Compare commits

..
Author SHA1 Message Date
panxiao81 3ec528623a 修复只读根文件系统下的服务账号令牌挂载
test / python (pull_request) Successful in 10s
test / shell (pull_request) Successful in 21s
2026-09-16 15:18:21 +00:00
panxiao81 11669ea81a Merge pull request:统一动态 Runner 命名并记录协议调度路线
test / python (push) Successful in 9s
test / shell (push) Successful in 17s
2026-09-16 14:54:44 +00:00
panxiao81 83eb87bcec 重命名为动态 Runner 并记录协议调度路线
test / python (pull_request) Successful in 9s
test / shell (pull_request) Successful in 15s
2026-09-16 14:51:06 +00:00
panxiao81 75798f4c89 Merge pull request:实现 Pod/VM 动态调度与 Pod Runner 后端
test / python (push) Successful in 8s
test / shell (push) Successful in 15s
2026-09-16 14:28:14 +00:00
16 changed files with 82 additions and 16 deletions
+5
View File
@@ -32,6 +32,11 @@ runs-on: [self-hosted, vm]
消息流使用一个 `WorkQueuePolicy` stream。相同 runner label 的所有 worker 共享同一
durable consumer;扩容只需要增加 worker 或提高单机 capacity。
当前 webhook → NATS 流程是用于尽快验证 Pod/VM 生命周期的 bootstrap 实现,不是
长期调度接口。长期目标是让 controller 作为兼容 Gitea Runner 协议的调度器直接注册、
声明 labels、领取 task,并把已领取 task 交给 Pod/VM executor;路线与迁移边界见
[`docs/runner-protocol-roadmap.md`](docs/runner-protocol-roadmap.md)。
## 开发
```bash
+4 -2
View File
@@ -7,9 +7,11 @@ COPY src ./src
RUN python -m venv /venv && /venv/bin/pip install --no-cache-dir .
FROM python:3.12.11-alpine3.22
RUN addgroup -S -g 65532 runner && adduser -S -D -H -u 65532 -G runner runner
RUN addgroup -S -g 65532 runner \
&& adduser -S -D -H -u 65532 -G runner runner \
&& install -d -o 65532 -g 65532 /var/run/secrets/kubernetes.io/serviceaccount
COPY --from=build /venv /venv
COPY --from=spire /opt/spire/bin/spire-agent /opt/spire/bin/spire-agent
USER 65532:65532
EXPOSE 8787
ENTRYPOINT ["/venv/bin/gitea-microvm-controller"]
ENTRYPOINT ["/venv/bin/gitea-dynamic-runner-controller"]
+59
View File
@@ -0,0 +1,59 @@
# Gitea Runner 协议调度器路线
## 目标
长期形态不依赖 `workflow_job` webhook 发现工作。controller 本身作为 Gitea Runner
协议客户端注册,并声明 `self-hosted`、`pod` 和 `vm` labels;它只在后端存在可用容量
时领取 task,然后将该 task 交给一个一次性 Pod 或 microVM 执行。
```text
Gitea RunnerService
│ Register / Declare / FetchTask
▼
dynamic-runner scheduler
│ 已领取的 task + lease
├── Pod executor
└── microVM executor
│ logs / state / result
└──────────────────────► Gitea
```
这与“收到 webhook 后临时注册另一个 act_runner”不同。`FetchTask` 已经完成任务分配,
不能再期待 Gitea 把同一个 task 分配给随后启动的 runner。协议调度器必须让 executor
执行已经领取的 task,并继续完成日志、状态、心跳、取消和最终结果上报。
## 设计约束
- 对 workflow 的接口保持 `[self-hosted, pod]` 和 `[self-hosted, vm]` 不变。
- scheduler 在没有对应 backend 容量时不领取 task,避免本地形成不可控积压。
- 每个 executor 只执行一个 task,完成后销毁。
- SPIFFE 身份从实际领取的 task 的 repository 和 job name 派生,不需要 queued 与
in-progress webhook 的二阶段关联。
- scheduler 的 runner registration credential 不进入 executor;executor 只得到执行
当前 task 所需的短期 lease/capability。
- task ACK、心跳和结果必须能够跨 scheduler 重启恢复;NATS 可以继续作为内部 handoff,
但不是 Gitea 任务事实来源。
- Pod 与 VM 共享 task/executor 协议,只有环境创建和销毁实现不同。
## 实现顺序
1. 固定当前 Gitea 版本所使用的 RunnerService protobuf 与 act_runner 版本,记录兼容
范围并建立协议契约测试。
2. 实现只注册、Declare labels 和容量感知 FetchTask 的 scheduler spike,暂不执行
task。
3. 从 act_runner 提取或复用 task 执行与日志上报能力,定义 scheduler 到 executor 的
单任务协议。
4. 首先接入 Pod executor,验证成功、失败、取消、超时和 scheduler 重启。
5. 接入 microVM executor,并复用同一 task 协议和身份派生逻辑。
6. 双轨运行并验证后,移除 webhook receiver、临时 runner 注册和 identity binding
subject。
## Bootstrap 实现的退出条件
只有同时满足以下条件才能删除 webhook 路径:
- scheduler 能通过 RunnerService 稳定领取并执行 Pod/VM task;
- Gitea UI 中的实时日志、取消、超时和结论与官方 runner 行为一致;
- scheduler 重启不会丢失已领取 task,也不会重复执行;
- SPIFFE 身份只来自实际领取 task;
- 同一套 workflow 无需修改 `runs-on` 即可从 bootstrap 迁移。
+7 -7
View File
@@ -3,9 +3,9 @@ requires = ["setuptools>=75"]
build-backend = "setuptools.build_meta"
[project]
name = "gitea-microvm-runner"
version = "0.2.0"
description = "On-demand Cloud Hypervisor runners for Gitea Actions"
name = "gitea-dynamic-runner"
version = "0.3.0"
description = "On-demand Pod and microVM execution environments for Gitea Actions"
requires-python = ">=3.11"
dependencies = ["aiohttp==3.12.15", "nats-py==2.11.0"]
@@ -13,10 +13,10 @@ dependencies = ["aiohttp==3.12.15", "nats-py==2.11.0"]
test = ["pytest==8.4.2", "pytest-asyncio==1.2.0"]
[project.scripts]
gitea-microvm-controller = "gitea_microvm_runner.controller:main"
gitea-microvm-worker = "gitea_microvm_runner.worker:cli"
gitea-pod-worker = "gitea_microvm_runner.pod_worker:cli"
gitea-spire-jwt-broker = "gitea_microvm_runner.jwt_broker:main"
gitea-dynamic-runner-controller = "gitea_dynamic_runner.controller:main"
gitea-dynamic-runner-pod-worker = "gitea_dynamic_runner.pod_worker:cli"
gitea-dynamic-runner-vm-worker = "gitea_dynamic_runner.worker:cli"
gitea-spire-jwt-broker = "gitea_dynamic_runner.jwt_broker:main"
[tool.pytest.ini_options]
asyncio_mode = "auto"
+1
View File
@@ -0,0 +1 @@
"""Dynamic Pod and microVM execution environments for Gitea Actions."""
-1
View File
@@ -1 +0,0 @@
"""Gitea microVM runner controller and worker."""
+1 -1
View File
@@ -7,7 +7,7 @@ Requires=microvm-runner-network.service
[Service]
Type=simple
EnvironmentFile=-/etc/microvm-runner/worker.env
ExecStart=/opt/gitea-microvm-runner/venv/bin/gitea-microvm-worker
ExecStart=/opt/gitea-dynamic-runner/venv/bin/gitea-dynamic-runner-vm-worker
Restart=on-failure
RestartSec=5s
SupplementaryGroups=kvm
+2 -2
View File
@@ -4,8 +4,8 @@ import json
import pytest
from gitea_microvm_runner import controller
from gitea_microvm_runner.models import IdentityBinding, RunnerRequest
from gitea_dynamic_runner import controller
from gitea_dynamic_runner.models import IdentityBinding, RunnerRequest
def queued_job(**overrides):
+1 -1
View File
@@ -1,6 +1,6 @@
import pytest
from gitea_microvm_runner.jwt_broker import extract_svid
from gitea_dynamic_runner.jwt_broker import extract_svid
def test_extract_svid_from_spire_json():
+2 -2
View File
@@ -1,7 +1,7 @@
import pytest
from gitea_microvm_runner.models import RunnerRequest
from gitea_microvm_runner import pod_worker
from gitea_dynamic_runner.models import RunnerRequest
from gitea_dynamic_runner import pod_worker
def request() -> RunnerRequest: