indexify 0.2.42__tar.gz → 0.2.44__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {indexify-0.2.42 → indexify-0.2.44}/PKG-INFO +3 -3
- {indexify-0.2.42 → indexify-0.2.44}/indexify/cli.py +3 -3
- indexify-0.2.44/indexify/executor/executor.py +155 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_worker.py +2 -4
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/task_fetcher.py +1 -2
- indexify-0.2.44/indexify/executor/task_reporter.py +215 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/graph.py +0 -2
- {indexify-0.2.42 → indexify-0.2.44}/pyproject.toml +3 -3
- indexify-0.2.42/indexify/executor/agent.py +0 -262
- indexify-0.2.42/indexify/executor/executor_tasks.py +0 -58
- indexify-0.2.42/indexify/executor/task_reporter.py +0 -155
- indexify-0.2.42/indexify/executor/task_store.py +0 -132
- indexify-0.2.42/indexify/functions_sdk/local_cache.py +0 -46
- {indexify-0.2.42 → indexify-0.2.44}/LICENSE.txt +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/README.md +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/__init__.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/common_util.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/data_loaders/__init__.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/data_loaders/local_directory_loader.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/data_loaders/url_loader.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/error.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/api_objects.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/downloader.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor_factory.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor_map.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/process_function_executor.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/process_function_executor_factory.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/executor/runtime_probes.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/function_executor_service.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/handlers/run_function/function_inputs_loader.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/handlers/run_function/handler.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/handlers/run_function/request_validator.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/handlers/run_function/response_helper.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/initialize_request_validator.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/configuration.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor.proto +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2.pyi +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2_grpc.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/message_validator.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/server.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/data_objects.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/graph_definition.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/graph_validation.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/image.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/indexify_functions.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/object_serializer.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/functions_sdk/pipeline.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/http_client.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/logging.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/remote_graph.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/remote_pipeline.py +0 -0
- {indexify-0.2.42 → indexify-0.2.44}/indexify/settings.py +0 -0
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.1
|
2
2
|
Name: indexify
|
3
|
-
Version: 0.2.
|
3
|
+
Version: 0.2.44
|
4
4
|
Summary: Python Client for Indexify
|
5
5
|
Home-page: https://github.com/tensorlakeai/indexify
|
6
6
|
License: Apache 2.0
|
@@ -21,11 +21,11 @@ Requires-Dist: grpcio-tools (==1.68.1)
|
|
21
21
|
Requires-Dist: httpx-sse (>=0.4.0,<0.5.0)
|
22
22
|
Requires-Dist: httpx[http2] (==0.27.2)
|
23
23
|
Requires-Dist: nanoid (>=2.0.0,<3.0.0)
|
24
|
-
Requires-Dist: pydantic (
|
24
|
+
Requires-Dist: pydantic (>=2.10.2,<3.0.0)
|
25
25
|
Requires-Dist: pyyaml (>=6,<7)
|
26
26
|
Requires-Dist: rich (>=13.9.2,<14.0.0)
|
27
27
|
Requires-Dist: structlog (>=24.4.0,<25.0.0)
|
28
|
-
Requires-Dist: typer (>=0.
|
28
|
+
Requires-Dist: typer (>=0.12,<0.13)
|
29
29
|
Project-URL: Repository, https://github.com/tensorlakeai/indexify
|
30
30
|
Description-Content-Type: text/markdown
|
31
31
|
|
@@ -22,7 +22,7 @@ from rich.panel import Panel
|
|
22
22
|
from rich.text import Text
|
23
23
|
from rich.theme import Theme
|
24
24
|
|
25
|
-
from indexify.executor.
|
25
|
+
from indexify.executor.executor import Executor
|
26
26
|
from indexify.function_executor.function_executor_service import (
|
27
27
|
FunctionExecutorService,
|
28
28
|
)
|
@@ -221,7 +221,7 @@ def executor(
|
|
221
221
|
shutil.rmtree(executor_cache)
|
222
222
|
Path(executor_cache).mkdir(parents=True, exist_ok=True)
|
223
223
|
|
224
|
-
|
224
|
+
executor = Executor(
|
225
225
|
id,
|
226
226
|
server_addr=server_addr,
|
227
227
|
config_path=config_path,
|
@@ -232,7 +232,7 @@ def executor(
|
|
232
232
|
)
|
233
233
|
|
234
234
|
try:
|
235
|
-
asyncio.get_event_loop().run_until_complete(
|
235
|
+
asyncio.get_event_loop().run_until_complete(executor.run())
|
236
236
|
except asyncio.CancelledError:
|
237
237
|
logger.info("graceful shutdown")
|
238
238
|
|
@@ -0,0 +1,155 @@
|
|
1
|
+
import asyncio
|
2
|
+
import signal
|
3
|
+
from pathlib import Path
|
4
|
+
from typing import Any, Optional
|
5
|
+
|
6
|
+
import structlog
|
7
|
+
|
8
|
+
from indexify.function_executor.proto.function_executor_pb2 import (
|
9
|
+
SerializedObject,
|
10
|
+
)
|
11
|
+
|
12
|
+
from .api_objects import Task
|
13
|
+
from .downloader import DownloadedInputs, Downloader
|
14
|
+
from .function_executor.process_function_executor_factory import (
|
15
|
+
ProcessFunctionExecutorFactory,
|
16
|
+
)
|
17
|
+
from .function_worker import (
|
18
|
+
FunctionWorker,
|
19
|
+
FunctionWorkerInput,
|
20
|
+
FunctionWorkerOutput,
|
21
|
+
)
|
22
|
+
from .task_fetcher import TaskFetcher
|
23
|
+
from .task_reporter import TaskReporter
|
24
|
+
|
25
|
+
|
26
|
+
class Executor:
|
27
|
+
def __init__(
|
28
|
+
self,
|
29
|
+
executor_id: str,
|
30
|
+
code_path: Path,
|
31
|
+
server_addr: str = "localhost:8900",
|
32
|
+
development_mode: bool = False,
|
33
|
+
config_path: Optional[str] = None,
|
34
|
+
name_alias: Optional[str] = None,
|
35
|
+
image_hash: Optional[str] = None,
|
36
|
+
):
|
37
|
+
self._logger = structlog.get_logger(module=__name__)
|
38
|
+
self._should_run = True
|
39
|
+
self._config_path = config_path
|
40
|
+
protocol: str = "http"
|
41
|
+
if config_path:
|
42
|
+
self._logger.info("running the extractor with TLS enabled")
|
43
|
+
protocol = "https"
|
44
|
+
|
45
|
+
self._function_worker = FunctionWorker(
|
46
|
+
function_executor_factory=ProcessFunctionExecutorFactory(
|
47
|
+
indexify_server_address=server_addr,
|
48
|
+
development_mode=development_mode,
|
49
|
+
config_path=config_path,
|
50
|
+
)
|
51
|
+
)
|
52
|
+
self._server_addr = server_addr
|
53
|
+
self._base_url = f"{protocol}://{self._server_addr}"
|
54
|
+
self._code_path = code_path
|
55
|
+
self._downloader = Downloader(
|
56
|
+
code_path=code_path, base_url=self._base_url, config_path=config_path
|
57
|
+
)
|
58
|
+
self._task_fetcher = TaskFetcher(
|
59
|
+
protocol=protocol,
|
60
|
+
indexify_server_addr=self._server_addr,
|
61
|
+
executor_id=executor_id,
|
62
|
+
name_alias=name_alias,
|
63
|
+
image_hash=image_hash,
|
64
|
+
config_path=config_path,
|
65
|
+
)
|
66
|
+
self._task_reporter = TaskReporter(
|
67
|
+
base_url=self._base_url,
|
68
|
+
executor_id=executor_id,
|
69
|
+
config_path=self._config_path,
|
70
|
+
)
|
71
|
+
|
72
|
+
async def run(self):
|
73
|
+
asyncio.get_event_loop().add_signal_handler(
|
74
|
+
signal.SIGINT, self.shutdown, asyncio.get_event_loop()
|
75
|
+
)
|
76
|
+
asyncio.get_event_loop().add_signal_handler(
|
77
|
+
signal.SIGTERM, self.shutdown, asyncio.get_event_loop()
|
78
|
+
)
|
79
|
+
|
80
|
+
while self._should_run:
|
81
|
+
try:
|
82
|
+
async for task in self._task_fetcher.run():
|
83
|
+
asyncio.create_task(self._run_task(task))
|
84
|
+
except Exception as e:
|
85
|
+
self._logger.error(
|
86
|
+
"failed fetching tasks, retrying in 5 seconds", exc_info=e
|
87
|
+
)
|
88
|
+
await asyncio.sleep(5)
|
89
|
+
|
90
|
+
async def _run_task(self, task: Task) -> None:
|
91
|
+
"""Runs the supplied task.
|
92
|
+
|
93
|
+
Doesn't raise any Exceptions. All errors are reported to the server."""
|
94
|
+
logger = self._task_logger(task)
|
95
|
+
output: Optional[FunctionWorkerOutput] = None
|
96
|
+
|
97
|
+
try:
|
98
|
+
graph: SerializedObject = await self._downloader.download_graph(task)
|
99
|
+
input: DownloadedInputs = await self._downloader.download_inputs(task)
|
100
|
+
output = await self._function_worker.run(
|
101
|
+
input=FunctionWorkerInput(
|
102
|
+
task=task,
|
103
|
+
graph=graph,
|
104
|
+
function_input=input,
|
105
|
+
)
|
106
|
+
)
|
107
|
+
logger.info("task_execution_finished", success=output.success)
|
108
|
+
except Exception as e:
|
109
|
+
logger.error("failed running the task", exc_info=e)
|
110
|
+
|
111
|
+
await self._report_task_outcome(task=task, output=output, logger=logger)
|
112
|
+
|
113
|
+
async def _report_task_outcome(
|
114
|
+
self, task: Task, output: Optional[FunctionWorkerOutput], logger: Any
|
115
|
+
) -> None:
|
116
|
+
"""Reports the task with the given output to the server.
|
117
|
+
|
118
|
+
None output means that the task execution didn't finish due to an internal error.
|
119
|
+
Doesn't raise any exceptions."""
|
120
|
+
reporting_retries: int = 0
|
121
|
+
|
122
|
+
while True:
|
123
|
+
logger = logger.bind(retries=reporting_retries)
|
124
|
+
try:
|
125
|
+
await self._task_reporter.report(
|
126
|
+
task=task, output=output, logger=logger
|
127
|
+
)
|
128
|
+
break
|
129
|
+
except Exception as e:
|
130
|
+
logger.error(
|
131
|
+
"failed_to_report_task",
|
132
|
+
exc_info=e,
|
133
|
+
)
|
134
|
+
reporting_retries += 1
|
135
|
+
await asyncio.sleep(5)
|
136
|
+
|
137
|
+
async def _shutdown(self, loop):
|
138
|
+
self._logger.info("shutting_down")
|
139
|
+
self._should_run = False
|
140
|
+
await self._function_worker.shutdown()
|
141
|
+
for task in asyncio.all_tasks(loop):
|
142
|
+
task.cancel()
|
143
|
+
|
144
|
+
def shutdown(self, loop):
|
145
|
+
loop.create_task(self._shutdown(loop))
|
146
|
+
|
147
|
+
def _task_logger(self, task: Task) -> Any:
|
148
|
+
return self._logger.bind(
|
149
|
+
namespace=task.namespace,
|
150
|
+
graph=task.compute_graph,
|
151
|
+
graph_version=task.graph_version,
|
152
|
+
invocation_id=task.invocation_id,
|
153
|
+
function_name=task.compute_fn,
|
154
|
+
task_id=task.id,
|
155
|
+
)
|
@@ -31,13 +31,11 @@ class FunctionWorkerInput:
|
|
31
31
|
def __init__(
|
32
32
|
self,
|
33
33
|
task: Task,
|
34
|
-
graph:
|
35
|
-
function_input:
|
34
|
+
graph: SerializedObject,
|
35
|
+
function_input: DownloadedInputs,
|
36
36
|
):
|
37
37
|
self.task = task
|
38
|
-
# Must not be None when running the task.
|
39
38
|
self.graph = graph
|
40
|
-
# Must not be None when running the task.
|
41
39
|
self.function_input = function_input
|
42
40
|
|
43
41
|
|
@@ -2,7 +2,6 @@ import json
|
|
2
2
|
from importlib.metadata import version
|
3
3
|
from typing import AsyncGenerator, Optional
|
4
4
|
|
5
|
-
import httpx
|
6
5
|
import structlog
|
7
6
|
from httpx_sse import aconnect_sse
|
8
7
|
|
@@ -66,7 +65,7 @@ class TaskFetcher:
|
|
66
65
|
except Exception as e:
|
67
66
|
await event_source.response.aread()
|
68
67
|
raise Exception(
|
69
|
-
"
|
68
|
+
"failed to register at server. "
|
70
69
|
f"Response code: {event_source.response.status_code}. "
|
71
70
|
f"Response text: '{event_source.response.text}'."
|
72
71
|
) from e
|
@@ -0,0 +1,215 @@
|
|
1
|
+
import asyncio
|
2
|
+
from typing import Any, List, Optional, Tuple
|
3
|
+
|
4
|
+
import nanoid
|
5
|
+
from httpx import Timeout
|
6
|
+
|
7
|
+
from indexify.common_util import get_httpx_client
|
8
|
+
from indexify.executor.api_objects import RouterOutput, Task, TaskResult
|
9
|
+
from indexify.function_executor.proto.function_executor_pb2 import (
|
10
|
+
FunctionOutput,
|
11
|
+
)
|
12
|
+
|
13
|
+
from .function_worker import FunctionWorkerOutput
|
14
|
+
|
15
|
+
|
16
|
+
# https://github.com/psf/requests/issues/1081#issuecomment-428504128
|
17
|
+
class ForceMultipartDict(dict):
|
18
|
+
def __bool__(self):
|
19
|
+
return True
|
20
|
+
|
21
|
+
|
22
|
+
FORCE_MULTIPART = ForceMultipartDict()
|
23
|
+
UTF_8_CONTENT_TYPE = "application/octet-stream"
|
24
|
+
|
25
|
+
|
26
|
+
class TaskOutputSummary:
|
27
|
+
def __init__(self):
|
28
|
+
self.output_count: int = 0
|
29
|
+
self.output_total_bytes: int = 0
|
30
|
+
self.router_output_count: int = 0
|
31
|
+
self.stdout_count: int = 0
|
32
|
+
self.stdout_total_bytes: int = 0
|
33
|
+
self.stderr_count: int = 0
|
34
|
+
self.stderr_total_bytes: int = 0
|
35
|
+
self.total_bytes: int = 0
|
36
|
+
|
37
|
+
|
38
|
+
class TaskReporter:
|
39
|
+
def __init__(
|
40
|
+
self, base_url: str, executor_id: str, config_path: Optional[str] = None
|
41
|
+
):
|
42
|
+
self._base_url = base_url
|
43
|
+
self._executor_id = executor_id
|
44
|
+
# Use thread-safe sync client due to issues with async client.
|
45
|
+
# Async client attempts to use connections it already closed.
|
46
|
+
# See e.g. https://github.com/encode/httpx/issues/2337.
|
47
|
+
# Creating a new async client for each request fixes this but it
|
48
|
+
# results in not reusing established TCP connections to server.
|
49
|
+
self._client = get_httpx_client(config_path, make_async=False)
|
50
|
+
|
51
|
+
async def report(
|
52
|
+
self, task: Task, output: Optional[FunctionWorkerOutput], logger: Any
|
53
|
+
):
|
54
|
+
"""Reports result of the supplied task.
|
55
|
+
|
56
|
+
If FunctionWorkerOutput is None this means that the task didn't finish and failed with internal error.
|
57
|
+
"""
|
58
|
+
logger = logger.bind(module=__name__)
|
59
|
+
task_result, output_files, output_summary = self._process_task_output(
|
60
|
+
task, output
|
61
|
+
)
|
62
|
+
task_result_data = task_result.model_dump_json(exclude_none=True)
|
63
|
+
|
64
|
+
logger.info(
|
65
|
+
"reporting task outcome",
|
66
|
+
total_bytes=output_summary.total_bytes,
|
67
|
+
total_files=output_summary.output_count
|
68
|
+
+ output_summary.stdout_count
|
69
|
+
+ output_summary.stderr_count,
|
70
|
+
output_files=output_summary.output_count,
|
71
|
+
output_bytes=output_summary.total_bytes,
|
72
|
+
router_output_count=output_summary.router_output_count,
|
73
|
+
stdout_bytes=output_summary.stdout_total_bytes,
|
74
|
+
stderr_bytes=output_summary.stderr_total_bytes,
|
75
|
+
)
|
76
|
+
|
77
|
+
kwargs = {
|
78
|
+
"data": {"task_result": task_result_data},
|
79
|
+
# Use httpx default timeout of 5s for all timeout types.
|
80
|
+
# For read timeouts, use 5 minutes to allow for large file uploads.
|
81
|
+
"timeout": Timeout(
|
82
|
+
5.0,
|
83
|
+
read=5.0 * 60,
|
84
|
+
),
|
85
|
+
"files": output_files if len(output_files) > 0 else FORCE_MULTIPART,
|
86
|
+
}
|
87
|
+
# Run in a separate thread to not block the main event loop.
|
88
|
+
response = await asyncio.to_thread(
|
89
|
+
self._client.post, url=f"{self._base_url}/internal/ingest_files", **kwargs
|
90
|
+
)
|
91
|
+
|
92
|
+
try:
|
93
|
+
response.raise_for_status()
|
94
|
+
except Exception as e:
|
95
|
+
# Caller catches and logs the exception.
|
96
|
+
raise Exception(
|
97
|
+
"failed to report task outcome. "
|
98
|
+
f"Response code: {response.status_code}. "
|
99
|
+
f"Response text: '{response.text}'."
|
100
|
+
) from e
|
101
|
+
|
102
|
+
def _process_task_output(
|
103
|
+
self, task: Task, output: Optional[FunctionWorkerOutput]
|
104
|
+
) -> Tuple[TaskResult, List[Any], TaskOutputSummary]:
|
105
|
+
task_result = TaskResult(
|
106
|
+
outcome="failure",
|
107
|
+
namespace=task.namespace,
|
108
|
+
compute_graph=task.compute_graph,
|
109
|
+
compute_fn=task.compute_fn,
|
110
|
+
invocation_id=task.invocation_id,
|
111
|
+
executor_id=self._executor_id,
|
112
|
+
task_id=task.id,
|
113
|
+
)
|
114
|
+
output_files: List[Any] = []
|
115
|
+
summary: TaskOutputSummary = TaskOutputSummary()
|
116
|
+
if output is None:
|
117
|
+
return task_result, output_files, summary
|
118
|
+
|
119
|
+
task_result.outcome = "success" if output.success else "failure"
|
120
|
+
task_result.reducer = output.reducer
|
121
|
+
|
122
|
+
_process_function_output(
|
123
|
+
function_output=output.function_output,
|
124
|
+
output_files=output_files,
|
125
|
+
summary=summary,
|
126
|
+
)
|
127
|
+
_process_router_output(
|
128
|
+
router_output=output.router_output, task_result=task_result, summary=summary
|
129
|
+
)
|
130
|
+
_process_stdout(
|
131
|
+
stdout=output.stdout, output_files=output_files, summary=summary
|
132
|
+
)
|
133
|
+
_process_stderr(
|
134
|
+
stderr=output.stderr, output_files=output_files, summary=summary
|
135
|
+
)
|
136
|
+
|
137
|
+
summary.total_bytes = (
|
138
|
+
summary.output_total_bytes
|
139
|
+
+ summary.stdout_total_bytes
|
140
|
+
+ summary.stderr_total_bytes
|
141
|
+
)
|
142
|
+
|
143
|
+
return task_result, output_files, summary
|
144
|
+
|
145
|
+
|
146
|
+
def _process_function_output(
|
147
|
+
function_output: Optional[FunctionOutput],
|
148
|
+
output_files: List[Any],
|
149
|
+
summary: TaskOutputSummary,
|
150
|
+
) -> None:
|
151
|
+
if function_output is None:
|
152
|
+
return
|
153
|
+
|
154
|
+
for output in function_output.outputs or []:
|
155
|
+
payload = output.bytes if output.HasField("bytes") else output.string
|
156
|
+
output_files.append(
|
157
|
+
(
|
158
|
+
"node_outputs",
|
159
|
+
(nanoid.generate(), payload, output.content_type),
|
160
|
+
)
|
161
|
+
)
|
162
|
+
summary.output_count += 1
|
163
|
+
summary.output_total_bytes += len(payload)
|
164
|
+
|
165
|
+
|
166
|
+
def _process_router_output(
|
167
|
+
router_output: Optional[RouterOutput],
|
168
|
+
task_result: TaskResult,
|
169
|
+
summary: TaskOutputSummary,
|
170
|
+
) -> None:
|
171
|
+
if router_output is None:
|
172
|
+
return
|
173
|
+
|
174
|
+
task_result.router_output = RouterOutput(edges=router_output.edges)
|
175
|
+
summary.router_output_count += 1
|
176
|
+
|
177
|
+
|
178
|
+
def _process_stdout(
|
179
|
+
stdout: Optional[str], output_files: List[Any], summary: TaskOutputSummary
|
180
|
+
) -> None:
|
181
|
+
if stdout is None:
|
182
|
+
return
|
183
|
+
|
184
|
+
output_files.append(
|
185
|
+
(
|
186
|
+
"stdout",
|
187
|
+
(
|
188
|
+
nanoid.generate(),
|
189
|
+
stdout.encode(),
|
190
|
+
UTF_8_CONTENT_TYPE,
|
191
|
+
),
|
192
|
+
)
|
193
|
+
)
|
194
|
+
summary.stdout_count += 1
|
195
|
+
summary.stdout_total_bytes += len(stdout)
|
196
|
+
|
197
|
+
|
198
|
+
def _process_stderr(
|
199
|
+
stderr: Optional[str], output_files: List[Any], summary: TaskOutputSummary
|
200
|
+
) -> None:
|
201
|
+
if stderr is None:
|
202
|
+
return
|
203
|
+
|
204
|
+
output_files.append(
|
205
|
+
(
|
206
|
+
"stderr",
|
207
|
+
(
|
208
|
+
nanoid.generate(),
|
209
|
+
stderr.encode(),
|
210
|
+
UTF_8_CONTENT_TYPE,
|
211
|
+
),
|
212
|
+
)
|
213
|
+
)
|
214
|
+
summary.stderr_count += 1
|
215
|
+
summary.stderr_total_bytes += len(stderr)
|
@@ -37,7 +37,6 @@ from .indexify_functions import (
|
|
37
37
|
IndexifyRouter,
|
38
38
|
RouterCallResult,
|
39
39
|
)
|
40
|
-
from .local_cache import CacheAwareFunctionWrapper
|
41
40
|
from .object_serializer import get_serializer
|
42
41
|
|
43
42
|
RouterFn = Annotated[
|
@@ -86,7 +85,6 @@ class Graph:
|
|
86
85
|
|
87
86
|
# Storage for local execution
|
88
87
|
self._results: Dict[str, Dict[str, List[IndexifyData]]] = {}
|
89
|
-
self._cache = CacheAwareFunctionWrapper("./indexify_local_runner_cache")
|
90
88
|
self._accumulator_values: Dict[str, IndexifyData] = {}
|
91
89
|
self._local_graph_ctx: Optional[GraphInvocationContext] = None
|
92
90
|
|
@@ -1,6 +1,6 @@
|
|
1
1
|
[tool.poetry]
|
2
2
|
name = "indexify"
|
3
|
-
version = "0.2.
|
3
|
+
version = "0.2.44"
|
4
4
|
description = "Python Client for Indexify"
|
5
5
|
authors = ["Tensorlake Inc. <support@tensorlake.ai>"]
|
6
6
|
license = "Apache 2.0"
|
@@ -15,12 +15,12 @@ indexify-cli = "indexify.cli:app"
|
|
15
15
|
python = "^3.9"
|
16
16
|
httpx = { version = "0.27.2", extras = ["http2"] }
|
17
17
|
pyyaml = "^6"
|
18
|
-
pydantic = "2.10.2"
|
18
|
+
pydantic = "^2.10.2"
|
19
19
|
cloudpickle = "^3.1.0"
|
20
20
|
rich = "^13.9.2"
|
21
21
|
nanoid = "^2.0.0"
|
22
22
|
docker = "^7.1.0"
|
23
|
-
typer = "^0.
|
23
|
+
typer = "^0.12"
|
24
24
|
httpx-sse = "^0.4.0"
|
25
25
|
structlog = "^24.4.0"
|
26
26
|
grpcio = "1.68.1"
|
@@ -1,262 +0,0 @@
|
|
1
|
-
import asyncio
|
2
|
-
from pathlib import Path
|
3
|
-
from typing import Dict, List, Optional
|
4
|
-
|
5
|
-
import structlog
|
6
|
-
|
7
|
-
from .downloader import Downloader
|
8
|
-
from .executor_tasks import DownloadGraphTask, DownloadInputsTask, RunTask
|
9
|
-
from .function_executor.process_function_executor_factory import (
|
10
|
-
ProcessFunctionExecutorFactory,
|
11
|
-
)
|
12
|
-
from .function_worker import (
|
13
|
-
FunctionWorker,
|
14
|
-
FunctionWorkerInput,
|
15
|
-
FunctionWorkerOutput,
|
16
|
-
)
|
17
|
-
from .task_fetcher import TaskFetcher
|
18
|
-
from .task_reporter import TaskReporter
|
19
|
-
from .task_store import CompletedTask, TaskStore
|
20
|
-
|
21
|
-
logger = structlog.get_logger(module=__name__)
|
22
|
-
|
23
|
-
|
24
|
-
class ExtractorAgent:
|
25
|
-
def __init__(
|
26
|
-
self,
|
27
|
-
executor_id: str,
|
28
|
-
code_path: Path,
|
29
|
-
server_addr: str = "localhost:8900",
|
30
|
-
development_mode: bool = False,
|
31
|
-
config_path: Optional[str] = None,
|
32
|
-
name_alias: Optional[str] = None,
|
33
|
-
image_hash: Optional[str] = None,
|
34
|
-
):
|
35
|
-
self._config_path = config_path
|
36
|
-
protocol: str = "http"
|
37
|
-
if config_path:
|
38
|
-
logger.info("running the extractor with TLS enabled")
|
39
|
-
protocol = "https"
|
40
|
-
|
41
|
-
self._task_store: TaskStore = TaskStore()
|
42
|
-
self._function_worker = FunctionWorker(
|
43
|
-
function_executor_factory=ProcessFunctionExecutorFactory(
|
44
|
-
indexify_server_address=server_addr,
|
45
|
-
development_mode=development_mode,
|
46
|
-
config_path=config_path,
|
47
|
-
)
|
48
|
-
)
|
49
|
-
self._has_registered = False
|
50
|
-
self._server_addr = server_addr
|
51
|
-
self._base_url = f"{protocol}://{self._server_addr}"
|
52
|
-
self._code_path = code_path
|
53
|
-
self._downloader = Downloader(
|
54
|
-
code_path=code_path, base_url=self._base_url, config_path=config_path
|
55
|
-
)
|
56
|
-
self._task_fetcher = TaskFetcher(
|
57
|
-
protocol=protocol,
|
58
|
-
indexify_server_addr=self._server_addr,
|
59
|
-
executor_id=executor_id,
|
60
|
-
name_alias=name_alias,
|
61
|
-
image_hash=image_hash,
|
62
|
-
config_path=config_path,
|
63
|
-
)
|
64
|
-
self._task_reporter = TaskReporter(
|
65
|
-
base_url=self._base_url,
|
66
|
-
executor_id=executor_id,
|
67
|
-
config_path=self._config_path,
|
68
|
-
)
|
69
|
-
|
70
|
-
async def task_completion_reporter(self):
|
71
|
-
logger.info("starting task completion reporter")
|
72
|
-
# We should copy only the keys and not the values
|
73
|
-
while True:
|
74
|
-
outcomes = await self._task_store.task_outcomes()
|
75
|
-
for task_outcome in outcomes:
|
76
|
-
logger.info(
|
77
|
-
"reporting_task_outcome",
|
78
|
-
task_id=task_outcome.task.id,
|
79
|
-
fn_name=task_outcome.task.compute_fn,
|
80
|
-
num_outputs=(
|
81
|
-
len(task_outcome.function_output.outputs)
|
82
|
-
if task_outcome.function_output is not None
|
83
|
-
else 0
|
84
|
-
),
|
85
|
-
router_output=task_outcome.router_output,
|
86
|
-
outcome=task_outcome.task_outcome,
|
87
|
-
retries=task_outcome.reporting_retries,
|
88
|
-
)
|
89
|
-
|
90
|
-
try:
|
91
|
-
# Send task outcome to the server
|
92
|
-
self._task_reporter.report_task_outcome(completed_task=task_outcome)
|
93
|
-
except Exception as e:
|
94
|
-
# The connection was dropped in the middle of the reporting, process, retry
|
95
|
-
logger.error(
|
96
|
-
"failed_to_report_task",
|
97
|
-
task_id=task_outcome.task.id,
|
98
|
-
exc_info=e,
|
99
|
-
retries=task_outcome.reporting_retries,
|
100
|
-
)
|
101
|
-
task_outcome.reporting_retries += 1
|
102
|
-
await asyncio.sleep(5)
|
103
|
-
continue
|
104
|
-
|
105
|
-
self._task_store.mark_reported(task_id=task_outcome.task.id)
|
106
|
-
|
107
|
-
async def task_launcher(self):
|
108
|
-
async_tasks: List[asyncio.Task] = [
|
109
|
-
asyncio.create_task(
|
110
|
-
self._task_store.get_runnable_tasks(), name="get_runnable_tasks"
|
111
|
-
)
|
112
|
-
]
|
113
|
-
|
114
|
-
while True:
|
115
|
-
done, pending = await asyncio.wait(
|
116
|
-
async_tasks, return_when=asyncio.FIRST_COMPLETED
|
117
|
-
)
|
118
|
-
|
119
|
-
async_tasks: List[asyncio.Task] = list(pending)
|
120
|
-
for async_task in done:
|
121
|
-
if async_task.get_name() == "get_runnable_tasks":
|
122
|
-
if async_task.exception():
|
123
|
-
logger.error(
|
124
|
-
"task_launcher_error, failed to get runnable tasks",
|
125
|
-
exc_info=async_task.exception(),
|
126
|
-
)
|
127
|
-
continue
|
128
|
-
result: Dict[str, Task] = await async_task
|
129
|
-
task: Task
|
130
|
-
for _, task in result.items():
|
131
|
-
async_tasks.append(
|
132
|
-
DownloadGraphTask(
|
133
|
-
function_worker_input=FunctionWorkerInput(task=task),
|
134
|
-
downloader=self._downloader,
|
135
|
-
)
|
136
|
-
)
|
137
|
-
async_tasks.append(
|
138
|
-
asyncio.create_task(
|
139
|
-
self._task_store.get_runnable_tasks(),
|
140
|
-
name="get_runnable_tasks",
|
141
|
-
)
|
142
|
-
)
|
143
|
-
elif async_task.get_name() == "download_graph":
|
144
|
-
if async_task.exception():
|
145
|
-
logger.error(
|
146
|
-
"task_launcher_error, failed to download graph",
|
147
|
-
exc_info=async_task.exception(),
|
148
|
-
)
|
149
|
-
completed_task = CompletedTask(
|
150
|
-
task=async_task.function_worker_input.task,
|
151
|
-
task_outcome="failure",
|
152
|
-
)
|
153
|
-
self._task_store.complete(outcome=completed_task)
|
154
|
-
continue
|
155
|
-
async_task: DownloadGraphTask
|
156
|
-
function_worker_input: FunctionWorkerInput = (
|
157
|
-
async_task.function_worker_input
|
158
|
-
)
|
159
|
-
function_worker_input.graph = await async_task
|
160
|
-
async_tasks.append(
|
161
|
-
DownloadInputsTask(
|
162
|
-
function_worker_input=function_worker_input,
|
163
|
-
downloader=self._downloader,
|
164
|
-
)
|
165
|
-
)
|
166
|
-
elif async_task.get_name() == "download_inputs":
|
167
|
-
if async_task.exception():
|
168
|
-
logger.error(
|
169
|
-
"task_launcher_error, failed to download inputs",
|
170
|
-
exc_info=async_task.exception(),
|
171
|
-
)
|
172
|
-
completed_task = CompletedTask(
|
173
|
-
task=async_task.function_worker_input.task,
|
174
|
-
task_outcome="failure",
|
175
|
-
)
|
176
|
-
self._task_store.complete(outcome=completed_task)
|
177
|
-
continue
|
178
|
-
async_task: DownloadInputsTask
|
179
|
-
function_worker_input: FunctionWorkerInput = (
|
180
|
-
async_task.function_worker_input
|
181
|
-
)
|
182
|
-
function_worker_input.function_input = await async_task
|
183
|
-
async_tasks.append(
|
184
|
-
RunTask(
|
185
|
-
function_worker=self._function_worker,
|
186
|
-
function_worker_input=function_worker_input,
|
187
|
-
)
|
188
|
-
)
|
189
|
-
elif async_task.get_name() == "run_task":
|
190
|
-
if async_task.exception():
|
191
|
-
completed_task = CompletedTask(
|
192
|
-
task=async_task.function_worker_input.task,
|
193
|
-
task_outcome="failure",
|
194
|
-
stderr=str(async_task.exception()),
|
195
|
-
)
|
196
|
-
self._task_store.complete(outcome=completed_task)
|
197
|
-
continue
|
198
|
-
async_task: RunTask
|
199
|
-
try:
|
200
|
-
outputs: FunctionWorkerOutput = await async_task
|
201
|
-
if not outputs.success:
|
202
|
-
task_outcome = "failure"
|
203
|
-
else:
|
204
|
-
task_outcome = "success"
|
205
|
-
|
206
|
-
completed_task = CompletedTask(
|
207
|
-
task=async_task.function_worker_input.task,
|
208
|
-
task_outcome=task_outcome,
|
209
|
-
function_output=outputs.function_output,
|
210
|
-
router_output=outputs.router_output,
|
211
|
-
stdout=outputs.stdout,
|
212
|
-
stderr=outputs.stderr,
|
213
|
-
reducer=outputs.reducer,
|
214
|
-
)
|
215
|
-
self._task_store.complete(outcome=completed_task)
|
216
|
-
except Exception as e:
|
217
|
-
logger.error(
|
218
|
-
"failed to execute task",
|
219
|
-
task_id=async_task.function_worker_input.task.id,
|
220
|
-
exc_info=e,
|
221
|
-
)
|
222
|
-
completed_task = CompletedTask(
|
223
|
-
task=async_task.function_worker_input.task,
|
224
|
-
task_outcome="failure",
|
225
|
-
)
|
226
|
-
self._task_store.complete(outcome=completed_task)
|
227
|
-
continue
|
228
|
-
|
229
|
-
async def _main_loop(self):
|
230
|
-
"""Fetches incoming tasks from the server and starts their processing."""
|
231
|
-
self._should_run = True
|
232
|
-
while self._should_run:
|
233
|
-
try:
|
234
|
-
async for task in self._task_fetcher.run():
|
235
|
-
self._task_store.add_tasks([task])
|
236
|
-
except Exception as e:
|
237
|
-
logger.error("failed fetching tasks, retrying in 5 seconds", exc_info=e)
|
238
|
-
await asyncio.sleep(5)
|
239
|
-
continue
|
240
|
-
|
241
|
-
async def run(self):
|
242
|
-
import signal
|
243
|
-
|
244
|
-
asyncio.get_event_loop().add_signal_handler(
|
245
|
-
signal.SIGINT, self.shutdown, asyncio.get_event_loop()
|
246
|
-
)
|
247
|
-
asyncio.get_event_loop().add_signal_handler(
|
248
|
-
signal.SIGTERM, self.shutdown, asyncio.get_event_loop()
|
249
|
-
)
|
250
|
-
asyncio.create_task(self.task_launcher())
|
251
|
-
asyncio.create_task(self.task_completion_reporter())
|
252
|
-
await self._main_loop()
|
253
|
-
|
254
|
-
async def _shutdown(self, loop):
|
255
|
-
logger.info("shutting_down")
|
256
|
-
self._should_run = False
|
257
|
-
await self._function_worker.shutdown()
|
258
|
-
for task in asyncio.all_tasks(loop):
|
259
|
-
task.cancel()
|
260
|
-
|
261
|
-
def shutdown(self, loop):
|
262
|
-
loop.create_task(self._shutdown(loop))
|
@@ -1,58 +0,0 @@
|
|
1
|
-
import asyncio
|
2
|
-
|
3
|
-
from pydantic import BaseModel
|
4
|
-
|
5
|
-
from .api_objects import Task
|
6
|
-
from .downloader import Downloader
|
7
|
-
from .function_worker import FunctionWorker, FunctionWorkerInput
|
8
|
-
|
9
|
-
|
10
|
-
class DownloadGraphTask(asyncio.Task):
|
11
|
-
def __init__(
|
12
|
-
self,
|
13
|
-
*,
|
14
|
-
function_worker_input: FunctionWorkerInput,
|
15
|
-
downloader: Downloader,
|
16
|
-
**kwargs,
|
17
|
-
):
|
18
|
-
kwargs["name"] = "download_graph"
|
19
|
-
kwargs["loop"] = asyncio.get_event_loop()
|
20
|
-
super().__init__(
|
21
|
-
downloader.download_graph(function_worker_input.task),
|
22
|
-
**kwargs,
|
23
|
-
)
|
24
|
-
self.function_worker_input = function_worker_input
|
25
|
-
|
26
|
-
|
27
|
-
class DownloadInputsTask(asyncio.Task):
|
28
|
-
def __init__(
|
29
|
-
self,
|
30
|
-
*,
|
31
|
-
function_worker_input: FunctionWorkerInput,
|
32
|
-
downloader: Downloader,
|
33
|
-
**kwargs,
|
34
|
-
):
|
35
|
-
kwargs["name"] = "download_inputs"
|
36
|
-
kwargs["loop"] = asyncio.get_event_loop()
|
37
|
-
super().__init__(
|
38
|
-
downloader.download_inputs(function_worker_input.task),
|
39
|
-
**kwargs,
|
40
|
-
)
|
41
|
-
self.function_worker_input = function_worker_input
|
42
|
-
|
43
|
-
|
44
|
-
class RunTask(asyncio.Task):
|
45
|
-
def __init__(
|
46
|
-
self,
|
47
|
-
*,
|
48
|
-
function_worker: FunctionWorker,
|
49
|
-
function_worker_input: FunctionWorkerInput,
|
50
|
-
**kwargs,
|
51
|
-
):
|
52
|
-
kwargs["name"] = "run_task"
|
53
|
-
kwargs["loop"] = asyncio.get_event_loop()
|
54
|
-
super().__init__(
|
55
|
-
function_worker.run(function_worker_input),
|
56
|
-
**kwargs,
|
57
|
-
)
|
58
|
-
self.function_worker_input = function_worker_input
|
@@ -1,155 +0,0 @@
|
|
1
|
-
from typing import Optional
|
2
|
-
|
3
|
-
import nanoid
|
4
|
-
import structlog
|
5
|
-
from httpx import Timeout
|
6
|
-
from pydantic import BaseModel
|
7
|
-
|
8
|
-
from indexify.common_util import get_httpx_client
|
9
|
-
from indexify.executor.api_objects import RouterOutput as ApiRouterOutput
|
10
|
-
from indexify.executor.api_objects import TaskResult
|
11
|
-
from indexify.executor.task_store import CompletedTask
|
12
|
-
|
13
|
-
logger = structlog.get_logger(__name__)
|
14
|
-
|
15
|
-
|
16
|
-
# https://github.com/psf/requests/issues/1081#issuecomment-428504128
|
17
|
-
class ForceMultipartDict(dict):
|
18
|
-
def __bool__(self):
|
19
|
-
return True
|
20
|
-
|
21
|
-
|
22
|
-
FORCE_MULTIPART = ForceMultipartDict()
|
23
|
-
UTF_8_CONTENT_TYPE = "application/octet-stream"
|
24
|
-
|
25
|
-
|
26
|
-
class ReportingData(BaseModel):
|
27
|
-
output_count: int = 0
|
28
|
-
output_total_bytes: int = 0
|
29
|
-
stdout_count: int = 0
|
30
|
-
stdout_total_bytes: int = 0
|
31
|
-
stderr_count: int = 0
|
32
|
-
stderr_total_bytes: int = 0
|
33
|
-
|
34
|
-
|
35
|
-
class TaskReporter:
|
36
|
-
def __init__(
|
37
|
-
self, base_url: str, executor_id: str, config_path: Optional[str] = None
|
38
|
-
):
|
39
|
-
self._base_url = base_url
|
40
|
-
self._executor_id = executor_id
|
41
|
-
self._client = get_httpx_client(config_path)
|
42
|
-
|
43
|
-
def report_task_outcome(self, completed_task: CompletedTask):
|
44
|
-
report = ReportingData()
|
45
|
-
fn_outputs = []
|
46
|
-
|
47
|
-
if completed_task.function_output:
|
48
|
-
for output in completed_task.function_output.outputs or []:
|
49
|
-
payload = output.bytes if output.HasField("bytes") else output.string
|
50
|
-
fn_outputs.append(
|
51
|
-
(
|
52
|
-
"node_outputs",
|
53
|
-
(nanoid.generate(), payload, output.content_type),
|
54
|
-
)
|
55
|
-
)
|
56
|
-
report.output_count += 1
|
57
|
-
report.output_total_bytes += len(payload)
|
58
|
-
|
59
|
-
if completed_task.stdout:
|
60
|
-
fn_outputs.append(
|
61
|
-
(
|
62
|
-
"stdout",
|
63
|
-
(
|
64
|
-
nanoid.generate(),
|
65
|
-
completed_task.stdout.encode(),
|
66
|
-
UTF_8_CONTENT_TYPE,
|
67
|
-
),
|
68
|
-
)
|
69
|
-
)
|
70
|
-
report.stdout_count += 1
|
71
|
-
report.stdout_total_bytes += len(completed_task.stdout)
|
72
|
-
|
73
|
-
if completed_task.stderr:
|
74
|
-
fn_outputs.append(
|
75
|
-
(
|
76
|
-
"stderr",
|
77
|
-
(
|
78
|
-
nanoid.generate(),
|
79
|
-
completed_task.stderr.encode(),
|
80
|
-
UTF_8_CONTENT_TYPE,
|
81
|
-
),
|
82
|
-
)
|
83
|
-
)
|
84
|
-
report.stderr_count += 1
|
85
|
-
report.stderr_total_bytes += len(completed_task.stderr)
|
86
|
-
|
87
|
-
router_output = (
|
88
|
-
ApiRouterOutput(edges=completed_task.router_output.edges)
|
89
|
-
if completed_task.router_output
|
90
|
-
else None
|
91
|
-
)
|
92
|
-
|
93
|
-
task_result = TaskResult(
|
94
|
-
router_output=router_output,
|
95
|
-
outcome=completed_task.task_outcome,
|
96
|
-
namespace=completed_task.task.namespace,
|
97
|
-
compute_graph=completed_task.task.compute_graph,
|
98
|
-
compute_fn=completed_task.task.compute_fn,
|
99
|
-
invocation_id=completed_task.task.invocation_id,
|
100
|
-
executor_id=self._executor_id,
|
101
|
-
task_id=completed_task.task.id,
|
102
|
-
reducer=completed_task.reducer,
|
103
|
-
)
|
104
|
-
task_result_data = task_result.model_dump_json(exclude_none=True)
|
105
|
-
|
106
|
-
total_bytes = (
|
107
|
-
report.output_total_bytes
|
108
|
-
+ report.stdout_total_bytes
|
109
|
-
+ report.stderr_total_bytes
|
110
|
-
)
|
111
|
-
|
112
|
-
logger.info(
|
113
|
-
"reporting task outcome",
|
114
|
-
task_id=completed_task.task.id,
|
115
|
-
retries=completed_task.reporting_retries,
|
116
|
-
total_bytes=total_bytes,
|
117
|
-
total_files=report.output_count + report.stdout_count + report.stderr_count,
|
118
|
-
output_files=report.output_count,
|
119
|
-
output_bytes=total_bytes,
|
120
|
-
stdout_bytes=report.stdout_total_bytes,
|
121
|
-
stderr_bytes=report.stderr_total_bytes,
|
122
|
-
)
|
123
|
-
#
|
124
|
-
kwargs = {
|
125
|
-
"data": {"task_result": task_result_data},
|
126
|
-
# Use httpx default timeout of 5s for all timeout types.
|
127
|
-
# For read timeouts, use 5 minutes to allow for large file uploads.
|
128
|
-
"timeout": Timeout(
|
129
|
-
5.0,
|
130
|
-
read=5.0 * 60,
|
131
|
-
),
|
132
|
-
}
|
133
|
-
if fn_outputs and len(fn_outputs) > 0:
|
134
|
-
kwargs["files"] = fn_outputs
|
135
|
-
else:
|
136
|
-
kwargs["files"] = FORCE_MULTIPART
|
137
|
-
|
138
|
-
response = self._client.post(
|
139
|
-
url=f"{self._base_url}/internal/ingest_files",
|
140
|
-
**kwargs,
|
141
|
-
)
|
142
|
-
|
143
|
-
try:
|
144
|
-
response.raise_for_status()
|
145
|
-
except Exception as e:
|
146
|
-
# Caller catches and logs the exception.
|
147
|
-
# Log response details here for easier debugging.
|
148
|
-
logger.error(
|
149
|
-
"failed to report task outcome",
|
150
|
-
task_id=completed_task.task.id,
|
151
|
-
retries=completed_task.reporting_retries,
|
152
|
-
status_code=response.status_code,
|
153
|
-
response_text=response.text,
|
154
|
-
)
|
155
|
-
raise e
|
@@ -1,132 +0,0 @@
|
|
1
|
-
import asyncio
|
2
|
-
from typing import Dict, List, Literal, Optional
|
3
|
-
|
4
|
-
import structlog
|
5
|
-
|
6
|
-
from indexify.function_executor.proto.function_executor_pb2 import (
|
7
|
-
FunctionOutput,
|
8
|
-
RouterOutput,
|
9
|
-
)
|
10
|
-
|
11
|
-
from .api_objects import Task
|
12
|
-
|
13
|
-
logger = structlog.get_logger(module=__name__)
|
14
|
-
|
15
|
-
|
16
|
-
class CompletedTask:
|
17
|
-
def __init__(
|
18
|
-
self,
|
19
|
-
task: Task,
|
20
|
-
task_outcome: Literal["success", "failure"],
|
21
|
-
function_output: Optional[FunctionOutput] = None,
|
22
|
-
router_output: Optional[RouterOutput] = None,
|
23
|
-
stdout: Optional[str] = None,
|
24
|
-
stderr: Optional[str] = None,
|
25
|
-
reducer: bool = False,
|
26
|
-
reporting_retries: int = 0,
|
27
|
-
):
|
28
|
-
self.task = task
|
29
|
-
self.task_outcome = task_outcome
|
30
|
-
self.function_output = function_output
|
31
|
-
self.router_output = router_output
|
32
|
-
self.stdout = stdout
|
33
|
-
self.stderr = stderr
|
34
|
-
self.reducer = reducer
|
35
|
-
self.reporting_retries = reporting_retries
|
36
|
-
|
37
|
-
|
38
|
-
class TaskStore:
|
39
|
-
def __init__(self) -> None:
|
40
|
-
self._tasks: Dict[str, Task] = {}
|
41
|
-
self._running_tasks: Dict[str, Task] = {}
|
42
|
-
self._finished: Dict[str, CompletedTask] = {}
|
43
|
-
self._retries: Dict[str, int] = {}
|
44
|
-
self._new_task_event = asyncio.Event()
|
45
|
-
self._finished_task_event = asyncio.Event()
|
46
|
-
|
47
|
-
def get_task(self, id) -> Task:
|
48
|
-
return self._tasks[id]
|
49
|
-
|
50
|
-
def add_tasks(self, tasks: List[Task]):
|
51
|
-
task: Task
|
52
|
-
for task in tasks:
|
53
|
-
if (
|
54
|
-
(task.id in self._tasks)
|
55
|
-
or (task.id in self._running_tasks)
|
56
|
-
or (task.id in self._finished)
|
57
|
-
):
|
58
|
-
continue
|
59
|
-
logger.info(
|
60
|
-
"added task",
|
61
|
-
task_id=task.id,
|
62
|
-
namespace=task.namespace,
|
63
|
-
graph=task.compute_graph,
|
64
|
-
fn=task.compute_fn,
|
65
|
-
)
|
66
|
-
self._tasks[task.id] = task
|
67
|
-
self._new_task_event.set()
|
68
|
-
|
69
|
-
async def get_runnable_tasks(self) -> Dict[str, Task]:
|
70
|
-
while True:
|
71
|
-
runnable_tasks = set(self._tasks) - set(self._running_tasks)
|
72
|
-
runnable_tasks = set(runnable_tasks) - set(self._finished)
|
73
|
-
if len(runnable_tasks) == 0:
|
74
|
-
await self._new_task_event.wait()
|
75
|
-
self._new_task_event.clear()
|
76
|
-
else:
|
77
|
-
break
|
78
|
-
out = {}
|
79
|
-
for task_id in runnable_tasks:
|
80
|
-
out[task_id] = self._tasks[task_id]
|
81
|
-
self._running_tasks[task_id] = self._tasks[task_id]
|
82
|
-
return out
|
83
|
-
|
84
|
-
def complete(self, outcome: CompletedTask):
|
85
|
-
self._retries.pop(outcome.task.id, None)
|
86
|
-
self._finished[outcome.task.id] = outcome
|
87
|
-
if outcome.task.id in self._running_tasks:
|
88
|
-
self._running_tasks.pop(outcome.task.id)
|
89
|
-
self._finished_task_event.set()
|
90
|
-
|
91
|
-
def retriable_failure(self, task_id: str):
|
92
|
-
self._running_tasks.pop(task_id)
|
93
|
-
if task_id not in self._retries:
|
94
|
-
self._retries[task_id] = 0
|
95
|
-
self._retries[task_id] += 1
|
96
|
-
if self._retries[task_id] > 3:
|
97
|
-
self._retries.pop(task_id)
|
98
|
-
self.complete(
|
99
|
-
outcome=CompletedTask(
|
100
|
-
task_id=task_id, task_outcome="failed", outputs=[]
|
101
|
-
)
|
102
|
-
)
|
103
|
-
else:
|
104
|
-
self._new_task_event.set()
|
105
|
-
|
106
|
-
def mark_reported(self, task_id: str):
|
107
|
-
self._tasks.pop(task_id)
|
108
|
-
self._finished.pop(task_id)
|
109
|
-
logger.info("removed task", task_id=task_id)
|
110
|
-
|
111
|
-
def report_failed(self, task_id: str):
|
112
|
-
if self._finished[task_id].task_outcome != "Failed":
|
113
|
-
# An error occurred while reporting the task, mark it as failed
|
114
|
-
# and try reporting again.
|
115
|
-
self._finished[task_id].task_outcome = "Failed"
|
116
|
-
else:
|
117
|
-
# If a task is already marked as failed, remove it from the queue.
|
118
|
-
# The only possible error at this point is task not present at
|
119
|
-
# the coordinator.
|
120
|
-
self._tasks.pop(task_id)
|
121
|
-
|
122
|
-
def num_pending_tasks(self) -> int:
|
123
|
-
return len(self._tasks) + len(self._running_tasks)
|
124
|
-
|
125
|
-
async def task_outcomes(self) -> List[CompletedTask]:
|
126
|
-
while True:
|
127
|
-
if len(self._finished) == 0:
|
128
|
-
await self._finished_task_event.wait()
|
129
|
-
self._finished_task_event.clear()
|
130
|
-
else:
|
131
|
-
break
|
132
|
-
return self._finished.copy().values()
|
@@ -1,46 +0,0 @@
|
|
1
|
-
import os
|
2
|
-
from hashlib import sha256
|
3
|
-
from typing import List, Optional
|
4
|
-
|
5
|
-
|
6
|
-
class CacheAwareFunctionWrapper:
|
7
|
-
def __init__(self, cache_dir: str):
|
8
|
-
self._cache_dir = cache_dir
|
9
|
-
if not os.path.exists(cache_dir):
|
10
|
-
os.makedirs(cache_dir)
|
11
|
-
|
12
|
-
def _get_key(self, input: bytes) -> str:
|
13
|
-
h = sha256()
|
14
|
-
h.update(input)
|
15
|
-
return h.hexdigest()
|
16
|
-
|
17
|
-
def get(self, graph: str, node_name: str, input: bytes) -> Optional[List[bytes]]:
|
18
|
-
key = self._get_key(input)
|
19
|
-
dir_path = os.path.join(self._cache_dir, graph, node_name, key)
|
20
|
-
if not os.path.exists(dir_path):
|
21
|
-
return None
|
22
|
-
|
23
|
-
files = os.listdir(dir_path)
|
24
|
-
outputs = []
|
25
|
-
for file in files:
|
26
|
-
with open(os.path.join(dir_path, file), "rb") as f:
|
27
|
-
return f.read()
|
28
|
-
|
29
|
-
return outputs
|
30
|
-
|
31
|
-
def set(
|
32
|
-
self,
|
33
|
-
graph: str,
|
34
|
-
node_name: str,
|
35
|
-
input: bytes,
|
36
|
-
output: List[bytes],
|
37
|
-
):
|
38
|
-
key = self._get_key(input)
|
39
|
-
dir_path = os.path.join(self._cache_dir, graph, node_name, key)
|
40
|
-
if not os.path.exists(dir_path):
|
41
|
-
os.makedirs(dir_path)
|
42
|
-
|
43
|
-
for i, output_item in enumerate(output):
|
44
|
-
file_path = os.path.join(dir_path, f"{i}.cbor")
|
45
|
-
with open(file_path, "wb") as f:
|
46
|
-
f.write(output_item)
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor.py
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor_factory.py
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/function_executor_map.py
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/executor/function_executor/process_function_executor.py
RENAMED
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/handlers/run_function/handler.py
RENAMED
File without changes
|
File without changes
|
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/initialize_request_validator.py
RENAMED
File without changes
|
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor.proto
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2.py
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2.pyi
RENAMED
File without changes
|
{indexify-0.2.42 → indexify-0.2.44}/indexify/function_executor/proto/function_executor_pb2_grpc.py
RENAMED
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|