLED_EXIT_CODE
|
||||
}
|
||||
return runCommand(build, branchConfig, branchConfig.buildCommand, workspace, stdoutLog, stderrLog, liveLog)
|
||||
@@ -220,15 +271,15 @@ class BuildExecutor(
|
||||
branchConfig = branchConfig,
|
||||
onAuxProcess = { aux ->
|
||||
// preparation phases (e.g. a Docker image build) must die on cancellation
|
||||
// too, otherwise a cancelled build blocks its slot until they finish
|
||||
// and shutdown too, otherwise they block their slot until they finish
|
||||
build.process = aux
|
||||
if (build.cancelled.get()) {
|
||||
if (build.cancelled.get() || shuttingDown.get()) {
|
||||
destroyProcessTree(aux)
|
||||
}
|
||||
},
|
||||
)
|
||||
build.process = process
|
||||
if (build.cancelled.get()) {
|
||||
if (build.cancelled.get() || shuttingDown.get()) {
|
||||
destroyProcessTree(process)
|
||||
}
|
||||
val stdoutPump = pump(process.inputStream, stdoutLog, liveLog)
|
||||
@@ -409,5 +460,6 @@ class BuildExecutor(
|
||||
const val LIVE_LOG_FILE = "build.log"
|
||||
private const val CANCELLED_EXIT_CODE = 130
|
||||
private const val PUMP_DRAIN_TIMEOUT_MILLIS = 10_000L
|
||||
private const val SHUTDOWN_DRAIN_TIMEOUT_MILLIS = 20_000L
|
||||
}
|
||||
}
|
||||
|
||||
@@ -2,12 +2,17 @@ package de.hoennig.gittally.gitea
|
||||
|
||||
import de.hoennig.gittally.build.BuildStatus
|
||||
|
||||
/** Gitea commit-status state published for this build status. */
|
||||
/**
|
||||
* Gitea commit-status state published for this build status.
|
||||
* INTERRUPTED publishes as `pending`, not `failure`: an interrupted build (e.g. a
|
||||
* server shutdown) is re-enqueued by the startup recovery, so the commit must not
|
||||
* turn red in between.
|
||||
*/
|
||||
fun BuildStatus.toGiteaState(): String =
|
||||
when (this) {
|
||||
BuildStatus.SUCCESS -> "success"
|
||||
BuildStatus.FAILED, BuildStatus.INTERRUPTED, BuildStatus.CANCELLED -> "failure"
|
||||
BuildStatus.PENDING, BuildStatus.RUNNING -> "pending"
|
||||
BuildStatus.FAILED, BuildStatus.CANCELLED -> "failure"
|
||||
BuildStatus.PENDING, BuildStatus.RUNNING, BuildStatus.INTERRUPTED -> "pending"
|
||||
}
|
||||
|
||||
/**
|
||||
|
||||
@@ -292,6 +292,74 @@ class BuildExecutorTest : FunSpec() {
|
||||
}
|
||||
}
|
||||
|
||||
test("shutdown kills an executing build and records INTERRUPTED, not FAILED") {
|
||||
val h = harness("echo \$\$ > pid-file; sleep 30")
|
||||
|
||||
val build = h.executor.startBuild("main", "abc123", h.workingDir)
|
||||
eventually(10.seconds) {
|
||||
Files.exists(h.workingDir.resolve("pid-file")).shouldBeTrue()
|
||||
}
|
||||
val pid =
|
||||
Files
|
||||
.readString(h.workingDir.resolve("pid-file"))
|
||||
.trim()
|
||||
.toLong()
|
||||
|
||||
h.executor.shutdown()
|
||||
|
||||
// shutdown() drains synchronously, so the result is already persisted
|
||||
h.repository.latestFor("main")?.status shouldBe BuildStatus.INTERRUPTED
|
||||
h.executor.currentBuilds().shouldBeEmpty()
|
||||
ProcessHandle.of(pid).map { it.isAlive }.orElse(false) shouldBe false
|
||||
h.events.map { it.result.status } shouldContainExactly
|
||||
listOf(BuildStatus.PENDING, BuildStatus.RUNNING, BuildStatus.INTERRUPTED)
|
||||
Files.readString(build.liveLogFile) shouldContain "build interrupted by shutdown"
|
||||
verify { h.giteaClient.publishStatus("abc123", BuildStatus.INTERRUPTED, any(), null, h.workingDir) }
|
||||
verify { h.artifactStore.persist(match { it.status == BuildStatus.INTERRUPTED }, build.stagingDir, any()) }
|
||||
}
|
||||
|
||||
test("a build still queued at shutdown stays PENDING for the startup recovery") {
|
||||
val h = harness("sleep 30")
|
||||
|
||||
val first = h.executor.startBuild("main", "sha-1", h.workingDir)
|
||||
val second = h.executor.startBuild("main", "sha-2", h.workingDir)
|
||||
eventually(10.seconds) {
|
||||
h.repository
|
||||
.history()
|
||||
.first { it.artifactKey == first.artifactKey }
|
||||
.status shouldBe BuildStatus.RUNNING
|
||||
}
|
||||
|
||||
h.executor.shutdown()
|
||||
|
||||
h.repository
|
||||
.history()
|
||||
.first { it.artifactKey == first.artifactKey }
|
||||
.status shouldBe BuildStatus.INTERRUPTED
|
||||
// the branch worker drops the queued build without a transition or a process
|
||||
awaitIdle(h)
|
||||
h.repository
|
||||
.history()
|
||||
.first { it.artifactKey == second.artifactKey }
|
||||
.status shouldBe BuildStatus.PENDING
|
||||
h.events
|
||||
.filter { it.result.artifactKey == second.artifactKey }
|
||||
.map { it.result.status } shouldContainExactly listOf(BuildStatus.PENDING)
|
||||
verify(exactly = 0) { h.giteaClient.publishStatus("sha-2", BuildStatus.RUNNING, any(), any(), any()) }
|
||||
}
|
||||
|
||||
test("shutdown without any build in flight is a no-op") {
|
||||
val h = harness("echo ok")
|
||||
|
||||
h.executor.startBuild("main", "abc123", h.workingDir)
|
||||
awaitStatus(h, "main", BuildStatus.SUCCESS)
|
||||
awaitIdle(h)
|
||||
|
||||
h.executor.shutdown()
|
||||
|
||||
h.repository.latestFor("main")?.status shouldBe BuildStatus.SUCCESS
|
||||
}
|
||||
|
||||
test("cancel with an unknown artifact key returns false") {
|
||||
val h = harness("echo ok")
|
||||
|
||||
|
||||
@@ -121,10 +121,11 @@ class GiteaClientTest :
|
||||
mapOf(
|
||||
BuildStatus.SUCCESS to "success",
|
||||
BuildStatus.FAILED to "failure",
|
||||
BuildStatus.INTERRUPTED to "failure",
|
||||
BuildStatus.CANCELLED to "failure",
|
||||
BuildStatus.PENDING to "pending",
|
||||
BuildStatus.RUNNING to "pending",
|
||||
// interrupted builds are re-enqueued on startup, so the commit must not turn red
|
||||
BuildStatus.INTERRUPTED to "pending",
|
||||
)
|
||||
|
||||
expectedStates.forEach { (status, state) ->
|
||||
|
||||
@@ -11,10 +11,11 @@ class GiteaStateMappingTest :
|
||||
test("maps each build status to its Gitea state") {
|
||||
BuildStatus.SUCCESS.toGiteaState() shouldBe "success"
|
||||
BuildStatus.FAILED.toGiteaState() shouldBe "failure"
|
||||
BuildStatus.INTERRUPTED.toGiteaState() shouldBe "failure"
|
||||
BuildStatus.CANCELLED.toGiteaState() shouldBe "failure"
|
||||
BuildStatus.PENDING.toGiteaState() shouldBe "pending"
|
||||
BuildStatus.RUNNING.toGiteaState() shouldBe "pending"
|
||||
// interrupted builds are re-enqueued on startup, so the commit must not turn red
|
||||
BuildStatus.INTERRUPTED.toGiteaState() shouldBe "pending"
|
||||
}
|
||||
|
||||
test("maps Gitea states back to build statuses") {
|
||||
|
||||
Reference in New Issue
Block a user