Service containers were started and then left alone, so a job's first step could run while a database was still starting up. The runner now waits for every service whose image or `options` declare a healthcheck, as GitHub does. An unhealthy service fails the job with its container log, one that never becomes healthy fails it after `container.service_ready_timeout` (default `5m`, negative disables the wait), and one that exits without a healthcheck only gets its log and a warning.
The started containers also fill the `job` context, whose fields existed but were never populated: `job.container.{id,network}` and `job.services.<id>.{id,network,ports}`. `ports` is keyed by the plain container port, so `job.services.postgres.ports['5432']` resolves to the host port Docker picked.
---------
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1107
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
`::set-env::` and `::add-path::` let a step rewrite the environment of every later step from its own output, which the runner honoured silently. They are now refused, as GitHub has done since 2020, and `ACTIONS_ALLOW_UNSECURE_COMMANDS` opts back in per step or job. Support for that variable is new here too, and is the only opt-in, matching GitHub rather than adding a runner config key on top.
Annotations keep their source location: Gitea has no annotation store and its web UI strips command properties, so `::error file=main.go,line=12::msg` is rendered as `::error::main.go:12: msg`.
`DEVELOPMENT.md` writes down the log line encoding rules this relies on.
---------
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1109
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
`ACTIONS_RESULTS_URL` names one origin serving every `github.actions.results.api.v1` service. Gitea serves the artifact half and this runner the cache half, so announcing `ACTIONS_CACHE_SERVICE_V2` while that URL pointed at Gitea was a promise the environment could not keep, and `docker buildx` posted its cache calls at Gitea and got a 404.
The cache server now forwards the artifact half to the instance each job registers with, so it is the whole results service and jobs are pointed at it. The announcement follows, and the bundle patch follows the cache URL instead.
Also fixes three things no JavaScript client reached: camelCase in the v2 responses where the Go clients read proto names, the missing `x-ms-request-id` on blob uploads that panics buildkit, and `cache.external_server` passed through without the trailing slash the v1 client concatenates onto.
Tests run the real actions against the services they look for: `actions/cache` over both API versions, the artifact actions up and back down through the forwarding, and `setup-node`. The regression itself is covered by asserting that whatever a job is handed as `ACTIONS_RESULTS_URL` answers a cache service call.
Fixes https://gitea.com/gitea/runner/issues/1139
Reviewed-on: https://gitea.com/gitea/runner/pulls/1141
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
`coerceToString` covered only a few kinds and returned the unconverted `reflect.Value` for the rest, so callers rendered contexts as `<*model.GithubContext Value>` and mangled sized integers and floats the same way. It now returns a string, which makes that placeholder unrepresentable, and is exported as `CoerceToString` so Gitea can drop its own diverging copy. It also takes an already reflected value, so internal callers need no conversion and no guard against the zero `Value`.
`format()` panicked whenever a lone `}` was followed by anything other than another `}`. Only a trailing `}` reached the existing unmatched-brace check, so an expression such as `format('a}b')`, which any workflow can write, took the process down instead. It now returns that same error.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1135
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
1. Drop `Masterminds/semver`, the Docker API version check needs dotted-numeric comparison and moby's client ships `versions.GreaterThanOrEqualTo`.
1. A malformed API version now reports as unsupported instead of panicking.
1. Drop `pkg/errors`, nothing used stack traces, `Wrap` or `Cause`.
1. Drop its depguard rule, and the `io/ioutil` one that only masked staticcheck's `SA1019`.
1. Re-sync `act/container/docker_cli.go` and `act/container/docker_cli_test.go`, copies of docker/cli's `opts.go` and `opts_test.go`, from a March and a 2022 commit to the version go.mod pins.
1. Record the local deviations in each header.
1. Fix `invalidParameter`, which lacked `Unwrap`, hiding the wrapped cause from `errors.Is` and `errors.As`.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1132
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
Docker 29.7 extracts copied archives through `os.Root`, which refuses to follow a symlink to an absolute target, so copying into `/var/run/act` fails with `path escapes from parent` on the many images that link `/var/run` to `/run`. The daemon now resolves every path component for us before any tar entry name is built, and the destination is created with one directory entry per missing component, which no daemon version rejects.
Verified against real daemons (29.4.0, 29.5.0, 29.5.1, 29.5.3, 29.6.2, 29.7.0-rc.1) with `debian:bookworm` (absolute symlink) and `alpine:3` (relative symlink), and against `moby/go-archive` v0.2.0 through the pending fix branch.
1. Fixes https://gitea.com/gitea/runner/issues/1128
1. Upstream bug: https://github.com/moby/moby/issues/53258
1. Supersedes the no-op change in https://gitea.com/gitea/runner/pulls/1129, which cannot help since the daemon strips leading slashes itself
Reviewed-on: https://gitea.com/gitea/runner/pulls/1130
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
`CopyTarStream` creates the destination directory by extracting a one-entry tarball with `DestinationPath: "/"`, but named that entry with the absolute `destPath`. Docker Engine 29.5+ tightened path validation on the copy API and rejects absolute entry names against a `/` destination with `statat var/run/act/actions/<sha>: path escapes from parent`, so the action directory never reached the job container and `actions/checkout` failed during "Set up job". Stripping the leading slash makes the entry relative, matching what the sibling `copyDir` already does and the upstream fix in nektos/act v0.2.89. Adds a regression test asserting the mkdir tarball entry is relative.
Fixes#1128
Reviewed-on: https://gitea.com/gitea/runner/pulls/1129
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Workflow-controlled `jobs.<job>.container.options` were merged directly into the
Docker `HostConfig`. When the runner's privileged mode is disabled, only
`Privileged` was forced to `false` — host namespace flags, capability expansion,
security-profile overrides, and device/runtime access from the workflow YAML
survived into the final `HostConfig`. A workflow author could therefore enter
host PID/IPC namespaces and execute commands as root on the runner host:
```yaml
container:
image: ubuntu:22.04
options: >-
--pid=host --ipc=host --cap-add=ALL
--security-opt seccomp=unconfined --security-opt apparmor=unconfined
```
## Fix
`mergeContainerConfigs()` now strips the dangerous options-derived `HostConfig`
fields before merging when privileged mode is off: `PidMode`, `IpcMode`,
`UTSMode`, `CgroupnsMode`, `UsernsMode`, `CapAdd`, `SecurityOpt`, `Devices`,
`DeviceCgroupRules`, `DeviceRequests`, `VolumesFrom`, `Runtime`, `CgroupParent`,
and `Sysctls`. Each strip emits a warning, matching the existing
`--network ignored` handling. Options remain honored when privileged mode is
enabled, since the administrator has already opted into host access.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1058
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Only the verbatim value of a secret was masked, so a secret leaked through an action that serialized it stayed readable: `toJSON(secrets)` escapes it, an Authorization header carries it base64-encoded, a URL percent-encodes it. Each secret and `::add-mask::` value is now masked in those forms too, matching the value encoders of GitHub's runner. Encodings that leave the value unchanged are skipped, so a plain token still costs a single replacement pair. Includes regression tests.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1108
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Serves `github.actions.results.api.v1.CacheService` next to the v1 cache API, sharing its store, plus the subset of the Azure blob protocol the toolkit uploads with. On by default via `cache.v2`, and works with `external_server`.
Clients reach it through two edits in the action's own bundle: the GHES check is opened, and the cache service URL is taken from `ACTIONS_CACHE_URL`.
The same GHES check is what makes the stock `actions/upload-artifact` and `download-artifact` abort on Gitea. Opening it makes them work without the `gitea-upload-artifact` fork, from `upload-artifact@v4.4.0` on.
Verified against 118 real bundles, every major version of 16 actions: 92 patched, the rest deliberately left alone, and every patched bundle checked with `node --check`. Also end to end against pinned `actions/cache@v6.1.0` with an unreachable results URL, so only the patch can make the cache work.
---------
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: bircni <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1110
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Leaked per-job networks each hold a subnet of the daemon's address pool and nothing reclaimed them, so a host eventually fails every job with `all predefined address pools have been fully subnetted`. This is what CI hit in https://gitea.com/gitea/runner/actions/runs/742177.
- teardown no longer loses a container, and its network with it: a failed `ContainerRemove` was reported as success, a container whose id was never learned was skipped, and the daemon's own `AutoRemove` teardown was raced
- the idle cleanup reclaims what teardown cannot: networks carry `com.gitea.runner.uuid`, so a runner only touches its own, and a cutoff keeps a job starting during the pass out of scope
- pull failures reported mid-stream were discarded, surfacing later as a confusing `No such image`; they now propagate, and fall back to a local copy instead of failing the job
- `NetworkCreate` retries pool exhaustion briefly, then says which knobs to turn
- digest-pinned images are not re-pulled, and removal kills first so it never waits out Podman's stop timeout (measured 10.1s → 0.09s per container)
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1124
Reviewed-by: bircni <bircni@icloud.com>
Set `http_proxy`, `https_proxy` and `no_proxy` in the runner's environment and everything the runner controls uses them.
Go already read them for the runner's own requests. This adds jobs, in lower and upper case, service containers, and Dockerfile action builds.
Some hosts are added to `no_proxy` for jobs so they stay direct: the cache server, loopback, the job's service containers, and a `tcp://` Docker daemon. Without the last one the Docker client sends its API calls to the proxy and docker-in-docker breaks. Gitea is not added.
Images are pulled by the Docker daemon, which has its own proxy setting. In the `dind` images it reads these same variables. The runner warns at startup if it has a proxy and the daemon does not.
Fixes https://gitea.com/gitea/runner/issues/1118, originally reported as https://gitea.com/gitea/runner/issues/708.
---------
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1112
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
An unchecked type assertion panics on input it did not expect, as a missing `tool_cache` key did in https://gitea.com/gitea/runner/pulls/1122.
Every flagged site is now handled where it can fail, or typed so it cannot: a `lock.Keyed` replaces the two `sync.Map` mutex registries, and the reporter's outputs carry an explicit sent flag. Mocks keep their assertions, a mismatch there is a setup error the panic names.
Bugs it turned up (only the first is reachable from workflows):
1. A scalar `matrix.include` or `matrix.exclude`, e.g. `include: foo` or `include: [1, 2]`, panicked the runner with `interface conversion: interface {} is string, not map[string]interface {}`. Verified against `main`, it is now a workflow error. `OnSchedule` panicked the same way on a malformed `on.schedule` entry.
1. `ExternalURL()` panicked on the nil listener after `Close()`, the port is now resolved once at startup.
1. `errors.Is(err, git.ErrShortRef)` followed by `err.(*git.Error)` panics as soon as anything wraps that error, so it is `errors.As` now.
1. An output name the server acknowledged without ever being sent one was recorded as sent forever, which silently dropped a later value for that name.
48576ab3e5 fixes one discovered issue: a matrix key holding a nested object was logged and then run as if the job had no matrix, so it now fails like an unknown `exclude` key.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1123
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
1. Docker rejects two mounts on one target, so a `container.volumes:` or `--volume` aimed at `/opt/hostedtoolcache` failed the job with `Duplicate mount point`. Job and service volumes now displace the mount on the same path, and `name:/target:ro` no longer mounts read-write at the literal path `/target:ro`.
1. Setting `RUNNER_TOOL_CACHE` only changed what the variable said, the cache stayed where it was, so tools writing to it landed outside the mount and `${{ runner.tool_cache }}` disagreed with the variable. It now relocates the cache. Leaving it unset behaves as before.
1. Unknown `config.yaml` keys now warn instead of being dropped without a trace.
Fixes https://gitea.com/gitea/runner/issues/813
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1122
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
Adds `runner.hooks.job_started` and `runner.hooks.job_completed`: operator scripts that run inside the job environment, before the job's first step and after its last one.
```yaml
runner:
hooks:
job_started: /hooks/started.sh
job_completed: /hooks/completed.sh
```
Equivalent to GitHub's `ACTIONS_RUNNER_HOOK_JOB_STARTED` / `ACTIONS_RUNNER_HOOK_JOB_COMPLETED`, which are read when unset: output is scanned for workflow commands, `$GITHUB_ENV` and `$GITHUB_PATH` are read back, and a non-zero exit fails the job.
Fixes: https://gitea.com/gitea/runner/issues/779
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1111
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Full suite 178s → 96s, and a run's log 12MB → 2KB. `act/runner` was 177s of the 178s, serialised behind one docker daemon.
- Its fixtures now run in parallel, bounded by a slot count instead of `go test -parallel`, with a per-test container name prefix and a pinned `MaxParallel`.
- Fixtures asserting a job failure leaked their container and network (`AutoRemove` was at the act-CLI default), filling the daemon's address pool over time.
- Replaced sleeps used as synchronisation in the parallel-executor and cache-handler tests.
- Dropped duplicate coverage: two files re-testing `NewParallelExecutor`, a test asserting on its own semaphore, and `TestDockerActionForcePullForceRebuild`, whose config `runTest` discarded.
- `fmt-check`/`security-check` move from `make test` to a `checks` target; `security-check` no longer installs `xgo` and `gxz`.
- Test flags follow gitea: `GOTEST_FLAGS ?= -race -timeout 20m -parallel 8`, with `-cover`/`-coverprofile` left in the target. Dropped `-v`, since a failing package still prints its full output without it.
Coverage unchanged at 73.4%.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1121
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
A docker action can declare `runs.pre-entrypoint` and `runs.post-entrypoint` next to `runs.entrypoint` — the docker equivalent of a javascript action's `runs.pre`/`runs.post`. Neither key existed on `ActionRuns`, so the YAML decoder dropped them and docker actions silently skipped their setup and cleanup stages.
Only the entrypoint is stage specific. [`ContainerActionHandler`](https://github.com/actions/runner/blob/main/src/Runner.Worker/Handlers/ContainerActionHandler.cs) selects `Data.Pre`/`Data.Post` per stage but evaluates `runs.args` and `runs.env` unconditionally, so `docker create` receives the action's args and env on every stage. The `entrypoint` input stays main only, because it is read inside the main branch.
The stage is carried as the existing `stepStage` value rather than a separate parameter, and the pre and post stages reuse the action path resolution that `runPreStep`/`runPostStep` already open-coded three times, so the diff also drops those copies.
Known gap: `stepActionLocal.pre()` is a no-op for every `using`, so `pre-entrypoint` does not run for `uses: ./local-action` while `post-entrypoint` does. Closing it requires reading the action model before the main stage and adding a pre case to `getIfExpression` for `pre-if`, which also changes local node, go and composite actions — better as its own change.
---------
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1106
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
Passes the `runner` context values act already knew but never reported to jobs:
1. `runner.name` / `RUNNER_NAME` — registered runner name, hostname for `exec`
2. `runner.environment` / `RUNNER_ENVIRONMENT` — `self-hosted`
3. `RUNNER_WORKSPACE` — parent of `GITHUB_WORKSPACE`
4. `runner.debug` / `RUNNER_DEBUG` — `1` when `ACTIONS_STEP_DEBUG` is set
`ImageOS` now prefers the release named in the resolved image tag, so `ubuntu-latest` mapped to `runner-images:ubuntu-24.04` reports `ubuntu24` instead of the hardcoded `ubuntu20`. The `runs-on` label stays the fallback.
---------
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: bircni <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1105
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
Command data is percent-escaped on the wire because Gitea's job log cannot carry a newline, and the renderer decodes it.
1. Unescaping iterated a map, so order was random per process: `%250A` decoded to a newline instead of a literal `%0A` about two runs in three. Now a precompiled `strings.NewReplacer`.
1. Escape the data of command lines the runner writes itself (`##[error]`, `::group::Run …`), so decoding returns the original text instead of mangling a `%`. Multi-line runner errors also get real line breaks instead of a literal `\n`.
1. Register secrets in escaped form too — one containing `%` reaches the log as `%25…` and was never masked.
1. Decode in `jobLogFormatter`, so `gitea-runner exec` matches the web view.
## Relation to the Gitea PR
https://github.com/go-gitea/gitea/pull/38659 makes the renderer decode command data. Point 2 is required by it; the rest stand alone. Either order works — until both land, a new runner on old Gitea shows `%25`, an old runner on new Gitea shows the mangling point 2 fixes.
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1120
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
Fixes https://gitea.com/gitea/runner/issues/667
Fixes https://gitea.com/gitea/runner/pulls/1103
`container.options` is parsed with docker/cli's `addFlags`, which omits the flags docker/cli registers on the `create` and `run` commands themselves. Those were rejected as unknown — most visibly `--platform`.
| Flag | Behavior | Why |
| --- | --- | --- |
| `--platform` | overrides the runner-wide container architecture | the only way to pick an image architecture per job, e.g. across a matrix |
| `--pull always\|missing\|never` | `always` forces a pull, `never` skips it | `force_pull` is runner-wide config today, with no per-job control |
Both are resolved in `NewContainer` because the image pull runs before the container is created and the two have to agree.
`--name`, `--quiet` and `--disable-content-trust` are now accepted and ignored, and `--use-api-socket` is rejected pointing at `container.docker_host`, so a valid `docker create` line no longer fails outright.
<sub>Written by Claude (Opus 4.8).</sub>
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1104
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
Fixes https://gitea.com/gitea/runner/issues/659
Docker teardown was chained with `Then`, which stops on the previous step's error and on a cancelled context, so the first failure orphaned everything downstream. A volume that is still in use, or a flaky daemon, left the service containers and the job network behind — matching the reports in the issue, where the leaks show up after failed or overloaded jobs.
Cleanup is now straight-line and best-effort: every step runs regardless of what failed before it, and the errors are joined. Two things follow from that:
- Service containers are removed before the job volumes, since a service can hold one via `--volumes-from` in its options. The network stays last, once every container has detached.
- Only job container and volume errors are returned. Service and network errors are logged, as before, because `stopJobContainer()` also runs as the pre-flight step of job start, where a network cleanup error must not abort the job.
The nil check on `rc.JobContainer` is kept, but it is not the leak reporters are hitting: `container.NewContainer` never returns nil in the docker build, so cleanup with no job container is only reachable in the `WITHOUT_DOCKER` stub.
Addresses https://gitea.com/gitea/runner/pulls/1066#issuecomment-1239073.
Not covered here: the `buildx_buildkit_*` volumes reported in the issue are created by buildx inside the job, not by the runner, so no runner-side teardown removes them.
---------
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1066
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
## Problem
On host-mode Windows runners, a job can leave processes running after it finishes,
and those leftovers hold file handles that block deletion of the workspace.
Today a step's tree is only torn down when the step is *cancelled*
(`process.Killer`). A step that completes leaves whatever it spawned alive, and
the existing workspace scan in `terminateRunningProcesses` misses two shapes of
leftover: orphans whose parent already exited (no tree to walk, and their
executable often lives outside the workspace), and processes that merely *run in*
the workspace but reference no path from it — `Win32_Process` exposes no working
directory, so the scan cannot match them.
## Solution
Two additions in `internal/pkg/process`, both no-ops outside Windows:
- **`process.Group`** — a job-scoped Windows Job Object created with
`JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE`. Step processes are assigned to it before
they get their own `Killer`, so the step's job nests inside the job's:
cancellation still kills exactly the step's tree, while `Remove` closing the
group makes the kernel terminate everything still assigned, whatever its
parentage. The kernel also drops the handle when the runner exits, so a crashed
runner cannot strand processes.
- **`process.KillProcessesWithCWDUnder`** — a best-effort net for processes that
never joined the job (started via a service or scheduled task). It reads each
process's working directory from its PEB and terminates those under a workspace
dir. Processes it cannot open are skipped.
---------
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1080
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Reshapes the job log's "Set up job" section to mirror `actions/runner`:
- runner name/version, then `Runner Information` (labels, task, job, repository, event) and `Operating System` groups
- every required action downloaded up front under `Prepare all required actions`, each as `Download action repository '<action>@<ref>' (SHA:<sha>)`
- `Complete job name` closes the section
Downloading up front is the one behavioral change: the same set was already fetched during the pre stage regardless of a step's `if`, now just before the first pre step, so a download failure is reported against the job rather than a step.
---------
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1089
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
`reportStepError` reported every step error as FAILURE, including a `context.Canceled` from a docker file-command read cancelled at job finalization — non-deterministic red CI. Classify `context.Canceled` as an interruption instead (deferring to the job context), so a genuine cancel reports cancelled and a stray teardown cancellation is ignored, never a failure.
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1095
Reviewed-by: bircni <bircni@icloud.com>
Containers set `HostConfig.AutoRemove` but act also removes them explicitly, so the two removers race and the loser logs a 409 `removal of container X is already in progress` — seen at the end of nearly every `uses: docker://` step.
The explicit remove is redundant for `docker://` steps and docker actions (`Start(true)` already awaited exit), so it's skipped. Job and service containers keep both removers — their `sleep` entrypoint needs `AutoRemove` as a fallback reaper — so there the race is inherent and `remove()` now treats `NotFound` and `Conflict` as success.
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1093
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
The `cancelled()`, `success()` and `failure()` expression functions dereferenced `Job.Status` unconditionally, so a nil `Job` context panicked the interpreter — which is why Gitea currently hands the runner a non-nil (empty) `JobContext` as a workaround. This routes all three through a `jobStatus()` helper that treats a nil `Job` as an empty status, keeping existing behaviour identical while removing the panic. Includes a regression test that panics on the old code and passes with the fix.
Related: https://github.com/go-gitea/gitea/pull/38495
Reviewed-on: https://gitea.com/gitea/runner/pulls/1092
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Gitea 1.27 resolves `workflow_call` inputs server-side and sends natively typed JSON values in `github.event.inputs`, so a `type: boolean` input now arrives as a real JSON boolean. The runner coerced booleans by comparing the `any` value against the string `"true"`, which a native bool never matches, so every boolean input evaluated to `false` — including when the callee relied on its default, since the server pre-fills defaults into the event payload and the string fallback is never reached. This accepts a native bool and keeps the string comparison as a fallback, since `workflow_dispatch` inputs are still strings and YAML defaults decode to strings; servers before 1.27 never put a native bool in the payload, so they take the exact same code path as before.
The same coercion is applied to `setupWorkflowInputs` (locally-called reusable workflows, `uses: ./.gitea/workflows/x.yml`), where a `type: boolean` input was previously a native bool when passed as `with: { flag: true }` but a string when interpolated or taken from `default:`. It is now always a bool, matching GitHub, whose `inputs` context "preserves Boolean values as Booleans instead of converting them to strings". **This is potentially breaking**: `inputs.flag == 'true'` now evaluates to `false` and must become `inputs.flag == true`. That pattern is already false on GitHub (a bool compared to a string coerces to `1 == NaN`), but it works on Gitea today, so I am happy to split this hunk into its own PR if you would rather keep this one backport-safe.
Fixes#1082
Reviewed-on: https://gitea.com/gitea/runner/pulls/1087
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Co-authored-by: Nicolas <bircni@icloud.com>
A batch of small, self-contained fixes and docs/example additions.
Fixes#625 - align the example config's `force_pull` with the actual default (`false`)
Fixes#804 - return an error instead of discarding `os.UserHomeDir()` when defaulting `cache.dir`/`host.workdir_parent`
Fixes#571 - send a `gitea-runner/<version>` User-Agent on API requests
Fixes#650 - detect an `Unauthenticated` fetch response and exit the daemon with an error instead of retrying forever
Fixes#766 - add `exec --eventpath` to supply a JSON event payload file
Fixes#256 - add a `bug-report` subcommand that prints version/Go/OS-arch/CPU info
Fixes#617 - add `runner.set_act_env` (default `true`) to optionally omit the `ACT=true` env var
Fixes#635 - record a failure result (and guard a nil reusable-workflow caller) when the job `if`-expression fails to evaluate
Fixes#1005 - remove README docs for config env-var overrides that were already removed from the code
Fixes#448 - clarify in `exec --job` help that `--workflows` may be needed to disambiguate
Fixes#209 - note that `host`-labelled runners still need Docker for `docker://` actions and service containers
Fixes#757 - add a systemd service example with automatic restart
Fixes#474 - add a Kubernetes StatefulSet example that persists the `.runner` registration across reschedules
Fixes#776 - build the basic (non-dind) docker image for `linux/riscv64`
Fixes#628 - build the basic (non-dind) docker image for `linux/s390x`Reviewed-on: https://gitea.com/gitea/runner/pulls/1075
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
GitHub Actions skips services whose image evaluates to an empty string, enabling conditional services via expressions like `image: ${{ matrix.image || '' }}`. Here such a service failed the job at `docker create`. Skip them before container creation, using GitHub's log message verbatim, and add a regression test.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1074
Reviewed-by: Nicolas <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
A composite action's `RunContext` is **seeded with its parent's `Masks` slice** in `newCompositeRunContext`. After the composite finished running, the *entire* `compositeRC.Masks` (the seeded parent masks plus any added during the run) was appended straight back into the parent:
```go
rc.Masks = append(rc.Masks, compositeRC.Masks...)
```
Because the composite already contained the parent's masks, every composite action roughly **doubled** the parent's `Masks` length. With many nested/repeated composite actions this grows exponentially:
```
before=1152 after=2304 compositeLen=1152
before=2304 after=4608 compositeLen=2304
before=4608 after=9216 compositeLen=4608
...
```
### Impact
The bloated `Masks` slice made per-log-line secret redaction (`strings.makeGenericReplacer`, rebuilt per line since #1001) extremely slow. Log writes fell behind enough to keep subprocess pipes open after the process had already exited, surfacing as seemingly random CI failures — most often in later steps of mask-heavy workflows:
```
Error: exec: WaitDelay expired before I/O complete
```
The split of `cmd.Run()` into `cmd.Start()` + `cmd.Wait()` (#883) is what turned this long-standing latency into a hard failure.
### Fix
Only append masks that aren't already present in the parent, via a small `appendUniqueMasks` helper. This keeps the mask count bounded by the number of *distinct* secrets instead of growing exponentially.
I kept `Masks` as a `[]string` rather than switching to a set type, since it's passed by pointer to loggers across `logger.go`, `run_context.go`, and `runner.go` — a type change would be a much wider, riskier refactor. The `O(n·m)` dedup is harmless precisely because `n` now stays small.
Fixes #1054Reviewed-on: https://gitea.com/gitea/runner/pulls/1059
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
When `DEFAULT_ACTIONS_URL=self`, action clone URLs (`uses: owner/repo@ref`) are
built from the Gitea **AppURL** (`gitea_default_actions_url`), but
`shouldCloneURLUseToken` compared the clone URL host only against the runner's
**registered address** (`GitHubInstance`).
When the runner registers with a different hostname than AppURL — same instance,
different DNS (e.g. `gitea.local` vs `gitea.my-nas.lan`, internal vs external) —
the strict `u1.Host == u2.Host` check returns false, so the task token is **not**
attached and the action clone goes out anonymously. Against an instance with
`REQUIRE_SIGNIN_VIEW=true` this fails with:
```
Unable to clone https://gitea.example/owner/action refs/heads/v1: authentication required
```
The current workaround is to make the runner's registered host exactly match
`AppURL`. This PR removes the need for that.
Refs: https://github.com/go-gitea/gitea/issues/27933
## Change
- `shouldCloneURLUseToken` now trusts the clone URL when its host matches **either**
the registered instance (`GitHubInstance`) **or** the self-hosted default-actions
instance (`DefaultActionInstance`). Embedded basic auth is still rejected, and the
empty-host cases are unchanged.
- A new `Config.DefaultActionInstanceIsSelfHosted` flag gates the second candidate.
It is set in the daemon layer (`run/runner.go`, `exec.go`), where `github.com` and
a configured `GithubMirror` are distinguishable, so the token is **never** attached
for off-instance hosts.Reviewed-on: https://gitea.com/gitea/runner/pulls/1056
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
## Summary
A nested `uses:` action inside a **local composite action** fails to clone with a bare `authentication required: Unauthorized` (401) when the instance resolves host-less action references against itself (`DEFAULT_ACTIONS_URL = self`). The job fails at the composite **main→post boundary**, even though the composite's own steps and all post-steps report success.
## Root cause
`newCompositeRunContext` nils `Config.Secrets` so composite steps do not see job secrets. But the action-clone path in `prepareActionExecutor` sourced its token via `getGitCloneToken` → `Config.GetToken()` → `Config.Secrets`, which is empty inside a composite RunContext. The nested action is therefore cloned anonymously → 401 against the authenticated instance.
Two details explain the exact symptom:
- It surfaces at the composite **main→post boundary** because the swallowed nested-step error is re-emitted by `common.JobError` at the end of the composite main pipeline.
- It carries **no clone URL** because, on a warm action cache, only `r.Fetch` runs (not `PlainClone`), and the go-git fetch error is returned verbatim.
## Fix
Source the clone token from `github.Token` instead of `Config.Secrets`. It is preserved across the composite config copy (`Config.Token` / `PresetGitHubContext`) and is identical to `Config.GetToken()` at the top level, so top-level and `act exec` behaviour is unchanged. The `shouldCloneURLUseToken` host gate is kept so the token is never sent to a foreign host. This also aligns the git-clone path with the ActionCache fetch path, which already uses `github.Token`.
Reusable workflows are unaffected — their RunContext keeps `Config.Secrets`.
## Before / after
Local composite action with a nested `uses:` (+ post step), followed by a marker step. Same workflow, same runner host — only the runner fix differs.
| Job step | Before fix | After fix |
| --- | --- | --- |
| `Run actions/checkout@v6` | ✅ success | ✅ success |
| `Run ./.gitea/actions/probe-composite` (nested `uses:` + post) | ❌ **failure** — bare 401 at the main→post boundary | ✅ success |
| `Step after composite` | ⊘ **skipped** | ✅ **ran** |
| Job result | ❌ **failed** | ✅ **succeeded** |
### Before — boundary log
```text
::endgroup::
##[error]authentication required: Unauthorized ← bare 401, no clone target
Run Post ./.gitea/actions/probe-composite
Success - Post ./.gitea/actions/probe-composite ← post steps run and succeed
Success - Post actions/checkout@v6
Job failed
```
The composite's own steps and both post-steps report `Success`; the job fails solely on the bare 401 emitted at the main→post boundary, and the next step is skipped.
### After — boundary log
```text
Run ./.gitea/actions/probe-composite
...composite steps...
Success - ./.gitea/actions/probe-composite
Run Marker after composite
PASSED composite boundary — no 401 (runner fix confirmed)
Success - Marker after composite
Job succeeded
```
## Reproduction
`.gitea/actions/probe-composite/action.yml`:
```yaml
name: probe-composite
runs:
using: composite
steps:
- uses: actions/setup-node@v6 # nested uses → has a post step
with: { node-version: 22 }
- run: echo "composite inner step OK"
shell: bash
```
`.gitea/workflows/probe.yaml`:
```yaml
on: [push]
jobs:
composite-boundary:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: ./.gitea/actions/probe-composite
- run: echo "step after composite" # skipped before the fix; runs after
```
## Verification
- **Before:** bare 401 at the boundary, reproduced with a `delay=0` tail — rules out a token-TTL / expiry effect; it is a missing credential.
- **After:** the composite step succeeds, the step after the composite runs, the job succeeds, and there is no `authentication required` in the log.
- A reusable workflow (`workflow_call`) with the same nested `uses:` + post step never hit the 401, which isolated the bug to the composite main/post path.
## Tests
Adds `TestStepActionRemoteCloneTokenSurvivesNilSecrets` (`act/runner`): asserts the clone token is forwarded when the RunContext mirrors a composite (`Secrets == nil`, token via `Config.Token`), and that the host gate still withholds the token for a foreign host. Verified to fail without the fix and pass with it.
---------
Reviewed-on: https://gitea.com/gitea/runner/pulls/1041
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Co-authored-by: Christian Heim <christian@heimdaheim.de>
Co-committed-by: Christian Heim <christian@heimdaheim.de>
Fixes#1039 — local docker actions referenced with `uses: ./` are incorrectly served from a cached image built for a *different* repository.
## Root cause
For a local docker action, the built image tag is derived from `actionName`, which is the **workspace-relative** path of the action. For a self-referencing action that is always `"./"`, regardless of which repository it lives in. In `execAsDocker` this collapsed to the constant tag `act-dockeraction:latest` for *every* repository.
Before building, the runner checks `ContainerImageExistsLocally` and skips the rebuild if a matching tag is present. On a shared docker daemon, once repository A built its image, repository B's `uses: ./` found the same tag, skipped its rebuild, and **ran repository A's action image** — the reported "the first action is run again, because the path is cached", including the cross-repository concern raised in the issue.
## Fix
Extracted the tag computation into a small, testable `dockerActionImageTag(repository, actionName, localAction)` function. For local actions the tag is now namespaced with the repository (`path.Join(repository, actionName)`):
- different repositories no longer collide (`act-owner-repo-a-dockeraction:latest` vs `act-owner-repo-b-dockeraction:latest`);
- image caching is preserved within a repository (tag stays stable);
- remote actions are unchanged — they already carry a unique, ref-scoped `actionName` (the uses hash).
---------
Co-authored-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1051
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Co-authored-by: bircni <bircni@icloud.com>
Co-committed-by: bircni <bircni@icloud.com>
- fix Windows-only golangci-lint findings in `lp_windows.go` and process killer handle cleanup
- add a `lint-go-windows` target that runs golangci-lint with `GOOS=windows`
- include the Windows lint pass in `make lint` so Ubuntu CI covers `_windows.go` files
Reviewed-on: https://gitea.com/gitea/runner/pulls/1052
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
## Problem
When a workflow run is **cancelled**, the runner diverged from GitHub Actions:
- Main-stage `if: ${{ always() }}` / `if: ${{ cancelled() }}` steps **did not run** at all (unlike a *failed* run, where they do).
- `if: ${{ cancelled() }}` was **structurally impossible** to satisfy — it could never be true.
GitHub Actions runs `always()`/`cancelled()` steps (and post cleanup) even when a job is cancelled. This runner only honored that for action *post* steps (since #1016), leaving main-stage cleanup steps silently skipped.
## Root causes (both in `act/`)
1. **`getJobContext()`** derived the job status purely from step conclusions, so it could only ever return `"success"` or `"failure"`. Since `cancelled()` checks `Job.Status == "cancelled"`, it was impossible — and `success()` stayed *true* on a cancelled run, so the wrong `if` branch was taken everywhere.
2. **The main step pipeline** is chained with `Executor.Then()`, which short-circuits the moment `ctx.Err() != nil`. Once the server cancelled, every not-yet-started main step (including `always()` ones) was abandoned.
## Fix
- Add a per-`RunContext` `jobCancelled` flag + `markCancelled()`. `getJobContext()` now reports `"cancelled"` (taking precedence over success/failure), so `cancelled()`/`always()` are true and `success()`/`failure()` are false — matching GitHub's "only always()/cancelled() run on cancel" semantics.
- Replace the plain main-steps pipeline with `newMainStepsExecutor`. On interruption (`context.Canceled` from a server cancel, or `context.DeadlineExceeded` from the job timeout) it marks the job cancelled and runs the **remaining** steps under a fresh context (`context.WithoutCancel` + bounded timeout) so `always()`/`cancelled()` steps run for cleanup, while default-`success()` steps skip themselves. The original interrupt error is still propagated upward.
- Backstop `markCancelled()` in the post-step `Finally` so cancellations landing outside the main loop still surface the cancelled status to post steps.
Pre-steps keep normal short-circuit behavior, and reporting (`RESULT_CANCELLED`) is untouched — that remains handled by #1016.
## Reporting semantics (unchanged by this PR)
| Run state | failing post/`always()` step reported as |
| --------- | --------------------------------------------------------------------------------------------- |
| Normal | **FAILURE** |
| Timeout | **FAILURE** (deadline path preserves the job-error container) |
| Cancelled | **CANCELLED** — cancellation wins; the failing step is logged but doesn't flip the conclusion |
The new `always()` path runs under `context.WithoutCancel`, so the job-error container is preserved — a failing `always()` step records its failure at step level and does not panic in `SetJobError`.
Fixes#657
Reviewed-on: https://gitea.com/gitea/runner/pulls/1043
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
## Summary
When a workflow references a remote action (e.g. `uses: actions/checkout@v4`) the runner clones that repository during job setup.
Previously this was always a full clone(every branch and the complete history) even though only a single ref is needed.
This PR makes the runner shallow-clone the requested ref by default (`--depth=1`, single branch), falling back to a full clone when a shallow clone fails.
Notes:
- Existing on-disk caches are reused as-is; there is no forced re-clone on upgrade.
## Changes
- A new `runner.action_shallow_clone` option (default `true`) lets operators opt back into full clones.
- `cloneAtDepth`: attempt a shallow clone; fall back to a full clone when shallow clone fails.
- Keep a shallow cache cheap on update: fetch the single requested ref at depth 1 and skip `pull`.
---------
Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1053
Reviewed-by: Nicolas <bircni@icloud.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>
Fixes https://gitea.com/gitea/runner/issues/1047
The runner panics during shutdown when jobs are interrupted:
```
panic in executor: interface conversion: interface {} is nil, not map[string]error
...
gitea.com/gitea/runner/act/common.SetJobError(...)
/data/gitea/runner/act/common/job_error.go:27
gitea.com/gitea/runner/act/runner.reportStepError(...)
/data/gitea/runner/act/runner/job_executor.go:62
```
`SetJobError` did an unchecked type assertion on the job-error container stored in the context:
```go
ctx.Value(jobErrorContextKeyVal).(map[string]error)["error"] = err
```
When the container is absent — e.g. on shutdown, when a job is interrupted and `reportStepError` → `SetJobError` runs on a context where `WithJobErrorContainer` was never called — `ctx.Value` returns `nil` and asserting `nil.(map[string]error)` panics.
## Fix
Use the comma-ok form so a missing container is a no-op, matching the safe pattern already used in the sibling `JobError` function. If there's no container, there's nowhere to record the error anyway, so skipping is correct.
```go
func SetJobError(ctx context.Context, err error) {
if container, ok := ctx.Value(jobErrorContextKeyVal).(map[string]error); ok {
container["error"] = err
}
}
```
Reviewed-on: https://gitea.com/gitea/runner/pulls/1050
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: bircni <bircni@icloud.com>
Co-committed-by: bircni <bircni@icloud.com>
Two `jobs.<job_id>` workflow syntax fields were parsed from YAML but silently ignored. This PR implements both:
- **`jobs.<job_id>.timeout-minutes`** — applies a context deadline around the entire job execution (container start, pre-steps, main steps, post-steps). Mirrors the existing step-level `evaluateStepTimeout`. Supports expression interpolation (e.g. `${{ env.MY_TIMEOUT }}`).
- **`jobs.<job_id>.continue-on-error`** — evaluates the expression when a job fails. If all failing matrix combinations had `continue-on-error: true`, the job does not cause the workflow run to fail (`handleFailure` skips it), and the tolerated failure reports `success` to dependent jobs through the `needs` context so jobs gated on the default `if: success()` still run (matching GitHub). The "any firm failure wins" rule is serialised under the existing per-job lock, so parallel matrix combinations are safe.
Both features follow the same patterns already used at the step level (`evaluateStepTimeout` / `isContinueOnError` in `act/runner/step.go`).
## Version compatibility
These changes are backward compatible. With mismatched versions the feature degrades silently to the previous behaviour (field ignored) — no errors on either side.
- `timeout-minutes`: runner-only, no server dependency.
- `continue-on-error`: requires both this runner PR and the matching Gitea server PR to take full effect. With only one side updated, the field continues to be ignored.
Related: [Github](https://github.com/go-gitea/gitea/pull/38100)
---------
Co-authored-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1032
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
- Adds `runner.post_task_script` and `runner.post_task_script_timeout` (default `5m`) to run a host executable after each task’s built-in cleanup (post-steps, container teardown, bind-workdir removal).
- Stops task heartbeats via `Reporter.StopHeartbeats()` while the script runs so Gitea won’t assign overlapping work; the final task acknowledgement still happens in `reporter.Close()`.
- Script output goes to the runner process log; non-zero exits are warned only and do not change the job result.
- Documents lifecycle, offline behavior, timeouts, and Windows limits (`.ps1` not supported yet) in `docs/post-task-script.md`.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1026
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Interpolate job container.volumes in GetBindsAndMounts(), matching service container volumes and other container fields (image, options).
Fixes expressions like ${{ secrets.MAME }}:/path:ro being passed literally and rejected as invalid bind mounts
Reviewed-on: https://gitea.com/gitea/runner/pulls/1036
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Adds `ssh://` to the list of recognized URL schemes in `newRemoteAction`, so a
step can reference an action over SSH, e.g.:
```yaml
uses: ssh://git@gitea.example.com/actions/checkout@v4
```
Previously only `https://` / `http://` prefixes were parsed; an `ssh://` URL
fell through to the bare `org/repo` parser and failed.
### How auth works
SSH auth is delegated entirely to go-git's defaults — the runner configures no
SSH-specific options:
- **Which key?** go-git falls back to the host's **ssh-agent** (`$SSH_AUTH_SOCK`).
There is no key-file fallback, so the agent must hold a usable key. The SSH
**username** comes from the URL, so use `ssh://git@host/...` (a bare
`ssh://host/...` authenticates as an empty user and most servers reject it).
- **Host key trust?** Established out-of-band via the host's `known_hosts`
(`$SSH_KNOWN_HOSTS`, `~/.ssh/known_hosts`, `/etc/ssh/ssh_known_hosts`). The
runner host must already trust the remote; there is no accept-on-first-use.
- **Host key changes?** The clone fails with a host-key-mismatch error and stays
failed until `known_hosts` is updated on the host. Note `InsecureSkipTLS` does
**not** apply to SSH.
### Caching
The action cache path is derived from `{org}/{repo}` only (scheme/host are not
part of the key), so an `ssh://` action shares cache storage with the same
`org/repo` fetched over HTTP. This is unchanged by this PR and works in practice
(fetches resolve by SHA), but is worth noting.
### Tests
Adds `ssh://` cases to `Test_newRemoteAction` covering the scheme prefix, the
`git@` username placement, and a malformed-URL rejection. The agent/known_hosts
behavior lives in go-git and is not unit-tested here.
Fixes#841
Reviewed-on: https://gitea.com/gitea/runner/pulls/1035
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Cancelling a job on a Linux/macOS host runner can leave the spawned process
tree running and hang the runner — the same failure mode fixed for Windows in
#1011, just on the other platforms.
Steps are launched as process-group leaders (`Setpgid`, or `Setsid` for the PTY
path), but the default `exec.CommandContext` cancellation only kills the
**direct child**. When a step launches a shell that starts a child which in turn
spawns further background processes, cancelling the job leaves the descendants
running. Because those orphans inherited the step's stdout/stderr pipe, the read
end never hits EOF and `cmd.Wait()` blocks forever.
Because the step executor never returns:
- the orphaned processes keep running (the cancelled work is not actually
stopped), and
- end-of-job cleanup is never reached, so the runner appears to go offline / stop
picking up jobs.
## Fix
Apply the same tree-kill approach as Windows, using the Unix counterpart of a Job
Object: the **process group**.
- Add a Unix `processKiller` (`process_unix.go`) that captures the step's PGID
(== PID, since the step is launched as a group leader) and sends `SIGKILL` to
the whole group on cancellation. This also closes the inherited pipe handles so
`cmd.Wait()` can return. `ESRCH` (group already gone) is not treated as an error.
- Restrict the previous no-op stub (`process_other.go`) to `plan9` and have it
fall back to a single-process kill, preserving plan9's prior behaviour.
- Wire `cmd.Cancel` (tree kill) and `cmd.WaitDelay` (10s) **unconditionally** in
`exec()` instead of Windows-only. `WaitDelay` also covers a step that
backgrounds a process holding the pipe open after the main process exits.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1025
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
## Problem
Several runner code paths could drop the **tail** of a step's log output, so a
failing (or cancelled) step would show output that is missing its last line(s).
This was observed in practice and traced to four independent issues.
## Root causes & fixes
### 1. Trailing line without a newline was never flushed
`common.lineWriter` buffers output until it sees a `\n`. A final line **without**
a trailing newline (e.g. an error message printed right before a process exits,
a panic, `printf` without `\n`) stayed in the internal buffer and was never
emitted — the writer exposed no flush at all.
- Added `lineWriter.Flush()` (idempotent), a `Flusher` interface, and a
`FlushWriter(io.Writer)` helper.
- Flush at every stream EOF: the exec copy goroutine, the container `attach()`
streaming goroutine, and at step end (`useStepLogger`).
### 2. Cancellation/timeout truncated output
`waitForCommand` returned immediately on `ctx.Done()` and abandoned the
output-copy goroutine, losing output the command had already produced. It now
drains with a bounded grace period before returning. The response channel is
buffered so the goroutine can't leak if the drain times out.
### 3. `attach()` raced the final bytes
Container output was streamed in a fire-and-forget goroutine that `wait()` did
not synchronize with, so the step could proceed before the last bytes were
written. `wait()` now blocks on the streaming goroutine (bounded) so output is
fully drained and flushed first.
### 4. `::stop-commands::` silently dropped lines from the step log
Lines between `::stop-commands::<token>` and its end token were echoed without
the `raw_output` field **and** short-circuited the handler chain (`return false`),
so they never reached the step log (non-raw entries aren't appended while a step
is running). Now returns `true` so they are still captured.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1028
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
Fixes#1023.
## Problem
In Windows host mode, a single stalled delete syscall (AV/EDR filter driver, unresponsive mount, dying disk) wedged the job forever at `Cleaning up container`. `HostEnvironment.Remove()` bounds every teardown phase (`terminateRunningProcesses`, both `removePathWithRetry` calls) except the `CleanUp` callback — an unbounded `os.RemoveAll(miscpath)` assigned in `startHostEnvironment`. The runner then held its capacity slot indefinitely, the task was reaped as a zombie, and there were no diagnostics.
## Fix
- **Bound the cleanup (availability):** `Remove()` now runs `CleanUp` under `hostCleanupTimeout` (30s) via `runWithTimeout`; on timeout it logs a warning and continues job completion. The stuck goroutine is left to finish (a delete syscall can't be interrupted). Added debug logs around the phase.
- **Reclaim the leak (disk hygiene):** a timed-out cleanup can leave a scratch dir behind, so the existing idle stale-dir sweep is extended to also remove orphaned host-mode scratch dirs (16-hex names) under `Host.WorkdirParent`, leaving the shared `tool_cache` and operator data untouched. The `bind_workdir` gate is dropped from `shouldRunIdleCleanup` so host-mode runners run the sweep.
Reviewed-on: https://gitea.com/gitea/runner/pulls/1024
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
