Import tasks getting stuck mid execution
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 25/100
- Issue 类型
- 缺陷
- 描述清晰度
- 需要澄清
- 活跃度
- 停滞
- 技术栈
- aws, go, rabbitmq
调研方向
从 task/runner.go 中的 10 分钟超时开始，然后跟踪 import task 的进度报告、管道传输和 S3 上传路径。复现或添加 instrument，以调查报告中的卡住任务案例，包括大型和常规 asset 示例。完成的标准是确定原因，并且 import task 不再保持无响应，也不会在 RabbitMQ redelivery 后被重复执行。
由索引模型根据 Issue 内容生成。
描述
We have some weird cases of tasks getting stuck mid execution. These tasks just keep getting
re-executed, since the RabbitMQ will eventually timeout (after 30m) waiting on an ACK and just
terminates the connection with the client (nacking all messages it had in-flight). The task is then
re-executed as if nothing happened [1].
[1] This is also another bug that we should address. Should just fail if the task had already tried
running before and just disappeared, which we can already tell from the metadata in the API. This
is not the root cause though so we still need to investigate and fix the stuck tasks.
No logs that indicate what is wrong, but I have a light suspicion on either:
- the "progress reporting" logic
- the "piping" logic in the import task which sends a stream both to
ffprobeand to the storage - the S3 upload client
On the first tasks I found this error, they were actually importing large stream recordings which take
12+ minutes to download on a good connection, due to the on-demand MP4 generation bottleneck.
It was already weird since we have a hard timeout of 10 minutes so the task runner
should have just failed the task, instead of gone silent.
Right now I just found an even weirder case though. It was from a regular "import" task, which is not
importing a recording but actually just another asset as a test that the user was making. This is the
task:
{
"id": "51ea2a1e-618e-452d-a024-7c5a0ace266f",
"type": "import",
"params": {
"import": {
"url": "https://livepeercdn.com/asset/REDACTED/video"
}
},
"status": {
"phase": "running",
"progress": 0.649,
"updatedAt": 1651269956139
},
"userId": "REDACTED",
"createdAt": 1650886712179,
"outputAssetId": "4582de3b-ead3-4ffe-8b6d-b130f61290a1"
}
The asset has around 5GB and takes less than a minute to download from a good connection, so there's
no clear reason of why the task-runner is getting stuck.
- 主要语言
- Go
- 星标
- 3
- 派生
- 1
- PR 合并指标
- 30 天内没有已合并 PR
环境准备
- 提供 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 没有贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
livepeer/task-runner 的其他 Issue
-
Out of bounds panics when processing catalyst callback messages可能重新可做 @gioelecerati 于 1353 天前认领，目前没有进行中的 PR。 未关闭
livepeer/task-runner#136 · 已指派 1 人 ·
-
"error processing ffprobe output: no video track found in file" for audio only input可能重新可做 @red-0ne 于 1381 天前认领，目前没有进行中的 PR。 未关闭
livepeer/task-runner#121 · 已指派 1 人 ·
-
catalyst task lost (last update 1m58.390845005s ago) after retrying for an Acceleration fault可能重新可做 @thomshutt 于 1402 天前认领，目前没有进行中的 PR。 未关闭
livepeer/task-runner#115 · 3 条评论 · 已指派 1 人 ·
-
难度 4/5 3-5 天 新手友好度 35/100
livepeer/task-runner#108 ·
-
难度 5/5 一周以上 新手友好度 35/100
livepeer/task-runner#105 ·
查看 livepeer/task-runner 的全部 Issue
相似的 Issue
-
Realtime: ASR EOU language tags leak into the LLM conversation (unintended reply-language switching)可能已有人在做 @nexxtmobile 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 70/100
维护者通常 1 天内回复
-
effort:high model:sonnet phase:3 type:docs
难度 2/5 1-3 小时 新手友好度 68/100
-
难度 2/5 1-3 小时 新手友好度 72/100
OffchainLabs/nitro#4771 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 82/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 65/100
BurntSushi/toml#530 ·