Bugfix/beat redux (#3639)

* WIP

* WIP

* try spinning out check for indexing into a system task

* check for the correct delimiter

* use constants

---------

Co-authored-by: Richard Kuo (Danswer) <rkuo@onyx.app>
Co-authored-by: Richard Kuo <rkuo@rkuo.com>
This commit is contained in:
rkuo-danswer
2025-01-17 12:59:43 -08:00
committed by GitHub
parent 1ad2128b2a
commit 6fc52c81ab
15 changed files with 1017 additions and 826 deletions

View File

@@ -20,6 +20,7 @@ from sqlalchemy.orm import Session
from onyx.background.celery.apps.task_formatters import CeleryTaskColoredFormatter
from onyx.background.celery.apps.task_formatters import CeleryTaskPlainFormatter
from onyx.background.celery.celery_utils import celery_is_worker_primary
from onyx.configs.constants import ONYX_CLOUD_CELERY_TASK_PREFIX
from onyx.configs.constants import OnyxRedisLocks
from onyx.db.engine import get_sqlalchemy_engine
from onyx.document_index.vespa.shared_utils.utils import get_vespa_http_client
@@ -100,6 +101,10 @@ def on_task_postrun(
if not task_id:
return
if task.name.startswith(ONYX_CLOUD_CELERY_TASK_PREFIX):
# this is a cloud / all tenant task ... no postrun is needed
return
# Get tenant_id directly from kwargs- each celery task has a tenant_id kwarg
if not kwargs:
logger.error(f"Task {task.name} (ID: {task_id}) is missing kwargs")

View File

@@ -1,5 +1,6 @@
from datetime import timedelta
from typing import Any
from typing import cast
from celery import Celery
from celery import signals
@@ -7,12 +8,14 @@ from celery.beat import PersistentScheduler # type: ignore
from celery.signals import beat_init
import onyx.background.celery.apps.app_base as app_base
from onyx.configs.constants import ONYX_CLOUD_CELERY_TASK_PREFIX
from onyx.configs.constants import POSTGRES_CELERY_BEAT_APP_NAME
from onyx.db.engine import get_all_tenant_ids
from onyx.db.engine import SqlEngine
from onyx.utils.logger import setup_logger
from onyx.utils.variable_functionality import fetch_versioned_implementation
from shared_configs.configs import IGNORED_SYNCING_TENANT_LIST
from shared_configs.configs import MULTI_TENANT
logger = setup_logger(__name__)
@@ -28,7 +31,7 @@ class DynamicTenantScheduler(PersistentScheduler):
self._last_reload = self.app.now() - self._reload_interval
# Let the parent class handle store initialization
self.setup_schedule()
self._update_tenant_tasks()
self._try_updating_schedule()
logger.info(f"Set reload interval to {self._reload_interval}")
def setup_schedule(self) -> None:
@@ -44,105 +47,154 @@ class DynamicTenantScheduler(PersistentScheduler):
or (now - self._last_reload) > self._reload_interval
):
logger.info("Reload interval reached, initiating task update")
self._update_tenant_tasks()
try:
self._try_updating_schedule()
except (AttributeError, KeyError) as e:
logger.exception(f"Failed to process task configuration: {str(e)}")
except Exception as e:
logger.exception(f"Unexpected error updating tasks: {str(e)}")
self._last_reload = now
logger.info("Task update completed, reset reload timer")
return retval
def _update_tenant_tasks(self) -> None:
logger.info("Starting task update process")
try:
logger.info("Fetching all IDs")
tenant_ids = get_all_tenant_ids()
logger.info(f"Found {len(tenant_ids)} IDs")
def _generate_schedule(
self, tenant_ids: list[str] | list[None]
) -> dict[str, dict[str, Any]]:
"""Given a list of tenant id's, generates a new beat schedule for celery."""
logger.info("Fetching tasks to schedule")
logger.info("Fetching tasks to schedule")
tasks_to_schedule = fetch_versioned_implementation(
"onyx.background.celery.tasks.beat_schedule", "get_tasks_to_schedule"
new_schedule: dict[str, dict[str, Any]] = {}
if MULTI_TENANT:
# cloud tasks only need the single task beat across all tenants
get_cloud_tasks_to_schedule = fetch_versioned_implementation(
"onyx.background.celery.tasks.beat_schedule",
"get_cloud_tasks_to_schedule",
)
new_beat_schedule: dict[str, dict[str, Any]] = {}
cloud_tasks_to_schedule: list[
dict[str, Any]
] = get_cloud_tasks_to_schedule()
for task in cloud_tasks_to_schedule:
task_name = task["name"]
cloud_task = {
"task": task["task"],
"schedule": task["schedule"],
"kwargs": {},
}
if options := task.get("options"):
logger.debug(f"Adding options to task {task_name}: {options}")
cloud_task["options"] = options
new_schedule[task_name] = cloud_task
current_schedule = self.schedule.items()
# regular task beats are multiplied across all tenants
get_tasks_to_schedule = fetch_versioned_implementation(
"onyx.background.celery.tasks.beat_schedule", "get_tasks_to_schedule"
)
existing_tenants = set()
for task_name, _ in current_schedule:
if "-" in task_name:
existing_tenants.add(task_name.split("-")[-1])
logger.info(f"Found {len(existing_tenants)} existing items in schedule")
tasks_to_schedule: list[dict[str, Any]] = get_tasks_to_schedule()
for tenant_id in tenant_ids:
if (
IGNORED_SYNCING_TENANT_LIST
and tenant_id in IGNORED_SYNCING_TENANT_LIST
):
logger.info(
f"Skipping tenant {tenant_id} as it is in the ignored syncing list"
)
continue
if tenant_id not in existing_tenants:
logger.info(f"Processing new item: {tenant_id}")
for task in tasks_to_schedule():
task_name = f"{task['name']}-{tenant_id}"
logger.debug(f"Creating task configuration for {task_name}")
new_task = {
"task": task["task"],
"schedule": task["schedule"],
"kwargs": {"tenant_id": tenant_id},
}
if options := task.get("options"):
logger.debug(f"Adding options to task {task_name}: {options}")
new_task["options"] = options
new_beat_schedule[task_name] = new_task
if self._should_update_schedule(current_schedule, new_beat_schedule):
for tenant_id in tenant_ids:
if IGNORED_SYNCING_TENANT_LIST and tenant_id in IGNORED_SYNCING_TENANT_LIST:
logger.info(
"Schedule update required",
extra={
"new_tasks": len(new_beat_schedule),
"current_tasks": len(current_schedule),
},
f"Skipping tenant {tenant_id} as it is in the ignored syncing list"
)
continue
# Create schedule entries
entries = {}
for name, entry in new_beat_schedule.items():
entries[name] = self.Entry(
name=name,
app=self.app,
task=entry["task"],
schedule=entry["schedule"],
options=entry.get("options", {}),
kwargs=entry.get("kwargs", {}),
for task in tasks_to_schedule:
task_name = task["name"]
tenant_task_name = f"{task['name']}-{tenant_id}"
logger.debug(f"Creating task configuration for {tenant_task_name}")
tenant_task = {
"task": task["task"],
"schedule": task["schedule"],
"kwargs": {"tenant_id": tenant_id},
}
if options := task.get("options"):
logger.debug(
f"Adding options to task {tenant_task_name}: {options}"
)
tenant_task["options"] = options
new_schedule[tenant_task_name] = tenant_task
# Update the schedule using the scheduler's methods
self.schedule.clear()
self.schedule.update(entries)
return new_schedule
# Ensure changes are persisted
self.sync()
def _try_updating_schedule(self) -> None:
"""Only updates the actual beat schedule on the celery app when it changes"""
logger.info("Schedule update completed successfully")
else:
logger.info("Schedule is up to date, no changes needed")
except (AttributeError, KeyError) as e:
logger.exception(f"Failed to process task configuration: {str(e)}")
except Exception as e:
logger.exception(f"Unexpected error updating tasks: {str(e)}")
logger.info("_try_updating_schedule starting")
def _should_update_schedule(
self, current_schedule: dict, new_schedule: dict
) -> bool:
"""Compare schedules to determine if an update is needed."""
logger.debug("Comparing current and new schedules")
current_tasks = set(name for name, _ in current_schedule)
new_tasks = set(new_schedule.keys())
needs_update = current_tasks != new_tasks
logger.debug(f"Schedule update needed: {needs_update}")
return needs_update
tenant_ids = get_all_tenant_ids()
logger.info(f"Found {len(tenant_ids)} IDs")
# get current schedule and extract current tenants
current_schedule = self.schedule.items()
current_tenants = set()
for task_name, _ in current_schedule:
task_name = cast(str, task_name)
if task_name.startswith(ONYX_CLOUD_CELERY_TASK_PREFIX):
continue
if "_" in task_name:
# example: "check-for-condition-tenant_12345678-abcd-efgh-ijkl-12345678"
# -> "12345678-abcd-efgh-ijkl-12345678"
current_tenants.add(task_name.split("_")[-1])
logger.info(f"Found {len(current_tenants)} existing items in schedule")
for tenant_id in tenant_ids:
if tenant_id not in current_tenants:
logger.info(f"Processing new tenant: {tenant_id}")
new_schedule = self._generate_schedule(tenant_ids)
if DynamicTenantScheduler._compare_schedules(current_schedule, new_schedule):
logger.info(
"_try_updating_schedule: Current schedule is up to date, no changes needed"
)
return
logger.info(
"Schedule update required",
extra={
"new_tasks": len(new_schedule),
"current_tasks": len(current_schedule),
},
)
# Create schedule entries
entries = {}
for name, entry in new_schedule.items():
entries[name] = self.Entry(
name=name,
app=self.app,
task=entry["task"],
schedule=entry["schedule"],
options=entry.get("options", {}),
kwargs=entry.get("kwargs", {}),
)
# Update the schedule using the scheduler's methods
self.schedule.clear()
self.schedule.update(entries)
# Ensure changes are persisted
self.sync()
logger.info("_try_updating_schedule: Schedule updated successfully")
@staticmethod
def _compare_schedules(schedule1: dict, schedule2: dict) -> bool:
"""Compare schedules to determine if an update is needed.
True if equivalent, False if not."""
current_tasks = set(name for name, _ in schedule1)
new_tasks = set(schedule2.keys())
if current_tasks != new_tasks:
return False
return True
@beat_init.connect

View File

@@ -17,7 +17,7 @@ from redis.lock import Lock as RedisLock
import onyx.background.celery.apps.app_base as app_base
from onyx.background.celery.apps.app_base import task_logger
from onyx.background.celery.celery_utils import celery_is_worker_primary
from onyx.background.celery.tasks.indexing.tasks import (
from onyx.background.celery.tasks.indexing.utils import (
get_unfenced_index_attempt_ids,
)
from onyx.configs.constants import CELERY_PRIMARY_WORKER_LOCK_TIMEOUT