refactor: add worker

This commit is contained in:
jialin
2025-11-15 22:37:44 +08:00
parent 864cf76c07
commit 0f3f31ef45
39 changed files with 1746 additions and 606 deletions
+31 -12
View File
@@ -1,14 +1,19 @@
import { clusterSessionAtom } from '@/atoms/clusters';
import DeleteModal from '@/components/delete-modal';
import IconFont from '@/components/icon-font';
import { FilterBar } from '@/components/page-tools';
import useTableFetch from '@/hooks/use-table-fetch';
import PageBox from '@/pages/_components/page-box';
import { queryClusterList } from '@/pages/cluster-management/apis';
import { useIntl, useNavigate } from '@umijs/max';
import { DockerStepsFromWorker } from '@/pages/cluster-management/components/add-worker/config';
import {
ClusterStatusValueMap,
ProviderValueMap
} from '@/pages/cluster-management/config';
import { ClusterListItem } from '@/pages/cluster-management/config/types';
import useAddWorker from '@/pages/cluster-management/hooks/use-add-worker';
import { useIntl } from '@umijs/max';
import { useMemoizedFn } from 'ahooks';
import { Button, ConfigProvider, Table, message } from 'antd';
import { useAtom } from 'jotai';
import React, { useEffect, useState } from 'react';
import NoResult from '../../_components/no-result';
import {
@@ -46,11 +51,9 @@ const Workers: React.FC = () => {
watch: true,
API: WORKERS_API
});
const [, setClusterSession] = useAtom(clusterSessionAtom);
const { MaintenanceModal, handleStopMaintenance, setOpenStatus } =
useWorkerMaintenance({ fetchData });
useWorkerMaintenance({ fetchData: handleSearch });
const navigate = useNavigate();
const intl = useIntl();
const [updateLabelsData, setUpdateLabelsData] = useState<{
open: boolean;
@@ -60,7 +63,7 @@ const Workers: React.FC = () => {
data: {} as ListItem
});
const [clusterData, setClusterData] = useState<{
list: Global.BaseOption<number>[];
list: Global.BaseOption<number, ClusterListItem>[];
data: Record<number, string>;
}>({
list: [],
@@ -73,6 +76,9 @@ const Workers: React.FC = () => {
open: false,
currentData: null
});
const { handleAddWorker, AddWorkerModal, setStepList } = useAddWorker({
clusterList: clusterData.list
});
const getClusterList = async () => {
try {
@@ -89,7 +95,10 @@ const Workers: React.FC = () => {
);
const list = res?.items?.map((item: any) => ({
label: item.name,
value: item.id
value: item.id,
id: item.id,
state: item.state,
provider: item.provider
}));
setClusterData({
list,
@@ -170,9 +179,16 @@ const Workers: React.FC = () => {
}
});
const handleAddWorker = () => {
setClusterSession({ firstAddWorker: true });
navigate('/cluster-management/clusters/list');
const handleOnAddWorker = () => {
let currentData = clusterData.list.find(
(item) =>
item.provider === ProviderValueMap.Docker &&
item.state === ClusterStatusValueMap.Ready
);
if (!currentData) {
currentData = clusterData.list[0];
}
handleAddWorker(currentData as ClusterListItem);
};
const renderEmpty = (type?: string) => {
@@ -188,7 +204,7 @@ const Workers: React.FC = () => {
title={intl.formatMessage({ id: 'noresult.workers.title' })}
subTitle={intl.formatMessage({ id: 'noresult.workers.subTitle' })}
>
<Button type="primary" onClick={handleAddWorker}>
<Button type="primary" onClick={handleOnAddWorker}>
{intl.formatMessage({ id: 'noresult.workers.button.add' })}
</Button>
</NoResult>
@@ -211,6 +227,7 @@ const Workers: React.FC = () => {
useEffect(() => {
getClusterList();
setStepList(DockerStepsFromWorker);
}, []);
return (
@@ -225,6 +242,7 @@ const Workers: React.FC = () => {
handleDeleteByBatch={handleDeleteBatch}
handleSearch={handleSearch}
handleSelectChange={handleClusterChange}
handleClickPrimary={handleOnAddWorker}
handleInputChange={handleNameChange}
rowSelection={rowSelection}
selectOptions={clusterData.list}
@@ -268,6 +286,7 @@ const Workers: React.FC = () => {
}
/>
{MaintenanceModal}
{AddWorkerModal}
</PageBox>
</>
);
+68 -96
View File
@@ -96,7 +96,7 @@ const generateHygonDockerEnvCommand = (config: {
// available for Kubernetes
export const generateKubernetesEnvCommand = (gpu: string) => {
const config = GPUsConfigs[gpu];
const config = GPUsConfigs[gpu] || {};
return `kubectl get runtimeclass ${config.runtime} > /dev/null 2>&1 && echo "${config.label} runtimeclass registered" || (echo "${config.label} runtimeclass issue"; exit 1)`;
};
@@ -126,6 +126,20 @@ export const dockerEnvCommandMap = {
GPUsConfigs[GPUDriverMap.METAX]
)
};
const setNormalArgs = () => {
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\`;
};
const setImageArgs = (params: any) => {
return `${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\`;
};
// avaliable for NVIDIA、AMD、MThreads
const registerWorker = (params: {
@@ -135,18 +149,17 @@ const registerWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
// remove empty enter lines and trailing backslash
return `${commonArgs} \\
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir} \\` : ''}
--runtime ${config.runtime} \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}`;
};
// avaliable for Ascend
@@ -157,20 +170,17 @@ const registerAscendWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--env "ASCEND_VISIBLE_DEVICES=$(npu-smi info -m | tail -n 1 | awk '{print $1}') \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
return `${commonArgs} \\
--env "ASCEND_VISIBLE_DEVICES=$(sudo ls /dev/davinci* | head -1 | grep -o '[0-9]\\+' || echo "0") \\
--runtime ${config.runtime} \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir}` : ''}`;
};
const registerHygonWorker = (params: {
@@ -180,22 +190,19 @@ const registerHygonWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--env ROCM_PATH=/opt/dtk \\
--env ROCM_SMI_LIB_PATH=/opt/hyhal/lib \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
return `${commonArgs} \\
--volume /opt/hyhal:/opt/hyhal:ro \\
--volume /opt/dtk:/opt/dtk:ro \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
--env ROCM_PATH=/opt/dtk \\
--env ROCM_SMI_LIB_PATH=/opt/hyhal/lib \\
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir}` : ''}`;
};
const registerIluvatarWorker = (params: {
@@ -205,22 +212,19 @@ const registerIluvatarWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
return `${commonArgs} \\
--volume /lib/modules:/lib/modules:ro \\
--volume /usr/local/corex:/usr/local/corex:ro \\
--volume /usr/bin/ixsmi:/usr/bin/ixsmi \\
--runtime ${config.runtime} \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir}` : ''}`;
};
const registerMetaXWorker = (params: {
@@ -230,20 +234,17 @@ const registerMetaXWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
return `${commonArgs} \\
--volume /opt/mxdriver:/opt/mxdriver:ro \\
--volume /opt/maca:/opt/maca:ro \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir}` : ''}`;
};
const registerCambriconWorker = (params: {
@@ -253,20 +254,17 @@ const registerCambriconWorker = (params: {
image: string;
gpu: string;
workerIP?: string;
modelDir?: string;
}) => {
const config = GPUsConfigs[params.gpu];
return `sudo docker run -d --name gpustack-worker \\
--restart=unless-stopped \\
--privileged \\
--network=host \\
--volume /var/run/docker.sock:/var/run/docker.sock \\
--volume gpustack-data:/var/lib/gpustack \\
const commonArgs = setNormalArgs();
const imageArgs = setImageArgs(params);
return `${commonArgs} \\
--volume /usr/local/neuware:/usr/local/neuware:ro \\
--volume /usr/bin/cnmon:/usr/bin/cnmon \\
${params.image} \\
--server-url ${params.server} \\
--token ${params.token} \\
${params.workerIP ? `--advertise-address ${params.workerIP}` : ''}`.trim();
${imageArgs}
${params.workerIP ? `--advertise-address ${params.workerIP} \\` : ''}
${params.modelDir ? `--volume ${params.modelDir}:${params.modelDir}` : ''}`;
};
export const registerAddWokerCommandMap = {
@@ -281,41 +279,15 @@ export const registerAddWokerCommandMap = {
};
export const AddWorkerDockerNotes: Record<string, string[]> = {
[GPUDriverMap.NVIDIA]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02'
],
[GPUDriverMap.AMD]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02'
],
[GPUDriverMap.MOORE_THREADS]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02'
],
[GPUDriverMap.ASCEND]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02'
],
[GPUDriverMap.NVIDIA]: [],
[GPUDriverMap.AMD]: [],
[GPUDriverMap.MOORE_THREADS]: [],
[GPUDriverMap.ASCEND]: [],
[GPUDriverMap.HYGON]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02',
'clusters.addworker.hygonNotes',
'clusters.addworker.hygonNotes-02'
],
[GPUDriverMap.ILUVATAR]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02',
'clusters.addworker.corexNotes'
],
[GPUDriverMap.CAMBRICON]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02',
'clusters.addworker.cambriconNotes'
],
[GPUDriverMap.METAX]: [
'clusters.addworker.nvidiaNotes-01',
'clusters.addworker.nvidiaNotes-02',
'clusters.addworker.metaxNotes'
]
[GPUDriverMap.ILUVATAR]: ['clusters.addworker.corexNotes'],
[GPUDriverMap.CAMBRICON]: ['clusters.addworker.cambriconNotes'],
[GPUDriverMap.METAX]: ['clusters.addworker.metaxNotes']
};