skypilot-nightly 1.0.0.dev20251203__py3-none-any.whl → 1.0.0.dev20260112__py3-none-any.whl

sky/schemas/generated/managed_jobsv1_pb2.py

@@ -14,7 +14,7 @@ _sym_db = _symbol_database.Default()
 
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n*sky/schemas/generated/managed_jobsv1.proto\x12\x0fmanaged_jobs.v1\"\x15\n\x06JobIds\x12\x0b\n\x03ids\x18\x01 \x03(\x03\"\x1c\n\nUserHashes\x12\x0e\n\x06hashes\x18\x01 \x03(\t\"\x1c\n\x08Statuses\x12\x10\n\x08statuses\x18\x01 \x03(\t\"\x18\n\x06\x46ields\x12\x0e\n\x06\x66ields\x18\x01 \x03(\t\" \n\nWorkspaces\x12\x12\n\nworkspaces\x18\x01 \x03(\t\"\x13\n\x11GetVersionRequest\"0\n\x12GetVersionResponse\x12\x1a\n\x12\x63ontroller_version\x18\x01 \x01(\t\"\xe1\x04\n\x12GetJobTableRequest\x12\x15\n\rskip_finished\x18\x01 \x01(\x08\x12?\n\x15\x61\x63\x63\x65ssible_workspaces\x18\x02 \x01(\x0b\x32\x1b.managed_jobs.v1.WorkspacesH\x00\x88\x01\x01\x12-\n\x07job_ids\x18\x03 \x01(\x0b\x32\x17.managed_jobs.v1.JobIdsH\x01\x88\x01\x01\x12\x1c\n\x0fworkspace_match\x18\x04 \x01(\tH\x02\x88\x01\x01\x12\x17\n\nname_match\x18\x05 \x01(\tH\x03\x88\x01\x01\x12\x17\n\npool_match\x18\x06 \x01(\tH\x04\x88\x01\x01\x12\x11\n\x04page\x18\x07 \x01(\x05H\x05\x88\x01\x01\x12\x12\n\x05limit\x18\x08 \x01(\x05H\x06\x88\x01\x01\x12\x35\n\x0buser_hashes\x18\t \x01(\x0b\x32\x1b.managed_jobs.v1.UserHashesH\x07\x88\x01\x01\x12\x30\n\x08statuses\x18\n \x01(\x0b\x32\x19.managed_jobs.v1.StatusesH\x08\x88\x01\x01\x12#\n\x1bshow_jobs_without_user_hash\x18\x0b \x01(\x08\x12,\n\x06\x66ields\x18\x0c \x01(\x0b\x32\x17.managed_jobs.v1.FieldsH\t\x88\x01\x01\x42\x18\n\x16_accessible_workspacesB\n\n\x08_job_idsB\x12\n\x10_workspace_matchB\r\n\x0b_name_matchB\r\n\x0b_pool_matchB\x07\n\x05_pageB\x08\n\x06_limitB\x0e\n\x0c_user_hashesB\x0b\n\t_statusesB\t\n\x07_fields\"\xcb\x08\n\x0eManagedJobInfo\x12\x0e\n\x06job_id\x18\x01 \x01(\x03\x12\x0f\n\x07task_id\x18\x02 \x01(\x03\x12\x10\n\x08job_name\x18\x03 \x01(\t\x12\x11\n\ttask_name\x18\x04 \x01(\t\x12\x14\n\x0cjob_duration\x18\x05 \x01(\x01\x12\x16\n\tworkspace\x18\x06 \x01(\tH\x00\x88\x01\x01\x12\x31\n\x06status\x18\x07 \x01(\x0e\x32!.managed_jobs.v1.ManagedJobStatus\x12@\n\x0eschedule_state\x18\x08 \x01(\x0e\x32(.managed_jobs.v1.ManagedJobScheduleState\x12\x11\n\tresources\x18\t \x01(\t\x12\x19\n\x11\x63luster_resources\x18\n \x01(\t\x12\x1e\n\x16\x63luster_resources_full\x18\x0b \x01(\t\x12\r\n\x05\x63loud\x18\x0c \x01(\t\x12\x0e\n\x06region\x18\r \x01(\t\x12\r\n\x05infra\x18\x0e \x01(\t\x12G\n\x0c\x61\x63\x63\x65lerators\x18\x0f \x03(\x0b\x32\x31.managed_jobs.v1.ManagedJobInfo.AcceleratorsEntry\x12\x16\n\x0erecovery_count\x18\x10 \x01(\x05\x12\x14\n\x07\x64\x65tails\x18\x11 \x01(\tH\x01\x88\x01\x01\x12\x1b\n\x0e\x66\x61ilure_reason\x18\x12 \x01(\tH\x02\x88\x01\x01\x12\x16\n\tuser_name\x18\x13 \x01(\tH\x03\x88\x01\x01\x12\x16\n\tuser_hash\x18\x14 \x01(\tH\x04\x88\x01\x01\x12\x19\n\x0csubmitted_at\x18\x15 \x01(\x01H\x05\x88\x01\x01\x12\x15\n\x08start_at\x18\x16 \x01(\x01H\x06\x88\x01\x01\x12\x13\n\x06\x65nd_at\x18\x17 \x01(\x01H\x07\x88\x01\x01\x12\x16\n\tuser_yaml\x18\x18 \x01(\tH\x08\x88\x01\x01\x12\x17\n\nentrypoint\x18\x19 \x01(\tH\t\x88\x01\x01\x12?\n\x08metadata\x18\x1a \x03(\x0b\x32-.managed_jobs.v1.ManagedJobInfo.MetadataEntry\x12\x11\n\x04pool\x18\x1b \x01(\tH\n\x88\x01\x01\x12\x16\n\tpool_hash\x18\x1c \x01(\tH\x0b\x88\x01\x01\x12\x14\n\x07_job_id\x18\x1d \x01(\x03H\x0c\x88\x01\x01\x1a\x33\n\x11\x41\x63\x63\x65leratorsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x1a/\n\rMetadataEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\x42\x0c\n\n_workspaceB\n\n\x08_detailsB\x11\n\x0f_failure_reasonB\x0c\n\n_user_nameB\x0c\n\n_user_hashB\x0f\n\r_submitted_atB\x0b\n\t_start_atB\t\n\x07_end_atB\x0c\n\n_user_yamlB\r\n\x0b_entrypointB\x07\n\x05_poolB\x0c\n\n_pool_hashB\n\n\x08X_job_id\"\xf0\x01\n\x13GetJobTableResponse\x12-\n\x04jobs\x18\x01 \x03(\x0b\x32\x1f.managed_jobs.v1.ManagedJobInfo\x12\r\n\x05total\x18\x02 \x01(\x05\x12\x17\n\x0ftotal_no_filter\x18\x03 \x01(\x05\x12M\n\rstatus_counts\x18\x04 \x03(\x0b\x32\x36.managed_jobs.v1.GetJobTableResponse.StatusCountsEntry\x1a\x33\n\x11StatusCountsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x05:\x02\x38\x01\"?\n\x19GetAllJobIdsByNameRequest\x12\x15\n\x08job_name\x18\x01 \x01(\tH\x00\x88\x01\x01\x42\x0b\n\t_job_name\"-\n\x1aGetAllJobIdsByNameResponse\x12\x0f\n\x07job_ids\x18\x01 \x03(\x03\"\xd7\x01\n\x11\x43\x61ncelJobsRequest\x12\x19\n\x11\x63urrent_workspace\x18\x01 \x01(\t\x12\x16\n\tuser_hash\x18\x02 \x01(\tH\x01\x88\x01\x01\x12\x13\n\tall_users\x18\x03 \x01(\x08H\x00\x12*\n\x07job_ids\x18\x04 \x01(\x0b\x32\x17.managed_jobs.v1.JobIdsH\x00\x12\x12\n\x08job_name\x18\x05 \x01(\tH\x00\x12\x13\n\tpool_name\x18\x06 \x01(\tH\x00\x42\x17\n\x15\x63\x61ncellation_criteriaB\x0c\n\n_user_hash\"%\n\x12\x43\x61ncelJobsResponse\x12\x0f\n\x07message\x18\x01 \x01(\t\"\x97\x01\n\x11StreamLogsRequest\x12\x15\n\x08job_name\x18\x01 \x01(\tH\x00\x88\x01\x01\x12\x13\n\x06job_id\x18\x02 \x01(\x03H\x01\x88\x01\x01\x12\x0e\n\x06\x66ollow\x18\x03 \x01(\x08\x12\x12\n\ncontroller\x18\x04 \x01(\x08\x12\x11\n\x04tail\x18\x05 \x01(\x05H\x02\x88\x01\x01\x42\x0b\n\t_job_nameB\t\n\x07_job_idB\x07\n\x05_tail\"L\n\x12StreamLogsResponse\x12\x10\n\x08log_line\x18\x01 \x01(\t\x12\x16\n\texit_code\x18\x02 \x01(\x05H\x00\x88\x01\x01\x42\x0c\n\n_exit_code*\x85\x04\n\x10ManagedJobStatus\x12\"\n\x1eMANAGED_JOB_STATUS_UNSPECIFIED\x10\x00\x12\x1e\n\x1aMANAGED_JOB_STATUS_PENDING\x10\x01\x12 \n\x1cMANAGED_JOB_STATUS_SUBMITTED\x10\x02\x12\x1f\n\x1bMANAGED_JOB_STATUS_STARTING\x10\x03\x12\x1e\n\x1aMANAGED_JOB_STATUS_RUNNING\x10\x04\x12!\n\x1dMANAGED_JOB_STATUS_RECOVERING\x10\x05\x12!\n\x1dMANAGED_JOB_STATUS_CANCELLING\x10\x06\x12 \n\x1cMANAGED_JOB_STATUS_SUCCEEDED\x10\x07\x12 \n\x1cMANAGED_JOB_STATUS_CANCELLED\x10\x08\x12\x1d\n\x19MANAGED_JOB_STATUS_FAILED\x10\t\x12#\n\x1fMANAGED_JOB_STATUS_FAILED_SETUP\x10\n\x12\'\n#MANAGED_JOB_STATUS_FAILED_PRECHECKS\x10\x0b\x12)\n%MANAGED_JOB_STATUS_FAILED_NO_RESOURCE\x10\x0c\x12(\n$MANAGED_JOB_STATUS_FAILED_CONTROLLER\x10\r*\x9e\x03\n\x17ManagedJobScheduleState\x12*\n&MANAGED_JOB_SCHEDULE_STATE_UNSPECIFIED\x10\x00\x12\x35\n-DEPRECATED_MANAGED_JOB_SCHEDULE_STATE_INVALID\x10\x01\x1a\x02\x08\x01\x12\'\n#MANAGED_JOB_SCHEDULE_STATE_INACTIVE\x10\x02\x12&\n\"MANAGED_JOB_SCHEDULE_STATE_WAITING\x10\x03\x12,\n(MANAGED_JOB_SCHEDULE_STATE_ALIVE_WAITING\x10\x04\x12(\n$MANAGED_JOB_SCHEDULE_STATE_LAUNCHING\x10\x05\x12,\n(MANAGED_JOB_SCHEDULE_STATE_ALIVE_BACKOFF\x10\x06\x12$\n MANAGED_JOB_SCHEDULE_STATE_ALIVE\x10\x07\x12#\n\x1fMANAGED_JOB_SCHEDULE_STATE_DONE\x10\x08\x32\xe4\x03\n\x12ManagedJobsService\x12U\n\nGetVersion\x12\".managed_jobs.v1.GetVersionRequest\x1a#.managed_jobs.v1.GetVersionResponse\x12X\n\x0bGetJobTable\x12#.managed_jobs.v1.GetJobTableRequest\x1a$.managed_jobs.v1.GetJobTableResponse\x12m\n\x12GetAllJobIdsByName\x12*.managed_jobs.v1.GetAllJobIdsByNameRequest\x1a+.managed_jobs.v1.GetAllJobIdsByNameResponse\x12U\n\nCancelJobs\x12\".managed_jobs.v1.CancelJobsRequest\x1a#.managed_jobs.v1.CancelJobsResponse\x12W\n\nStreamLogs\x12\".managed_jobs.v1.StreamLogsRequest\x1a#.managed_jobs.v1.StreamLogsResponse0\x01\x62\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n*sky/schemas/generated/managed_jobsv1.proto\x12\x0fmanaged_jobs.v1\"\x15\n\x06JobIds\x12\x0b\n\x03ids\x18\x01 \x03(\x03\"\x1c\n\nUserHashes\x12\x0e\n\x06hashes\x18\x01 \x03(\t\"\x1c\n\x08Statuses\x12\x10\n\x08statuses\x18\x01 \x03(\t\"\x18\n\x06\x46ields\x12\x0e\n\x06\x66ields\x18\x01 \x03(\t\" \n\nWorkspaces\x12\x12\n\nworkspaces\x18\x01 \x03(\t\"\x13\n\x11GetVersionRequest\"0\n\x12GetVersionResponse\x12\x1a\n\x12\x63ontroller_version\x18\x01 \x01(\t\"\xe1\x04\n\x12GetJobTableRequest\x12\x15\n\rskip_finished\x18\x01 \x01(\x08\x12?\n\x15\x61\x63\x63\x65ssible_workspaces\x18\x02 \x01(\x0b\x32\x1b.managed_jobs.v1.WorkspacesH\x00\x88\x01\x01\x12-\n\x07job_ids\x18\x03 \x01(\x0b\x32\x17.managed_jobs.v1.JobIdsH\x01\x88\x01\x01\x12\x1c\n\x0fworkspace_match\x18\x04 \x01(\tH\x02\x88\x01\x01\x12\x17\n\nname_match\x18\x05 \x01(\tH\x03\x88\x01\x01\x12\x17\n\npool_match\x18\x06 \x01(\tH\x04\x88\x01\x01\x12\x11\n\x04page\x18\x07 \x01(\x05H\x05\x88\x01\x01\x12\x12\n\x05limit\x18\x08 \x01(\x05H\x06\x88\x01\x01\x12\x35\n\x0buser_hashes\x18\t \x01(\x0b\x32\x1b.managed_jobs.v1.UserHashesH\x07\x88\x01\x01\x12\x30\n\x08statuses\x18\n \x01(\x0b\x32\x19.managed_jobs.v1.StatusesH\x08\x88\x01\x01\x12#\n\x1bshow_jobs_without_user_hash\x18\x0b \x01(\x08\x12,\n\x06\x66ields\x18\x0c \x01(\x0b\x32\x17.managed_jobs.v1.FieldsH\t\x88\x01\x01\x42\x18\n\x16_accessible_workspacesB\n\n\x08_job_idsB\x12\n\x10_workspace_matchB\r\n\x0b_name_matchB\r\n\x0b_pool_matchB\x07\n\x05_pageB\x08\n\x06_limitB\x0e\n\x0c_user_hashesB\x0b\n\t_statusesB\t\n\x07_fields\"\xb4\t\n\x0eManagedJobInfo\x12\x0e\n\x06job_id\x18\x01 \x01(\x03\x12\x0f\n\x07task_id\x18\x02 \x01(\x03\x12\x10\n\x08job_name\x18\x03 \x01(\t\x12\x11\n\ttask_name\x18\x04 \x01(\t\x12\x14\n\x0cjob_duration\x18\x05 \x01(\x01\x12\x16\n\tworkspace\x18\x06 \x01(\tH\x00\x88\x01\x01\x12\x31\n\x06status\x18\x07 \x01(\x0e\x32!.managed_jobs.v1.ManagedJobStatus\x12@\n\x0eschedule_state\x18\x08 \x01(\x0e\x32(.managed_jobs.v1.ManagedJobScheduleState\x12\x11\n\tresources\x18\t \x01(\t\x12\x19\n\x11\x63luster_resources\x18\n \x01(\t\x12\x1e\n\x16\x63luster_resources_full\x18\x0b \x01(\t\x12\r\n\x05\x63loud\x18\x0c \x01(\t\x12\x0e\n\x06region\x18\r \x01(\t\x12\r\n\x05infra\x18\x0e \x01(\t\x12G\n\x0c\x61\x63\x63\x65lerators\x18\x0f \x03(\x0b\x32\x31.managed_jobs.v1.ManagedJobInfo.AcceleratorsEntry\x12\x16\n\x0erecovery_count\x18\x10 \x01(\x05\x12\x14\n\x07\x64\x65tails\x18\x11 \x01(\tH\x01\x88\x01\x01\x12\x1b\n\x0e\x66\x61ilure_reason\x18\x12 \x01(\tH\x02\x88\x01\x01\x12\x16\n\tuser_name\x18\x13 \x01(\tH\x03\x88\x01\x01\x12\x16\n\tuser_hash\x18\x14 \x01(\tH\x04\x88\x01\x01\x12\x19\n\x0csubmitted_at\x18\x15 \x01(\x01H\x05\x88\x01\x01\x12\x15\n\x08start_at\x18\x16 \x01(\x01H\x06\x88\x01\x01\x12\x13\n\x06\x65nd_at\x18\x17 \x01(\x01H\x07\x88\x01\x01\x12\x16\n\tuser_yaml\x18\x18 \x01(\tH\x08\x88\x01\x01\x12\x17\n\nentrypoint\x18\x19 \x01(\tH\t\x88\x01\x01\x12?\n\x08metadata\x18\x1a \x03(\x0b\x32-.managed_jobs.v1.ManagedJobInfo.MetadataEntry\x12\x11\n\x04pool\x18\x1b \x01(\tH\n\x88\x01\x01\x12\x16\n\tpool_hash\x18\x1c \x01(\tH\x0b\x88\x01\x01\x12\x14\n\x07_job_id\x18\x1d \x01(\x03H\x0c\x88\x01\x01\x12\x39\n\x05links\x18\x1e \x03(\x0b\x32*.managed_jobs.v1.ManagedJobInfo.LinksEntry\x1a\x33\n\x11\x41\x63\x63\x65leratorsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x1a/\n\rMetadataEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\x1a,\n\nLinksEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\x42\x0c\n\n_workspaceB\n\n\x08_detailsB\x11\n\x0f_failure_reasonB\x0c\n\n_user_nameB\x0c\n\n_user_hashB\x0f\n\r_submitted_atB\x0b\n\t_start_atB\t\n\x07_end_atB\x0c\n\n_user_yamlB\r\n\x0b_entrypointB\x07\n\x05_poolB\x0c\n\n_pool_hashB\n\n\x08X_job_id\"\xf0\x01\n\x13GetJobTableResponse\x12-\n\x04jobs\x18\x01 \x03(\x0b\x32\x1f.managed_jobs.v1.ManagedJobInfo\x12\r\n\x05total\x18\x02 \x01(\x05\x12\x17\n\x0ftotal_no_filter\x18\x03 \x01(\x05\x12M\n\rstatus_counts\x18\x04 \x03(\x0b\x32\x36.managed_jobs.v1.GetJobTableResponse.StatusCountsEntry\x1a\x33\n\x11StatusCountsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x05:\x02\x38\x01\"?\n\x19GetAllJobIdsByNameRequest\x12\x15\n\x08job_name\x18\x01 \x01(\tH\x00\x88\x01\x01\x42\x0b\n\t_job_name\"-\n\x1aGetAllJobIdsByNameResponse\x12\x0f\n\x07job_ids\x18\x01 \x03(\x03\"\xd7\x01\n\x11\x43\x61ncelJobsRequest\x12\x19\n\x11\x63urrent_workspace\x18\x01 \x01(\t\x12\x16\n\tuser_hash\x18\x02 \x01(\tH\x01\x88\x01\x01\x12\x13\n\tall_users\x18\x03 \x01(\x08H\x00\x12*\n\x07job_ids\x18\x04 \x01(\x0b\x32\x17.managed_jobs.v1.JobIdsH\x00\x12\x12\n\x08job_name\x18\x05 \x01(\tH\x00\x12\x13\n\tpool_name\x18\x06 \x01(\tH\x00\x42\x17\n\x15\x63\x61ncellation_criteriaB\x0c\n\n_user_hash\"%\n\x12\x43\x61ncelJobsResponse\x12\x0f\n\x07message\x18\x01 \x01(\t\"\x97\x01\n\x11StreamLogsRequest\x12\x15\n\x08job_name\x18\x01 \x01(\tH\x00\x88\x01\x01\x12\x13\n\x06job_id\x18\x02 \x01(\x03H\x01\x88\x01\x01\x12\x0e\n\x06\x66ollow\x18\x03 \x01(\x08\x12\x12\n\ncontroller\x18\x04 \x01(\x08\x12\x11\n\x04tail\x18\x05 \x01(\x05H\x02\x88\x01\x01\x42\x0b\n\t_job_nameB\t\n\x07_job_idB\x07\n\x05_tail\"L\n\x12StreamLogsResponse\x12\x10\n\x08log_line\x18\x01 \x01(\t\x12\x16\n\texit_code\x18\x02 \x01(\x05H\x00\x88\x01\x01\x42\x0c\n\n_exit_code*\x85\x04\n\x10ManagedJobStatus\x12\"\n\x1eMANAGED_JOB_STATUS_UNSPECIFIED\x10\x00\x12\x1e\n\x1aMANAGED_JOB_STATUS_PENDING\x10\x01\x12 \n\x1cMANAGED_JOB_STATUS_SUBMITTED\x10\x02\x12\x1f\n\x1bMANAGED_JOB_STATUS_STARTING\x10\x03\x12\x1e\n\x1aMANAGED_JOB_STATUS_RUNNING\x10\x04\x12!\n\x1dMANAGED_JOB_STATUS_RECOVERING\x10\x05\x12!\n\x1dMANAGED_JOB_STATUS_CANCELLING\x10\x06\x12 \n\x1cMANAGED_JOB_STATUS_SUCCEEDED\x10\x07\x12 \n\x1cMANAGED_JOB_STATUS_CANCELLED\x10\x08\x12\x1d\n\x19MANAGED_JOB_STATUS_FAILED\x10\t\x12#\n\x1fMANAGED_JOB_STATUS_FAILED_SETUP\x10\n\x12\'\n#MANAGED_JOB_STATUS_FAILED_PRECHECKS\x10\x0b\x12)\n%MANAGED_JOB_STATUS_FAILED_NO_RESOURCE\x10\x0c\x12(\n$MANAGED_JOB_STATUS_FAILED_CONTROLLER\x10\r*\x9e\x03\n\x17ManagedJobScheduleState\x12*\n&MANAGED_JOB_SCHEDULE_STATE_UNSPECIFIED\x10\x00\x12\x35\n-DEPRECATED_MANAGED_JOB_SCHEDULE_STATE_INVALID\x10\x01\x1a\x02\x08\x01\x12\'\n#MANAGED_JOB_SCHEDULE_STATE_INACTIVE\x10\x02\x12&\n\"MANAGED_JOB_SCHEDULE_STATE_WAITING\x10\x03\x12,\n(MANAGED_JOB_SCHEDULE_STATE_ALIVE_WAITING\x10\x04\x12(\n$MANAGED_JOB_SCHEDULE_STATE_LAUNCHING\x10\x05\x12,\n(MANAGED_JOB_SCHEDULE_STATE_ALIVE_BACKOFF\x10\x06\x12$\n MANAGED_JOB_SCHEDULE_STATE_ALIVE\x10\x07\x12#\n\x1fMANAGED_JOB_SCHEDULE_STATE_DONE\x10\x08\x32\xe4\x03\n\x12ManagedJobsService\x12U\n\nGetVersion\x12\".managed_jobs.v1.GetVersionRequest\x1a#.managed_jobs.v1.GetVersionResponse\x12X\n\x0bGetJobTable\x12#.managed_jobs.v1.GetJobTableRequest\x1a$.managed_jobs.v1.GetJobTableResponse\x12m\n\x12GetAllJobIdsByName\x12*.managed_jobs.v1.GetAllJobIdsByNameRequest\x1a+.managed_jobs.v1.GetAllJobIdsByNameResponse\x12U\n\nCancelJobs\x12\".managed_jobs.v1.CancelJobsRequest\x1a#.managed_jobs.v1.CancelJobsResponse\x12W\n\nStreamLogs\x12\".managed_jobs.v1.StreamLogsRequest\x1a#.managed_jobs.v1.StreamLogsResponse0\x01\x62\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -27,12 +27,14 @@ if not _descriptor._USE_C_DESCRIPTORS:
   _globals['_MANAGEDJOBINFO_ACCELERATORSENTRY']._serialized_options = b'8\001'
   _globals['_MANAGEDJOBINFO_METADATAENTRY']._loaded_options = None
   _globals['_MANAGEDJOBINFO_METADATAENTRY']._serialized_options = b'8\001'
+  _globals['_MANAGEDJOBINFO_LINKSENTRY']._loaded_options = None
+  _globals['_MANAGEDJOBINFO_LINKSENTRY']._serialized_options = b'8\001'
   _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._loaded_options = None
   _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._serialized_options = b'8\001'
-  _globals['_MANAGEDJOBSTATUS']._serialized_start=2836
-  _globals['_MANAGEDJOBSTATUS']._serialized_end=3353
-  _globals['_MANAGEDJOBSCHEDULESTATE']._serialized_start=3356
-  _globals['_MANAGEDJOBSCHEDULESTATE']._serialized_end=3770
+  _globals['_MANAGEDJOBSTATUS']._serialized_start=2941
+  _globals['_MANAGEDJOBSTATUS']._serialized_end=3458
+  _globals['_MANAGEDJOBSCHEDULESTATE']._serialized_start=3461
+  _globals['_MANAGEDJOBSCHEDULESTATE']._serialized_end=3875
   _globals['_JOBIDS']._serialized_start=63
   _globals['_JOBIDS']._serialized_end=84
   _globals['_USERHASHES']._serialized_start=86
@@ -50,27 +52,29 @@ if not _descriptor._USE_C_DESCRIPTORS:
   _globals['_GETJOBTABLEREQUEST']._serialized_start=278
   _globals['_GETJOBTABLEREQUEST']._serialized_end=887
   _globals['_MANAGEDJOBINFO']._serialized_start=890
-  _globals['_MANAGEDJOBINFO']._serialized_end=1989
-  _globals['_MANAGEDJOBINFO_ACCELERATORSENTRY']._serialized_start=1711
-  _globals['_MANAGEDJOBINFO_ACCELERATORSENTRY']._serialized_end=1762
-  _globals['_MANAGEDJOBINFO_METADATAENTRY']._serialized_start=1764
-  _globals['_MANAGEDJOBINFO_METADATAENTRY']._serialized_end=1811
-  _globals['_GETJOBTABLERESPONSE']._serialized_start=1992
-  _globals['_GETJOBTABLERESPONSE']._serialized_end=2232
-  _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._serialized_start=2181
-  _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._serialized_end=2232
-  _globals['_GETALLJOBIDSBYNAMEREQUEST']._serialized_start=2234
-  _globals['_GETALLJOBIDSBYNAMEREQUEST']._serialized_end=2297
-  _globals['_GETALLJOBIDSBYNAMERESPONSE']._serialized_start=2299
-  _globals['_GETALLJOBIDSBYNAMERESPONSE']._serialized_end=2344
-  _globals['_CANCELJOBSREQUEST']._serialized_start=2347
-  _globals['_CANCELJOBSREQUEST']._serialized_end=2562
-  _globals['_CANCELJOBSRESPONSE']._serialized_start=2564
-  _globals['_CANCELJOBSRESPONSE']._serialized_end=2601
-  _globals['_STREAMLOGSREQUEST']._serialized_start=2604
-  _globals['_STREAMLOGSREQUEST']._serialized_end=2755
-  _globals['_STREAMLOGSRESPONSE']._serialized_start=2757
-  _globals['_STREAMLOGSRESPONSE']._serialized_end=2833
-  _globals['_MANAGEDJOBSSERVICE']._serialized_start=3773
-  _globals['_MANAGEDJOBSSERVICE']._serialized_end=4257
+  _globals['_MANAGEDJOBINFO']._serialized_end=2094
+  _globals['_MANAGEDJOBINFO_ACCELERATORSENTRY']._serialized_start=1770
+  _globals['_MANAGEDJOBINFO_ACCELERATORSENTRY']._serialized_end=1821
+  _globals['_MANAGEDJOBINFO_METADATAENTRY']._serialized_start=1823
+  _globals['_MANAGEDJOBINFO_METADATAENTRY']._serialized_end=1870
+  _globals['_MANAGEDJOBINFO_LINKSENTRY']._serialized_start=1872
+  _globals['_MANAGEDJOBINFO_LINKSENTRY']._serialized_end=1916
+  _globals['_GETJOBTABLERESPONSE']._serialized_start=2097
+  _globals['_GETJOBTABLERESPONSE']._serialized_end=2337
+  _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._serialized_start=2286
+  _globals['_GETJOBTABLERESPONSE_STATUSCOUNTSENTRY']._serialized_end=2337
+  _globals['_GETALLJOBIDSBYNAMEREQUEST']._serialized_start=2339
+  _globals['_GETALLJOBIDSBYNAMEREQUEST']._serialized_end=2402
+  _globals['_GETALLJOBIDSBYNAMERESPONSE']._serialized_start=2404
+  _globals['_GETALLJOBIDSBYNAMERESPONSE']._serialized_end=2449
+  _globals['_CANCELJOBSREQUEST']._serialized_start=2452
+  _globals['_CANCELJOBSREQUEST']._serialized_end=2667
+  _globals['_CANCELJOBSRESPONSE']._serialized_start=2669
+  _globals['_CANCELJOBSRESPONSE']._serialized_end=2706
+  _globals['_STREAMLOGSREQUEST']._serialized_start=2709
+  _globals['_STREAMLOGSREQUEST']._serialized_end=2860
+  _globals['_STREAMLOGSRESPONSE']._serialized_start=2862
+  _globals['_STREAMLOGSRESPONSE']._serialized_end=2938
+  _globals['_MANAGEDJOBSSERVICE']._serialized_start=3878
+  _globals['_MANAGEDJOBSSERVICE']._serialized_end=4362
 # @@protoc_insertion_point(module_scope)

sky/schemas/generated/managed_jobsv1_pb2.pyi

@@ -127,7 +127,7 @@ class GetJobTableRequest(_message.Message):
     def __init__(self, skip_finished: bool = ..., accessible_workspaces: _Optional[_Union[Workspaces, _Mapping]] = ..., job_ids: _Optional[_Union[JobIds, _Mapping]] = ..., workspace_match: _Optional[str] = ..., name_match: _Optional[str] = ..., pool_match: _Optional[str] = ..., page: _Optional[int] = ..., limit: _Optional[int] = ..., user_hashes: _Optional[_Union[UserHashes, _Mapping]] = ..., statuses: _Optional[_Union[Statuses, _Mapping]] = ..., show_jobs_without_user_hash: bool = ..., fields: _Optional[_Union[Fields, _Mapping]] = ...) -> None: ...
 
 class ManagedJobInfo(_message.Message):
-    __slots__ = ("job_id", "task_id", "job_name", "task_name", "job_duration", "workspace", "status", "schedule_state", "resources", "cluster_resources", "cluster_resources_full", "cloud", "region", "infra", "accelerators", "recovery_count", "details", "failure_reason", "user_name", "user_hash", "submitted_at", "start_at", "end_at", "user_yaml", "entrypoint", "metadata", "pool", "pool_hash", "_job_id")
+    __slots__ = ("job_id", "task_id", "job_name", "task_name", "job_duration", "workspace", "status", "schedule_state", "resources", "cluster_resources", "cluster_resources_full", "cloud", "region", "infra", "accelerators", "recovery_count", "details", "failure_reason", "user_name", "user_hash", "submitted_at", "start_at", "end_at", "user_yaml", "entrypoint", "metadata", "pool", "pool_hash", "_job_id", "links")
     class AcceleratorsEntry(_message.Message):
         __slots__ = ("key", "value")
         KEY_FIELD_NUMBER: _ClassVar[int]
@@ -142,6 +142,13 @@ class ManagedJobInfo(_message.Message):
         key: str
         value: str
         def __init__(self, key: _Optional[str] = ..., value: _Optional[str] = ...) -> None: ...
+    class LinksEntry(_message.Message):
+        __slots__ = ("key", "value")
+        KEY_FIELD_NUMBER: _ClassVar[int]
+        VALUE_FIELD_NUMBER: _ClassVar[int]
+        key: str
+        value: str
+        def __init__(self, key: _Optional[str] = ..., value: _Optional[str] = ...) -> None: ...
     JOB_ID_FIELD_NUMBER: _ClassVar[int]
     TASK_ID_FIELD_NUMBER: _ClassVar[int]
     JOB_NAME_FIELD_NUMBER: _ClassVar[int]
@@ -171,6 +178,7 @@ class ManagedJobInfo(_message.Message):
     POOL_FIELD_NUMBER: _ClassVar[int]
     POOL_HASH_FIELD_NUMBER: _ClassVar[int]
     _JOB_ID_FIELD_NUMBER: _ClassVar[int]
+    LINKS_FIELD_NUMBER: _ClassVar[int]
     job_id: int
     task_id: int
     job_name: str
@@ -200,7 +208,8 @@ class ManagedJobInfo(_message.Message):
     pool: str
     pool_hash: str
     _job_id: int
-    def __init__(self, job_id: _Optional[int] = ..., task_id: _Optional[int] = ..., job_name: _Optional[str] = ..., task_name: _Optional[str] = ..., job_duration: _Optional[float] = ..., workspace: _Optional[str] = ..., status: _Optional[_Union[ManagedJobStatus, str]] = ..., schedule_state: _Optional[_Union[ManagedJobScheduleState, str]] = ..., resources: _Optional[str] = ..., cluster_resources: _Optional[str] = ..., cluster_resources_full: _Optional[str] = ..., cloud: _Optional[str] = ..., region: _Optional[str] = ..., infra: _Optional[str] = ..., accelerators: _Optional[_Mapping[str, float]] = ..., recovery_count: _Optional[int] = ..., details: _Optional[str] = ..., failure_reason: _Optional[str] = ..., user_name: _Optional[str] = ..., user_hash: _Optional[str] = ..., submitted_at: _Optional[float] = ..., start_at: _Optional[float] = ..., end_at: _Optional[float] = ..., user_yaml: _Optional[str] = ..., entrypoint: _Optional[str] = ..., metadata: _Optional[_Mapping[str, str]] = ..., pool: _Optional[str] = ..., pool_hash: _Optional[str] = ..., _job_id: _Optional[int] = ...) -> None: ...
+    links: _containers.ScalarMap[str, str]
+    def __init__(self, job_id: _Optional[int] = ..., task_id: _Optional[int] = ..., job_name: _Optional[str] = ..., task_name: _Optional[str] = ..., job_duration: _Optional[float] = ..., workspace: _Optional[str] = ..., status: _Optional[_Union[ManagedJobStatus, str]] = ..., schedule_state: _Optional[_Union[ManagedJobScheduleState, str]] = ..., resources: _Optional[str] = ..., cluster_resources: _Optional[str] = ..., cluster_resources_full: _Optional[str] = ..., cloud: _Optional[str] = ..., region: _Optional[str] = ..., infra: _Optional[str] = ..., accelerators: _Optional[_Mapping[str, float]] = ..., recovery_count: _Optional[int] = ..., details: _Optional[str] = ..., failure_reason: _Optional[str] = ..., user_name: _Optional[str] = ..., user_hash: _Optional[str] = ..., submitted_at: _Optional[float] = ..., start_at: _Optional[float] = ..., end_at: _Optional[float] = ..., user_yaml: _Optional[str] = ..., entrypoint: _Optional[str] = ..., metadata: _Optional[_Mapping[str, str]] = ..., pool: _Optional[str] = ..., pool_hash: _Optional[str] = ..., _job_id: _Optional[int] = ..., links: _Optional[_Mapping[str, str]] = ...) -> None: ...
 
 class GetJobTableResponse(_message.Message):
     __slots__ = ("jobs", "total", "total_no_filter", "status_counts")

sky/serve/serve_utils.py

@@ -23,6 +23,7 @@ import filelock
 from sky import backends
 from sky import exceptions
 from sky import global_user_state
+from sky import resources as resources_lib
 from sky import sky_logging
 from sky import skypilot_config
 from sky.adaptors import common as adaptors_common
@@ -350,6 +351,13 @@ def validate_service_task(task: 'sky.Task', pool: bool) -> None:
                              f'file does not match the pool argument. '
                              f'To fix, add a valid `{field_name}` field.')
 
+    # Validate that pools do not use ordered resources
+    if pool and isinstance(task.resources, list):
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(
+                'Ordered resources are not supported for pools. '
+                'Use `any_of` instead, or specify a single resource.')
+
     policy_description = ('on-demand'
                           if task.service.dynamic_ondemand_fallback else 'spot')
     for resource in list(task.resources):
@@ -360,22 +368,6 @@ def validate_service_task(task: 'sky.Task', pool: bool) -> None:
                                  f'{sys_name} will replenish preempted spot '
                                  f'with {policy_description} instances.')
 
-    if pool:
-        accelerators = set()
-        for resource in task.resources:
-            if resource.accelerators is not None:
-                if isinstance(resource.accelerators, str):
-                    accelerators.add(resource.accelerators)
-                elif isinstance(resource.accelerators, dict):
-                    accelerators.update(resource.accelerators.keys())
-                elif isinstance(resource.accelerators, list):
-                    accelerators.update(resource.accelerators)
-        if len(accelerators) > 1:
-            with ux_utils.print_exception_no_traceback():
-                raise ValueError('Heterogeneous clusters are not supported for '
-                                 'pools please specify one accelerator '
-                                 'for all workers.')
-
     # Try to create a spot placer from the task yaml. Check if the task yaml
     # is valid for spot placer.
     spot_placer.SpotPlacer.from_task(task.service, task)
@@ -730,7 +722,7 @@ def _get_service_status(
             for replica_info in record['replica_info']:
                 job_ids = managed_job_state.get_nonterminal_job_ids_by_pool(
                     service_name, replica_info['name'])
-                replica_info['used_by'] = job_ids[0] if job_ids else None
+                replica_info['used_by'] = job_ids
     return record
 
 
@@ -810,16 +802,112 @@ def get_ready_replicas(
     ]
 
 
-def get_next_cluster_name(service_name: str, job_id: int) -> Optional[str]:
-    """Get the next available cluster name from idle replicas.
+def _task_fits(task_resources: 'resources_lib.Resources',
+               free_resources: 'resources_lib.Resources') -> bool:
+    """Check if the task resources fit in the free resources."""
+    if not task_resources.less_demanding_than(free_resources,
+                                              check_cloud=False):
+        return False
+    if task_resources.cpus is not None:
+        if (free_resources.cpus is None or
+                task_resources.cpus > free_resources.cpus):
+            return False
+    if task_resources.memory is not None:
+        if (free_resources.memory is None or
+                task_resources.memory > free_resources.memory):
+            return False
+    return True
+
+
+def _is_empty_resource(resource: 'resources_lib.Resources') -> bool:
+    # Returns True if this resource object does not specify any resources.
+    return (resource.cpus is None and resource.memory is None and
+            resource.accelerators is None)
+
+
+def get_free_worker_resources(
+        pool: str) -> Optional[Dict[str, Optional[resources_lib.Resources]]]:
+    """Get free resources for each worker in a pool.
+
+    Args:
+        pool: Pool name (service name)
+
+    Returns:
+        Dictionary mapping cluster_name (worker) to free Resources object (or
+        None if worker is not available or has no free resources).
+    """
+
+    free_resources: Dict[str, Optional[resources_lib.Resources]] = {}
+    replicas = serve_state.get_replica_infos(pool)
+
+    for replica_info in replicas:
+        cluster_name = replica_info.cluster_name
+
+        # Get cluster handle
+        handle = replica_info.handle()
+        if handle is None or handle.launched_resources is None:
+            free_resources[cluster_name] = None
+            continue
+
+        total_resources = handle.launched_resources
+
+        # Get job IDs running on this worker
+        job_ids = managed_job_state.get_nonterminal_job_ids_by_pool(
+            pool, cluster_name)
+
+        if len(job_ids) == 0:
+            free_resources[cluster_name] = total_resources
+            continue
+
+        # Get used resources
+        # TODO(lloyd): We should batch the database calls here so that we
+        # make a single call to get all the used resources for all the jobs.
+        used_resources = managed_job_state.get_pool_worker_used_resources(
+            set(job_ids))
+        if used_resources is None:
+            # We failed to get the used resources. We should return None since
+            # we can't make any guarantees about what resources are being used.
+            logger.warning(
+                f'Failed to get used resources for cluster {cluster_name!r}')
+            return None
+
+        if _is_empty_resource(used_resources):
+            # We encountered a job that has no resources specified. We
+            # will not consider it for resource-aware scheduling so it must
+            # be scheduled on its own. To do this we will set the free
+            # worker resources to nothing by returning an empty resource
+            # object.
+            logger.debug(f'Job {job_ids} has no resources specified. '
+                         'Skipping resource-aware scheduling for cluster '
+                         f'{cluster_name!r}')
+            free_resources[cluster_name] = resources_lib.Resources()
+        else:
+            # Calculate free resources using - operator
+            free = total_resources - used_resources
+            free_resources[cluster_name] = free
+
+    return free_resources
+
+
+def get_next_cluster_name(
+    service_name: str,
+    job_id: int,
+    task_resources: Optional[typing.Union[
+        'resources_lib.Resources', typing.Set['resources_lib.Resources'],
+        typing.List['resources_lib.Resources']]] = None
+) -> Optional[str]:
+    """Get the next available cluster name from replicas with sufficient
+    resources.
 
     Args:
         service_name: The name of the service.
-        job_id: Optional job ID to associate with the acquired cluster.
-                If None, a placeholder will be used.
+        job_id: Job ID to associate with the acquired cluster.
+        task_resources: Optional task resource requirements. If provided, will
+                check if resources fit in free worker resources. Can be
+                a single Resources object or a set/list of Resources objects.
 
     Returns:
-        The cluster name if an idle replica is found, None otherwise.
+        The cluster name if a suitable replica is found, None otherwise.
     """
     # Check if service exists
     service_status = _get_service_status(service_name,
@@ -831,36 +919,126 @@ def get_next_cluster_name(service_name: str, job_id: int) -> Optional[str]:
     if not service_status['pool']:
         logger.error(f'Service {service_name!r} is not a pool.')
         return None
+
     with filelock.FileLock(get_service_filelock_path(service_name)):
+        free_resources = get_free_worker_resources(service_name)
+        logger.debug(f'Free resources: {free_resources!r}')
         logger.debug(f'Get next cluster name for pool {service_name!r}')
         ready_replicas = get_ready_replicas(service_name)
+
+        logger.debug(f'Ready replicas: {ready_replicas!r}')
+
         idle_replicas: List['replica_managers.ReplicaInfo'] = []
-        for replica_info in ready_replicas:
-            jobs_on_replica = managed_job_state.get_nonterminal_job_ids_by_pool(
-                service_name, replica_info.cluster_name)
-            # TODO(tian): Make it resources aware. Currently we allow and only
-            # allow one job per replica. In the following PR, we should:
-            #  i) When the replica is launched with `any_of` resources (
-            #     replicas can have different resources), we should check if
-            #     the resources that jobs require are available on the replica.
-            #     e.g., if a job requires A100:1 on a {L4:1, A100:1} pool, it
-            #     should only goes to replica with A100.
-            # ii) When a job only requires a subset of the resources on the
-            #     replica, each replica should be able to handle multiple jobs
-            #     at the same time. e.g., if a job requires A100:1 on a A100:8
-            #     pool, it should be able to run 4 jobs at the same time.
-            if not jobs_on_replica:
-                idle_replicas.append(replica_info)
+
+        # If task_resources is provided, use resource-aware scheduling
+        # Normalize task_resources to a list
+        if isinstance(task_resources, resources_lib.Resources):
+            task_resources_list = [task_resources]
+        elif isinstance(task_resources, (set, list)):
+            task_resources_list = list(task_resources)
+        else:
+            task_resources_list = []
+
+        # We should do resource aware scheduling if:
+        # 1. There are task resources.
+        # 2. The first task resource has some resources listed.
+        # 3. There are free resources.
+        # 4. Any free resource has some resources listed.
+        resource_aware = len(task_resources_list) > 0
+        resource_aware = (resource_aware and
+                          not _is_empty_resource(task_resources_list[0]))
+        resource_aware = resource_aware and free_resources is not None
+        if free_resources is not None:
+            for free_resource in free_resources.values():
+                if free_resource is not None and not _is_empty_resource(
+                        free_resource):
+                    resource_aware = True
+                    break
+            else:
+                resource_aware = False
+        else:
+            resource_aware = False
+
+        if resource_aware:
+            logger.debug('Doing resource aware scheduling')
+            for replica_info in ready_replicas:
+                cluster_name = replica_info.cluster_name
+                assert free_resources is not None
+                free_resources_on_worker = free_resources.get(cluster_name)
+                logger.debug(f'Free resources for cluster {cluster_name!r}: '
+                             f'{free_resources_on_worker!r}')
+
+                # Skip if worker has no free resources available
+                if free_resources_on_worker is None:
+                    logger.debug(f'Worker {cluster_name!r} has no free '
+                                 'resources')
+                    continue
+
+                # Check if any of the task resource options fit
+                fits = False
+                for task_res in task_resources_list:
+                    logger.debug(f'Task resources: {task_res!r}')
+                    if _task_fits(task_res, free_resources_on_worker):
+                        logger.debug(f'Task resources {task_res!r} fits'
+                                     ' in free resources '
+                                     f'{free_resources_on_worker!r}')
+                        fits = True
+                        break
+                    else:
+                        logger.debug(f'Task resources {task_res!r} does not fit'
+                                     ' in free resources '
+                                     f'{free_resources_on_worker!r}')
+                if fits:
+                    idle_replicas.append(replica_info)
+        # Also fall back to resource unaware scheduling if no idle replicas are
+        # found. This might be because our launched resources were improperly
+        # set. If that's the case then jobs will fail to schedule in a resource
+        # aware way because one of the resources will be `None` so we can just
+        # fallback to 1 job per replica. If we are truly resource bottlenecked
+        # then we will see that there are jobs running on the replica and will
+        # not schedule another.
+        if len(idle_replicas) == 0:
+            logger.debug('Falling back to resource unaware scheduling')
+            # Fall back to resource unaware scheduling if no task resources
+            # are provided.
+            for replica_info in ready_replicas:
+                jobs_on_replica = (
+                    managed_job_state.get_nonterminal_job_ids_by_pool(
+                        service_name, replica_info.cluster_name))
+                if not jobs_on_replica:
+                    idle_replicas.append(replica_info)
+
         if not idle_replicas:
             logger.info(f'No idle replicas found for pool {service_name!r}')
             return None
 
         # Select the first idle replica.
-        # TODO(tian): "Load balancing" policy.
         replica_info = idle_replicas[0]
         logger.info(f'Selected replica {replica_info.replica_id} with cluster '
                     f'{replica_info.cluster_name!r} for job {job_id!r} in pool '
                     f'{service_name!r}')
+
+        # If job has heterogeneous resources (any_of/ordered), update
+        # full_resources to the specific resource that was selected for this
+        # worker. This must happen before releasing the filelock to ensure
+        # atomicity with the scheduling decision.
+        if resource_aware and len(task_resources_list) > 1:
+            assert free_resources is not None
+            free_resources_on_worker = free_resources.get(
+                replica_info.cluster_name)
+            if free_resources_on_worker is not None:
+                # Find which task resource fits on this worker
+                for task_res in task_resources_list:
+                    if _task_fits(task_res, free_resources_on_worker):
+                        # Update full_resources in database to this specific
+                        # resource
+                        logger.debug(
+                            f'Updating full_resources for job {job_id!r} '
+                            f'to selected resource: {task_res!r}')
+                        managed_job_state.update_job_full_resources(
+                            job_id, task_res.to_yaml_config())
+                        break
+
         managed_job_state.set_current_cluster_name(job_id,
                                                    replica_info.cluster_name)
         return replica_info.cluster_name
@@ -1541,7 +1719,21 @@ def _format_replica_table(replica_records: List[Dict[str, Any]], show_all: bool,
         replica_status = record['status']
         status_str = replica_status.colored_str()
         used_by = record.get('used_by', None)
-        used_by_str = str(used_by) if used_by is not None else '-'
+        if used_by is None:
+            used_by_str = '-'
+        elif isinstance(used_by, str):
+            used_by_str = used_by
+        else:
+            if len(used_by) > 2:
+                used_by_str = (
+                    f'{used_by[0]}, {used_by[1]}, +{len(used_by) - 2}'
+                    ' more')
+            elif len(used_by) == 2:
+                used_by_str = f'{used_by[0]}, {used_by[1]}'
+            elif len(used_by) == 1:
+                used_by_str = str(used_by[0])
+            else:
+                used_by_str = '-'
 
         replica_handle: Optional['backends.CloudVmRayResourceHandle'] = record[
             'handle']

sky/serve/server/impl.py

@@ -517,7 +517,7 @@ def update(
                     f'{workers} is not supported. Ignoring the update.')
 
         # Load the existing task configuration from the service's YAML file
-        yaml_content = service_record['yaml_content']
+        yaml_content = service_record['pool_yaml']
 
         # Load the existing task configuration
         task = task_lib.Task.from_yaml_str(yaml_content)

sky/server/common.py

@@ -17,6 +17,7 @@ import time
 import typing
 from typing import (Any, Callable, cast, Dict, Generic, Literal, Optional,
                     Tuple, TypeVar, Union)
+from urllib.request import Request
 import uuid
 
 import cachetools
@@ -147,6 +148,22 @@ def get_api_cookie_jar() -> requests.cookies.RequestsCookieJar:
     return cookie_jar
 
 
+def get_cookie_header_for_url(url: str) -> Dict[str, str]:
+    """Extract Cookie header value from a cookie jar for a specific URL"""
+    cookies = get_api_cookie_jar()
+    if not cookies:
+        return {}
+
+    # Use urllib Request to do URL-aware cookie filtering
+    request = Request(url)
+    cookies.add_cookie_header(request)
+    cookie_header = request.get_header('Cookie')
+
+    if cookie_header is None:
+        return {}
+    return {'Cookie': cookie_header}
+
+
 def set_api_cookie_jar(cookie_jar: CookieJar,
                        create_if_not_exists: bool = True) -> None:
     """Updates the file cookie jar with the given cookie jar."""

sky/server/constants.py

@@ -10,7 +10,7 @@ from sky.skylet import constants
 # based on version info is needed.
 # For more details and code guidelines, refer to:
 # https://docs.skypilot.co/en/latest/developers/CONTRIBUTING.html#backward-compatibility-guidelines
-API_VERSION = 24
+API_VERSION = 26
 
 # The minimum peer API version that the code should still work with.
 # Notes (dev):

sky/server/metrics.py

@@ -48,10 +48,12 @@ async def gpu_metrics() -> fastapi.Response:
     all_metrics: List[str] = []
     successful_contexts = 0
 
+    remote_contexts = [
+        context for context in contexts if context != 'in-cluster'
+    ]
     tasks = [
         asyncio.create_task(metrics_utils.get_metrics_for_context(context))
-        for context in contexts
-        if context != 'in-cluster'
+        for context in remote_contexts
     ]
 
     results = await asyncio.gather(*tasks, return_exceptions=True)
@@ -59,7 +61,8 @@ async def gpu_metrics() -> fastapi.Response:
     for i, result in enumerate(results):
         if isinstance(result, Exception):
             logger.error(
-                f'Failed to get metrics for context {contexts[i]}: {result}')
+                f'Failed to get metrics for context {remote_contexts[i]}: '
+                f'{result}')
         elif isinstance(result, BaseException):
             # Avoid changing behavior for non-Exception BaseExceptions
             # like KeyboardInterrupt/SystemExit: re-raise them.