|
| 1 | +"""CLI commands for kernel scheduling history.""" |
| 2 | + |
| 3 | +from __future__ import annotations |
| 4 | + |
| 5 | +import asyncio |
| 6 | +from typing import TYPE_CHECKING |
| 7 | +from uuid import UUID |
| 8 | + |
| 9 | +import click |
| 10 | + |
| 11 | +from ai.backend.client.cli.v2.helpers import ( |
| 12 | + create_v2_registry, |
| 13 | + load_v2_config, |
| 14 | + parse_order_options, |
| 15 | + print_result, |
| 16 | +) |
| 17 | + |
| 18 | +if TYPE_CHECKING: |
| 19 | + from ai.backend.common.dto.manager.v2.scheduling_history.request import KernelHistoryFilter |
| 20 | + |
| 21 | +# Shared result choices for scheduling history filters |
| 22 | +_RESULT_CHOICES = click.Choice( |
| 23 | + ["SUCCESS", "FAILURE", "STALE", "NEED_RETRY", "EXPIRED", "GIVE_UP", "SKIPPED"], |
| 24 | + case_sensitive=False, |
| 25 | +) |
| 26 | + |
| 27 | + |
| 28 | +def _build_kernel_history_filter( |
| 29 | + kernel_id: str | None, |
| 30 | + session_id: str | None, |
| 31 | + phase: str | None, |
| 32 | + from_status: tuple[str, ...], |
| 33 | + to_status: tuple[str, ...], |
| 34 | + result: str | None, |
| 35 | + error_code: str | None, |
| 36 | + message: str | None, |
| 37 | +) -> KernelHistoryFilter | None: |
| 38 | + """Build a KernelHistoryFilter from explicit CLI options. |
| 39 | +
|
| 40 | + Returns None if no filter options were provided. |
| 41 | + """ |
| 42 | + from ai.backend.common.dto.manager.query import StringFilter, UUIDFilter |
| 43 | + from ai.backend.common.dto.manager.v2.scheduling_history.request import ( |
| 44 | + KernelHistoryFilter, |
| 45 | + SchedulingResultFilter, |
| 46 | + ) |
| 47 | + from ai.backend.common.dto.manager.v2.scheduling_history.types import SchedulingResultType |
| 48 | + |
| 49 | + has_any = any( |
| 50 | + opt is not None for opt in (kernel_id, session_id, phase, result, error_code, message) |
| 51 | + ) |
| 52 | + if not has_any and not from_status and not to_status: |
| 53 | + return None |
| 54 | + |
| 55 | + return KernelHistoryFilter( |
| 56 | + kernel_id=UUIDFilter(equals=UUID(kernel_id)) if kernel_id is not None else None, |
| 57 | + session_id=UUIDFilter(equals=UUID(session_id)) if session_id is not None else None, |
| 58 | + phase=StringFilter(contains=phase) if phase is not None else None, |
| 59 | + from_status=list(from_status) if from_status else None, |
| 60 | + to_status=list(to_status) if to_status else None, |
| 61 | + result=( |
| 62 | + SchedulingResultFilter(equals=SchedulingResultType(result)) |
| 63 | + if result is not None |
| 64 | + else None |
| 65 | + ), |
| 66 | + error_code=StringFilter(contains=error_code) if error_code is not None else None, |
| 67 | + message=StringFilter(contains=message) if message is not None else None, |
| 68 | + ) |
| 69 | + |
| 70 | + |
| 71 | +@click.group() |
| 72 | +def kernel() -> None: |
| 73 | + """Kernel scheduling history commands.""" |
| 74 | + |
| 75 | + |
| 76 | +@kernel.command() |
| 77 | +@click.option("--limit", type=int, default=None, help="Maximum items to return.") |
| 78 | +@click.option("--offset", type=int, default=None, help="Number of items to skip.") |
| 79 | +@click.option("--kernel-id", type=str, default=None, help="Filter by kernel ID (UUID).") |
| 80 | +@click.option("--session-id", type=str, default=None, help="Filter by session ID (UUID).") |
| 81 | +@click.option("--phase", type=str, default=None, help="Filter by scheduling phase (contains).") |
| 82 | +@click.option( |
| 83 | + "--from-status", |
| 84 | + type=str, |
| 85 | + multiple=True, |
| 86 | + help="Filter by from_status values (repeatable).", |
| 87 | +) |
| 88 | +@click.option( |
| 89 | + "--to-status", |
| 90 | + type=str, |
| 91 | + multiple=True, |
| 92 | + help="Filter by to_status values (repeatable).", |
| 93 | +) |
| 94 | +@click.option("--result", type=_RESULT_CHOICES, default=None, help="Filter by scheduling result.") |
| 95 | +@click.option("--error-code", type=str, default=None, help="Filter by error code (contains).") |
| 96 | +@click.option("--message", type=str, default=None, help="Filter by message (contains).") |
| 97 | +@click.option( |
| 98 | + "--order-by", |
| 99 | + multiple=True, |
| 100 | + help="Order by field:direction (e.g., created_at:desc). Fields: created_at, updated_at.", |
| 101 | +) |
| 102 | +def search( |
| 103 | + limit: int | None, |
| 104 | + offset: int | None, |
| 105 | + kernel_id: str | None, |
| 106 | + session_id: str | None, |
| 107 | + phase: str | None, |
| 108 | + from_status: tuple[str, ...], |
| 109 | + to_status: tuple[str, ...], |
| 110 | + result: str | None, |
| 111 | + error_code: str | None, |
| 112 | + message: str | None, |
| 113 | + order_by: tuple[str, ...], |
| 114 | +) -> None: |
| 115 | + """Search kernel scheduling histories (superadmin only).""" |
| 116 | + from ai.backend.common.dto.manager.v2.scheduling_history.request import ( |
| 117 | + AdminSearchKernelHistoriesInput, |
| 118 | + KernelHistoryOrder, |
| 119 | + ) |
| 120 | + from ai.backend.common.dto.manager.v2.scheduling_history.types import KernelHistoryOrderField |
| 121 | + |
| 122 | + history_filter = _build_kernel_history_filter( |
| 123 | + kernel_id, session_id, phase, from_status, to_status, result, error_code, message |
| 124 | + ) |
| 125 | + |
| 126 | + orders = ( |
| 127 | + parse_order_options(order_by, KernelHistoryOrderField, KernelHistoryOrder) |
| 128 | + if order_by |
| 129 | + else None |
| 130 | + ) |
| 131 | + |
| 132 | + async def _run() -> None: |
| 133 | + registry = await create_v2_registry(load_v2_config()) |
| 134 | + try: |
| 135 | + result_data = await registry.scheduling_history.search_kernel_history( |
| 136 | + AdminSearchKernelHistoriesInput( |
| 137 | + filter=history_filter, |
| 138 | + order=orders, |
| 139 | + limit=limit, |
| 140 | + offset=offset, |
| 141 | + ), |
| 142 | + ) |
| 143 | + print_result(result_data) |
| 144 | + finally: |
| 145 | + await registry.close() |
| 146 | + |
| 147 | + asyncio.run(_run()) |
| 148 | + |
| 149 | + |
| 150 | +@kernel.command(name="search-scoped") |
| 151 | +@click.option("--session-id", type=str, default=None, help="Scope to the kernels of this session.") |
| 152 | +@click.option("--kernel-id", type=str, default=None, help="Scope to this kernel.") |
| 153 | +@click.option("--limit", type=int, default=None, help="Maximum items to return.") |
| 154 | +@click.option("--offset", type=int, default=None, help="Number of items to skip.") |
| 155 | +@click.option("--phase", type=str, default=None, help="Filter by scheduling phase (contains).") |
| 156 | +@click.option( |
| 157 | + "--from-status", |
| 158 | + type=str, |
| 159 | + multiple=True, |
| 160 | + help="Filter by from_status values (repeatable).", |
| 161 | +) |
| 162 | +@click.option( |
| 163 | + "--to-status", |
| 164 | + type=str, |
| 165 | + multiple=True, |
| 166 | + help="Filter by to_status values (repeatable).", |
| 167 | +) |
| 168 | +@click.option("--result", type=_RESULT_CHOICES, default=None, help="Filter by scheduling result.") |
| 169 | +@click.option("--error-code", type=str, default=None, help="Filter by error code (contains).") |
| 170 | +@click.option("--message", type=str, default=None, help="Filter by message (contains).") |
| 171 | +@click.option( |
| 172 | + "--order-by", |
| 173 | + multiple=True, |
| 174 | + help="Order by field:direction (e.g., created_at:desc). Fields: created_at, updated_at.", |
| 175 | +) |
| 176 | +def search_scoped( |
| 177 | + session_id: str | None, |
| 178 | + kernel_id: str | None, |
| 179 | + limit: int | None, |
| 180 | + offset: int | None, |
| 181 | + phase: str | None, |
| 182 | + from_status: tuple[str, ...], |
| 183 | + to_status: tuple[str, ...], |
| 184 | + result: str | None, |
| 185 | + error_code: str | None, |
| 186 | + message: str | None, |
| 187 | + order_by: tuple[str, ...], |
| 188 | +) -> None: |
| 189 | + """Search kernel scheduling history within a session and/or kernel scope. |
| 190 | +
|
| 191 | + The scope takes two axes rather than one positional ID, so it is given as options. |
| 192 | + At least one of --session-id / --kernel-id is required. |
| 193 | + """ |
| 194 | + from ai.backend.common.dto.manager.v2.scheduling_history.request import ( |
| 195 | + KernelHistoryOrder, |
| 196 | + ScopedSearchKernelHistoriesInput, |
| 197 | + ) |
| 198 | + from ai.backend.common.dto.manager.v2.scheduling_history.types import ( |
| 199 | + KernelHistoryOrderField, |
| 200 | + KernelHistoryScopeDTO, |
| 201 | + ) |
| 202 | + |
| 203 | + if session_id is None and kernel_id is None: |
| 204 | + raise click.UsageError("At least one of --session-id or --kernel-id is required.") |
| 205 | + |
| 206 | + scope = KernelHistoryScopeDTO( |
| 207 | + session_id=UUID(session_id) if session_id is not None else None, |
| 208 | + kernel_id=UUID(kernel_id) if kernel_id is not None else None, |
| 209 | + ) |
| 210 | + |
| 211 | + # The scope already narrows by id, so ids are not repeated in the filter. |
| 212 | + history_filter = _build_kernel_history_filter( |
| 213 | + None, None, phase, from_status, to_status, result, error_code, message |
| 214 | + ) |
| 215 | + |
| 216 | + orders = ( |
| 217 | + parse_order_options(order_by, KernelHistoryOrderField, KernelHistoryOrder) |
| 218 | + if order_by |
| 219 | + else None |
| 220 | + ) |
| 221 | + |
| 222 | + async def _run() -> None: |
| 223 | + registry = await create_v2_registry(load_v2_config()) |
| 224 | + try: |
| 225 | + result_data = await registry.scheduling_history.kernel_scoped_search( |
| 226 | + ScopedSearchKernelHistoriesInput( |
| 227 | + scope=scope, |
| 228 | + filter=history_filter, |
| 229 | + order=orders, |
| 230 | + limit=limit, |
| 231 | + offset=offset, |
| 232 | + ), |
| 233 | + ) |
| 234 | + print_result(result_data) |
| 235 | + finally: |
| 236 | + await registry.close() |
| 237 | + |
| 238 | + asyncio.run(_run()) |
0 commit comments