|
| 1 | +import { useSuspendedBackendaiClient } from '.'; |
| 2 | +import { useSetBAINotification } from './useBAINotification'; |
| 3 | +import { |
| 4 | + useCurrentProjectValue, |
| 5 | + useCurrentResourceGroupState, |
| 6 | +} from './useCurrentProject'; |
| 7 | +import { toGlobalId } from 'backend.ai-ui'; |
| 8 | +import _ from 'lodash'; |
| 9 | +import { useTranslation } from 'react-i18next'; |
| 10 | +import { fetchQuery, graphql, useRelayEnvironment } from 'react-relay'; |
| 11 | +import { useStartSessionCreationQuery } from 'src/__generated__/useStartSessionCreationQuery.graphql'; |
| 12 | +import { transformPortValuesToNumbers } from 'src/components/PortSelectFormItem'; |
| 13 | +import { RESOURCE_ALLOCATION_INITIAL_FORM_VALUES } from 'src/components/SessionFormItems/ResourceAllocationFormItems'; |
| 14 | +import { generateRandomString } from 'src/helper'; |
| 15 | +import { |
| 16 | + SessionLauncherFormValue, |
| 17 | + SessionResources, |
| 18 | +} from 'src/pages/SessionLauncherPage'; |
| 19 | + |
| 20 | +// Type for successful session creation result |
| 21 | +type SessionCreationSuccess = { |
| 22 | + kernelId?: string; |
| 23 | + sessionId: string; |
| 24 | + sessionName: string; |
| 25 | + servicePorts: Array<{ name: string }>; |
| 26 | +}; |
| 27 | +interface CreateSessionInfo { |
| 28 | + kernelName: string; |
| 29 | + sessionName: string; |
| 30 | + architecture: string; |
| 31 | + batchTimeout?: string; |
| 32 | + resources: SessionResources; |
| 33 | +} |
| 34 | + |
| 35 | +export const SESSION_LAUNCHER_NOTI_PREFIX = 'session-launcher:'; |
| 36 | + |
| 37 | +export const useStartSession = () => { |
| 38 | + 'use memo'; |
| 39 | + |
| 40 | + const { t } = useTranslation(); |
| 41 | + const currentProject = useCurrentProjectValue(); |
| 42 | + const { upsertNotification } = useSetBAINotification(); |
| 43 | + |
| 44 | + const relayEnv = useRelayEnvironment(); |
| 45 | + const baiClient = useSuspendedBackendaiClient(); |
| 46 | + const supportsMountById = baiClient.supports('mount-by-id'); |
| 47 | + const supportBatchTimeout = baiClient?.supports('batch-timeout') ?? false; |
| 48 | + |
| 49 | + const [currentGlobalResourceGroup] = useCurrentResourceGroupState(); |
| 50 | + |
| 51 | + const defaultFormValues: DeepPartial<SessionLauncherFormValue> = { |
| 52 | + sessionType: 'interactive', |
| 53 | + // If you set `allocationPreset` to 'custom', `allocationPreset` is not changed automatically any more. |
| 54 | + allocationPreset: 'auto-select', |
| 55 | + hpcOptimization: { |
| 56 | + autoEnabled: true, |
| 57 | + }, |
| 58 | + batch: { |
| 59 | + enabled: false, |
| 60 | + command: undefined, |
| 61 | + scheduleDate: undefined, |
| 62 | + ...(supportBatchTimeout && { |
| 63 | + timeoutEnabled: false, |
| 64 | + timeout: undefined, |
| 65 | + timeoutUnit: 's', |
| 66 | + }), |
| 67 | + }, |
| 68 | + envvars: [], |
| 69 | + // set default_session_environment only if set |
| 70 | + ...(baiClient._config?.default_session_environment && { |
| 71 | + environments: { |
| 72 | + environment: baiClient._config?.default_session_environment, |
| 73 | + }, |
| 74 | + }), |
| 75 | + ...RESOURCE_ALLOCATION_INITIAL_FORM_VALUES, |
| 76 | + resourceGroup: currentGlobalResourceGroup || undefined, |
| 77 | + }; |
| 78 | + |
| 79 | + const startSession = async (values: SessionLauncherFormValue) => { |
| 80 | + // If manual image is selected, use it as kernelName |
| 81 | + const imageFullName = |
| 82 | + values.environments.manual || values.environments.version; |
| 83 | + const [kernelName, architecture] = imageFullName |
| 84 | + ? imageFullName.split('@') |
| 85 | + : ['', '']; |
| 86 | + |
| 87 | + const sessionName = _.isEmpty(values.sessionName) |
| 88 | + ? generateSessionId() |
| 89 | + : values.sessionName; |
| 90 | + |
| 91 | + const sessionInfo: CreateSessionInfo = { |
| 92 | + // Basic session information |
| 93 | + sessionName: sessionName, |
| 94 | + kernelName, |
| 95 | + architecture, |
| 96 | + resources: { |
| 97 | + enqueueOnly: true, |
| 98 | + // Project and domain settings |
| 99 | + group_name: values.owner?.enabled |
| 100 | + ? values.owner.project |
| 101 | + : currentProject.name, |
| 102 | + domain: values.owner?.enabled |
| 103 | + ? values.owner.domainName |
| 104 | + : baiClient._config.domainName, |
| 105 | + |
| 106 | + // Session configuration |
| 107 | + type: values.sessionType, |
| 108 | + cluster_mode: values.cluster_mode, |
| 109 | + cluster_size: values.cluster_size, |
| 110 | + maxWaitSeconds: 15, |
| 111 | + |
| 112 | + // Owner settings (optional) |
| 113 | + // FYI, `config.scaling_group` also changes based on owner settings |
| 114 | + ...(values.owner?.enabled |
| 115 | + ? { |
| 116 | + owner_access_key: values.owner.accesskey, |
| 117 | + } |
| 118 | + : {}), |
| 119 | + |
| 120 | + // Batch mode settings (optional) |
| 121 | + ...(values.sessionType === 'batch' |
| 122 | + ? { |
| 123 | + starts_at: values.batch.enabled |
| 124 | + ? values.batch.scheduleDate |
| 125 | + : undefined, |
| 126 | + startupCommand: values.batch.command, |
| 127 | + } |
| 128 | + : {}), |
| 129 | + |
| 130 | + // Bootstrap script (optional) |
| 131 | + ...(values.bootstrap_script |
| 132 | + ? { bootstrap_script: values.bootstrap_script } |
| 133 | + : {}), |
| 134 | + |
| 135 | + // Batch timeout configuration (optional) |
| 136 | + ...(supportBatchTimeout && |
| 137 | + values?.batch?.timeoutEnabled && |
| 138 | + !_.isUndefined(values?.batch?.timeout) |
| 139 | + ? { |
| 140 | + batchTimeout: |
| 141 | + _.toString(values.batch.timeout) + values?.batch?.timeoutUnit, |
| 142 | + } |
| 143 | + : undefined), |
| 144 | + |
| 145 | + config: { |
| 146 | + // Resource allocation |
| 147 | + resources: { |
| 148 | + cpu: values.resource.cpu, |
| 149 | + mem: values.resource.mem, |
| 150 | + // Add accelerator only if specified |
| 151 | + ...(values.resource.accelerator > 0 |
| 152 | + ? { |
| 153 | + [values.resource.acceleratorType]: |
| 154 | + values.resource.accelerator, |
| 155 | + } |
| 156 | + : undefined), |
| 157 | + }, |
| 158 | + scaling_group: values.owner?.enabled |
| 159 | + ? values.owner.project |
| 160 | + : values.resourceGroup, |
| 161 | + resource_opts: { |
| 162 | + shmem: values.resource.shmem, |
| 163 | + // allow_fractional_resource_fragmentation can be added here if needed |
| 164 | + }, |
| 165 | + |
| 166 | + // Storage configuration |
| 167 | + [supportsMountById ? 'mount_ids' : 'mounts']: values.mount_ids, |
| 168 | + [supportsMountById ? 'mount_id_map' : 'mount_map']: |
| 169 | + values.mount_id_map, |
| 170 | + |
| 171 | + // Environment variables |
| 172 | + environ: { |
| 173 | + ..._.fromPairs(values.envvars.map((v) => [v.variable, v.value])), |
| 174 | + // set hpcOptimization options: "OMP_NUM_THREADS", "OPENBLAS_NUM_THREADS" |
| 175 | + ...(values.hpcOptimization.autoEnabled |
| 176 | + ? {} |
| 177 | + : _.omit(values.hpcOptimization, 'autoEnabled')), |
| 178 | + }, |
| 179 | + |
| 180 | + // Networking |
| 181 | + preopen_ports: transformPortValuesToNumbers(values.ports), |
| 182 | + |
| 183 | + // Agent selection (optional) |
| 184 | + ...(baiClient.supports('agent-select') && |
| 185 | + !baiClient?._config?.hideAgents && |
| 186 | + values.agent !== 'auto' |
| 187 | + ? { |
| 188 | + // Filter out undefined values |
| 189 | + agent_list: [values.agent].filter( |
| 190 | + (agent): agent is string => !!agent, |
| 191 | + ), |
| 192 | + } |
| 193 | + : undefined), |
| 194 | + }, |
| 195 | + }, |
| 196 | + }; |
| 197 | + const sessionPromises = _.map(_.range(values.num_of_sessions || 1), (i) => { |
| 198 | + const formattedSessionName = |
| 199 | + (values.num_of_sessions || 1) > 1 |
| 200 | + ? `${sessionInfo.sessionName}-${generateRandomString()}-${i}` |
| 201 | + : sessionInfo.sessionName; |
| 202 | + return baiClient |
| 203 | + .createIfNotExists( |
| 204 | + sessionInfo.kernelName, |
| 205 | + formattedSessionName, |
| 206 | + sessionInfo.resources, |
| 207 | + undefined, |
| 208 | + sessionInfo.architecture, |
| 209 | + ) |
| 210 | + .then((res: { created: boolean; status: string }) => { |
| 211 | + // // When session is already created with the same name, the status code |
| 212 | + // // is 200, but the response body has 'created' field as false. For better |
| 213 | + // // user experience, we show the notification message. |
| 214 | + if (!res?.created) { |
| 215 | + // message.warning(t('session.launcher.SessionAlreadyExists')); |
| 216 | + throw new Error(t('session.launcher.SessionAlreadyExists')); |
| 217 | + } |
| 218 | + if (res?.status === 'CANCELLED') { |
| 219 | + // Case about failed to start new session kind of "docker image not found" or etc. |
| 220 | + throw new Error(t('session.launcher.FailedToStartNewSession')); |
| 221 | + } |
| 222 | + return res; |
| 223 | + }) |
| 224 | + .catch((err: any) => { |
| 225 | + if (err?.message?.includes('The session already exists')) { |
| 226 | + throw new Error(t('session.launcher.SessionAlreadyExists')); |
| 227 | + } else { |
| 228 | + throw err; |
| 229 | + } |
| 230 | + }); |
| 231 | + }); |
| 232 | + |
| 233 | + return Promise.allSettled(sessionPromises).then( |
| 234 | + async ( |
| 235 | + sessionCreations: PromiseSettledResult<SessionCreationSuccess>[], |
| 236 | + ) => { |
| 237 | + // Group session creations by their status |
| 238 | + const results = _.groupBy(sessionCreations, 'status') as { |
| 239 | + fulfilled?: PromiseFulfilledResult<SessionCreationSuccess>[]; |
| 240 | + rejected?: PromiseRejectedResult[]; |
| 241 | + }; |
| 242 | + |
| 243 | + return results; |
| 244 | + }, |
| 245 | + ); |
| 246 | + }; |
| 247 | + |
| 248 | + const upsertSessionNotification = async ( |
| 249 | + successCreations: PromiseFulfilledResult<SessionCreationSuccess>[], |
| 250 | + upsertNotificationOverrides?: Parameters<typeof upsertNotification>, |
| 251 | + ) => { |
| 252 | + const promises = _.map(successCreations, (creation) => { |
| 253 | + const session = creation.value; |
| 254 | + return fetchQuery<useStartSessionCreationQuery>( |
| 255 | + relayEnv, |
| 256 | + graphql` |
| 257 | + query useStartSessionCreationQuery($id: GlobalIDField!) { |
| 258 | + compute_session_node(id: $id) { |
| 259 | + ...BAINodeNotificationItemFragment |
| 260 | + } |
| 261 | + } |
| 262 | + `, |
| 263 | + { |
| 264 | + id: toGlobalId('ComputeSessionNode', session.sessionId), |
| 265 | + }, |
| 266 | + ) |
| 267 | + .toPromise() |
| 268 | + .then((queryResult) => { |
| 269 | + const createdSession = queryResult?.compute_session_node ?? null; |
| 270 | + |
| 271 | + if (createdSession) { |
| 272 | + upsertNotification( |
| 273 | + { |
| 274 | + key: `${SESSION_LAUNCHER_NOTI_PREFIX}${session.sessionId}`, |
| 275 | + node: createdSession, |
| 276 | + open: true, |
| 277 | + duration: 0, |
| 278 | + ...upsertNotificationOverrides?.[0], |
| 279 | + }, |
| 280 | + upsertNotificationOverrides?.[1], |
| 281 | + ); |
| 282 | + } |
| 283 | + }); |
| 284 | + }); |
| 285 | + return Promise.allSettled(promises); |
| 286 | + }; |
| 287 | + |
| 288 | + return { startSession, defaultFormValues, upsertSessionNotification }; |
| 289 | +}; |
| 290 | + |
| 291 | +const generateSessionId = () => { |
| 292 | + let text = ''; |
| 293 | + const possible = |
| 294 | + 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789'; |
| 295 | + for (let i = 0; i < 8; i++) { |
| 296 | + text += possible.charAt(Math.floor(Math.random() * possible.length)); |
| 297 | + } |
| 298 | + return text + '-session'; |
| 299 | +}; |
0 commit comments