|
| 1 | +# Copyright (c) ZenML GmbH 2022. All Rights Reserved. |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at: |
| 6 | +# |
| 7 | +# https://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express |
| 12 | +# or implied. See the License for the specific language governing |
| 13 | +# permissions and limitations under the License. |
| 14 | +"""ZenML Step HeartBeat functionality.""" |
| 15 | + |
| 16 | +import _thread |
| 17 | +import logging |
| 18 | +import threading |
| 19 | +import time |
| 20 | +from typing import Annotated |
| 21 | +from uuid import UUID |
| 22 | + |
| 23 | +from pydantic import BaseModel, conint, model_validator |
| 24 | + |
| 25 | +from zenml.enums import ExecutionStatus |
| 26 | + |
| 27 | +logger = logging.getLogger(__name__) |
| 28 | + |
| 29 | + |
| 30 | +class StepHeartBeatTerminationException(Exception): |
| 31 | + """Custom exception class for heartbeat termination.""" |
| 32 | + |
| 33 | + pass |
| 34 | + |
| 35 | + |
| 36 | +class StepHeartBeatOptions(BaseModel): |
| 37 | + """Options group for step heartbeat execution.""" |
| 38 | + |
| 39 | + step_id: UUID |
| 40 | + interval: Annotated[int, conint(ge=10, le=60)] |
| 41 | + name: str | None = None |
| 42 | + |
| 43 | + @model_validator(mode="after") |
| 44 | + def set_default_name(self) -> "StepHeartBeatOptions": |
| 45 | + """Model validator - set name value if missing. |
| 46 | +
|
| 47 | + Returns: |
| 48 | + The validated step heartbeat options. |
| 49 | + """ |
| 50 | + if not self.name: |
| 51 | + self.name = f"HeartBeatWorker-{self.step_id}" |
| 52 | + |
| 53 | + return self |
| 54 | + |
| 55 | + |
| 56 | +class HeartbeatWorker: |
| 57 | + """Worker class implementing heartbeat polling and remote termination.""" |
| 58 | + |
| 59 | + def __init__(self, options: StepHeartBeatOptions): |
| 60 | + """Heartbeat worker constructor. |
| 61 | +
|
| 62 | + Args: |
| 63 | + options: Parameter group - polling interval, step id, etc. |
| 64 | + """ |
| 65 | + self.options = options |
| 66 | + |
| 67 | + self._thread: threading.Thread | None = None |
| 68 | + self._running: bool = False |
| 69 | + self._terminated: bool = ( |
| 70 | + False # one-shot guard to avoid repeated interrupts |
| 71 | + ) |
| 72 | + |
| 73 | + # properties |
| 74 | + |
| 75 | + @property |
| 76 | + def interval(self) -> int: |
| 77 | + """Property function for heartbeat interval. |
| 78 | +
|
| 79 | + Returns: |
| 80 | + The heartbeat polling interval value. |
| 81 | + """ |
| 82 | + return self.options.interval |
| 83 | + |
| 84 | + @property |
| 85 | + def name(self) -> str: |
| 86 | + """Property function for heartbeat worker name. |
| 87 | +
|
| 88 | + Returns: |
| 89 | + The name of the heartbeat worker. |
| 90 | + """ |
| 91 | + return str(self.options.name) |
| 92 | + |
| 93 | + @property |
| 94 | + def step_id(self) -> UUID: |
| 95 | + """Property function for heartbeat worker step ID. |
| 96 | +
|
| 97 | + Returns: |
| 98 | + The id of the step heartbeat is running for. |
| 99 | + """ |
| 100 | + return self.options.step_id |
| 101 | + |
| 102 | + # public functions |
| 103 | + |
| 104 | + def start(self) -> None: |
| 105 | + """Start the heartbeat worker on a background thread.""" |
| 106 | + if self._thread and self._thread.is_alive(): |
| 107 | + logger.info("%s already running; start() is a no-op", self.name) |
| 108 | + return |
| 109 | + |
| 110 | + self._running = True |
| 111 | + self._terminated = False |
| 112 | + self._thread = threading.Thread( |
| 113 | + target=self._run, name=self.name, daemon=True |
| 114 | + ) |
| 115 | + self._thread.start() |
| 116 | + logger.info( |
| 117 | + "Daemon thread %s started (interval=%s)", self.name, self.interval |
| 118 | + ) |
| 119 | + |
| 120 | + def stop(self) -> None: |
| 121 | + """Stops the heartbeat worker.""" |
| 122 | + if not self._running: |
| 123 | + return |
| 124 | + self._running = False |
| 125 | + logger.info("%s stop requested", self.name) |
| 126 | + |
| 127 | + def is_alive(self) -> bool: |
| 128 | + """Liveness of the heartbeat worker thread. |
| 129 | +
|
| 130 | + Returns: |
| 131 | + True if the heartbeat worker thread is alive, False otherwise. |
| 132 | + """ |
| 133 | + t = self._thread |
| 134 | + return bool(t and t.is_alive()) |
| 135 | + |
| 136 | + def _run(self) -> None: |
| 137 | + logger.info("%s run() loop entered", self.name) |
| 138 | + try: |
| 139 | + while self._running: |
| 140 | + try: |
| 141 | + self._heartbeat() |
| 142 | + except StepHeartBeatTerminationException: |
| 143 | + # One-shot: signal the main thread and stop the loop. |
| 144 | + if not self._terminated: |
| 145 | + self._terminated = True |
| 146 | + logger.info( |
| 147 | + "%s received HeartBeatTerminationException; " |
| 148 | + "interrupting main thread", |
| 149 | + self.name, |
| 150 | + ) |
| 151 | + _thread.interrupt_main() # raises KeyboardInterrupt in main thread |
| 152 | + # Ensure we stop our own loop as well. |
| 153 | + self._running = False |
| 154 | + except Exception: |
| 155 | + # Log-and-continue policy for all other errors. |
| 156 | + logger.exception( |
| 157 | + "%s heartbeat() failed; continuing", self.name |
| 158 | + ) |
| 159 | + # Sleep after each attempt (even after errors, unless stopped). |
| 160 | + if self._running: |
| 161 | + time.sleep(self.interval) |
| 162 | + finally: |
| 163 | + logger.info("%s run() loop exiting", self.name) |
| 164 | + |
| 165 | + def _heartbeat(self) -> None: |
| 166 | + from zenml.config.global_config import GlobalConfiguration |
| 167 | + |
| 168 | + store = GlobalConfiguration().zen_store |
| 169 | + |
| 170 | + response = store.update_step_heartbeat(step_run_id=self.step_id) |
| 171 | + |
| 172 | + if response.status in { |
| 173 | + ExecutionStatus.STOPPED, |
| 174 | + ExecutionStatus.STOPPING, |
| 175 | + }: |
| 176 | + raise StepHeartBeatTerminationException( |
| 177 | + f"Step {self.step_id} remotely stopped with status {response.status}." |
| 178 | + ) |
0 commit comments